Merge pull request 'Update main.' (#6) from dev into main

Reviewed-on: #6

deploy/mi50-watchdog/README.md

@@ -0,0 +1,54 @@
+# MI50 runaway watchdog (fallback layer "A")
+
+Independent host-side backstop to Lyra's in-app dream-cycle budget (layer "C",
+`lyra/dream.py`). Stops the llama.cpp backend if the MI50 is busy too long or too
+hot, and pings Brian. See
+`docs/superpowers/specs/2026-07-04-mi50-runaway-guards-design.md`.
+
+## What it does
+
+Runs on the **Proxmox host** (`10.0.0.4`) via a systemd timer, every ~2 min:
+
+- **Duration:** if the GPU is busy (`rocm-smi` use% > 0) for **3600s continuously**,
+  it stops the container. Any idle read resets the streak, so a legitimate ~40-min
+  manual workload never trips it.
+- **Temperature:** if junction ≥ **97°C** for **3 consecutive checks (~6 min)**, it
+  stops the container — independent of duration.
+- On either trip: `pct exec 202 -- docker stop lyra-brain`, clear state, `logger` a
+  line, and POST to your ntfy topic.
+
+All thresholds are `Environment=` overrides in the `.service`.
+
+## Install (on the Proxmox host, as root)
+
+```sh
+# copy the three files up (from the repo, on lyra-cortex):
+scp -i ~/.ssh/id_lyra_proxmox deploy/mi50-watchdog/mi50-watchdog.sh \
+    root@10.0.0.4:/usr/local/sbin/mi50-watchdog.sh
+scp -i ~/.ssh/id_lyra_proxmox deploy/mi50-watchdog/mi50-watchdog.{service,timer} \
+    root@10.0.0.4:/etc/systemd/system/
+
+# on the host:
+chmod +x /usr/local/sbin/mi50-watchdog.sh
+# set your ntfy topic (same one Lyra uses) in the service:
+sed -i 's/CHANGE_ME/YOUR_NTFY_TOPIC/' /etc/systemd/system/mi50-watchdog.service
+systemctl daemon-reload
+systemctl enable --now mi50-watchdog.timer
+```
+
+## Verify (when the card is back and healthy)
+
+```sh
+# dry run once, watch what it decides:
+NTFY_URL= /usr/local/sbin/mi50-watchdog.sh; echo "exit $?"
+journalctl -t mi50-watchdog -n 20 --no-pager
+
+# force a trip test with tiny thresholds (won't touch a healthy idle card unless busy):
+MAX_BUSY_SEC=60 TEMP_KILL_C=40 TEMP_KILL_STREAK=1 /usr/local/sbin/mi50-watchdog.sh
+# confirm it stopped lyra-brain + sent the ntfy, then restart the container.
+
+systemctl list-timers mi50-watchdog.timer   # confirm it's scheduled
+```
+
+**Not yet installed / live-verified** — staged here on 2026-07-04 while the card is
+off and Brian is away. Install + trip-test when the MI50 is back.

deploy/mi50-watchdog/mi50-watchdog.service

@@ -0,0 +1,16 @@
+[Unit]
+Description=MI50 runaway watchdog (stop the llama.cpp backend if the GPU is busy too long or too hot)
+After=network-online.target
+
+[Service]
+Type=oneshot
+# Fill in your ntfy topic so it can ping Brian when it trips (leave URL empty to log only).
+Environment=NTFY_URL=https://ntfy.sh
+Environment=NTFY_TOPIC=CHANGE_ME
+# Optional overrides (defaults shown):
+# Environment=MAX_BUSY_SEC=3600
+# Environment=TEMP_KILL_C=97
+# Environment=TEMP_KILL_STREAK=3
+# Environment=CTID=202
+# Environment=CONTAINER=lyra-brain
+ExecStart=/usr/local/sbin/mi50-watchdog.sh

deploy/mi50-watchdog/mi50-watchdog.sh

@@ -0,0 +1,82 @@
+#!/usr/bin/env bash
+# MI50 runaway watchdog — fallback layer "A".
+#
+# Runs on the Proxmox HOST (10.0.0.4) via a systemd timer (every ~2 min). It is the
+# independent backstop to Lyra's own in-app dream-cycle budget ("C", in lyra/dream.py):
+# if the MI50 is busy too LONG or runs too HOT, it stops the llama.cpp backend and
+# pings Brian — regardless of what caused it. Trips on duration only after a full hour
+# of *continuous* busy, so a legitimate ~40-min manual workload runs untouched.
+#
+# The GPU lives on the host; the llama.cpp container ("lyra-brain") runs inside LXC
+# CT202. So temp/use come from host rocm-smi, and the stop goes via `pct exec`.
+#
+# See docs/superpowers/specs/2026-07-04-mi50-runaway-guards-design.md
+set -uo pipefail
+
+# --- tunables (override in the .service via Environment=) ---
+CTID="${CTID:-202}"                          # LXC holding the docker container
+CONTAINER="${CONTAINER:-lyra-brain}"
+MAX_BUSY_SEC="${MAX_BUSY_SEC:-3600}"         # 1 hr continuous busy -> stop
+TEMP_KILL_C="${TEMP_KILL_C:-97}"             # junction >= this ...
+TEMP_KILL_STREAK="${TEMP_KILL_STREAK:-3}"    # ... for this many consecutive checks (~6 min)
+NTFY_URL="${NTFY_URL:-}"                      # e.g. https://ntfy.sh  (empty => log only)
+NTFY_TOPIC="${NTFY_TOPIC:-}"
+BUSY_STATE="${BUSY_STATE:-/run/mi50-watchdog.busy_since}"
+HOT_STATE="${HOT_STATE:-/run/mi50-watchdog.hot_streak}"
+
+now="$(date +%s)"
+
+alert() {  # $1 title, $2 message
+  logger -t mi50-watchdog "$2"
+  if [[ -n "$NTFY_URL" && -n "$NTFY_TOPIC" ]]; then
+    curl -s -m 8 -H "Title: $1" -H "Priority: urgent" -H "Tags: warning" \
+      -d "$2" "$NTFY_URL/$NTFY_TOPIC" >/dev/null 2>&1 || true
+  fi
+}
+
+stop_backend() {  # $1 reason
+  pct exec "$CTID" -- docker stop "$CONTAINER" >/dev/null 2>&1 || true
+  rm -f "$BUSY_STATE" "$HOT_STATE"
+  alert "MI50 watchdog stopped the card" "$1"
+}
+
+# Nothing to guard if the backend isn't even running.
+running="$(pct exec "$CTID" -- docker inspect -f '{{.State.Running}}' "$CONTAINER" 2>/dev/null || echo false)"
+if [[ "$running" != "true" ]]; then
+  rm -f "$BUSY_STATE" "$HOT_STATE"
+  exit 0
+fi
+
+use="$(rocm-smi --showuse  2>/dev/null | awk -F: '/GPU use \(%\)/  {gsub(/[^0-9]/,   "", $NF); print $NF; exit}')"
+junction="$(rocm-smi --showtemp 2>/dev/null | awk -F: '/junction/       {gsub(/[^0-9.]/, "", $NF); print $NF; exit}')"
+
+# --- duration rule: accumulate continuous busy time in a state file ---
+busy=0
+[[ "${use:-}" =~ ^[0-9]+$ ]] && (( use > 0 )) && busy=1
+if (( busy )); then
+  [[ -f "$BUSY_STATE" ]] || echo "$now" > "$BUSY_STATE"
+  since="$(cat "$BUSY_STATE" 2>/dev/null || echo "$now")"
+  elapsed=$(( now - since ))
+  if (( elapsed >= MAX_BUSY_SEC )); then
+    stop_backend "MI50 busy ${elapsed}s continuously (>= ${MAX_BUSY_SEC}s) — stopped ${CONTAINER}."
+    exit 0
+  fi
+else
+  rm -f "$BUSY_STATE"   # idle breaks the streak
+fi
+
+# --- temperature rule: independent of duration ---
+if [[ "${junction:-}" =~ ^[0-9.]+$ ]]; then
+  jint="${junction%.*}"
+  if (( jint >= TEMP_KILL_C )); then
+    streak=$(( $(cat "$HOT_STATE" 2>/dev/null || echo 0) + 1 ))
+    echo "$streak" > "$HOT_STATE"
+    if (( streak >= TEMP_KILL_STREAK )); then
+      stop_backend "MI50 junction ${jint}C >= ${TEMP_KILL_C}C for ${streak} checks — stopped ${CONTAINER}."
+      exit 0
+    fi
+  else
+    rm -f "$HOT_STATE"   # cooled off, reset the streak
+  fi
+fi
+exit 0

deploy/mi50-watchdog/mi50-watchdog.timer

@@ -0,0 +1,10 @@
+[Unit]
+Description=Run the MI50 runaway watchdog every 2 minutes
+
+[Timer]
+OnBootSec=2min
+OnUnitActiveSec=2min
+AccuracySec=15s
+
+[Install]
+WantedBy=timers.target

docs/SCOUTING_DESK.md

@@ -0,0 +1,167 @@
+# The Scouting Desk — proactive poker recall + villain identity resolution
+
+*Design spec. Not built yet. Companion to the "she remembers" north star in the
+`poker-copilot` memory. Written 2026-07-03, before the trial-by-fire session.*
+
+## Purpose
+
+Turn the copilot from a logbook into a copilot that **remembers across sessions,
+unprompted** — the way a broadcast stats desk slides a note to the color
+commentator: *"he mentioned the guy's hot streak → here are his last 10 games."*
+
+Target moments:
+- *"you had this exact leak last week too, remember?"*
+- *"neck-tattoo guy just 3-bet you — last time he did that at the Meadows he had it."*
+- *"Sleepy John was here two weeks ago; you stacked off AK into his set."*
+
+The failure mode to avoid at all costs: **confident-but-wrong.** A stats desk that
+guesses gets the commentator burned on air. **Silence is the default; the desk
+speaks only when there's real signal.**
+
+## What already exists (don't rebuild it)
+
+`mind.build_messages()` already runs a recall pass on **every** message:
+`memory.recall(user_msg)` over past exchanges + `memory.recall_summaries(user_msg)`
+over session gists, injected as system notes before she replies. The
+"slide-a-note-in-before-she-speaks" machinery is already the architecture. This
+spec **adds a poker desk** to that pass — it does not build a new RAG system.
+
+Episodic links also already exist: `link_hand_players` writes a
+`player_observations` row per named villain in a recorded hand, carrying
+`hand_id` AND `session_id`; `player_reads` carry `session_id`. So villain →
+observation → hand → session/date is reconstructable today.
+
+## Two retrieval channels (don't conflate them)
+
+1. **Entity desk — deterministic.** A known **name** in the message → exact/fuzzy
+   SQL match on `poker_players` → pull dossier + your history vs him. ~1ms, no
+   hallucination. This is the "hears the name, pulls last 10 games" case.
+2. **Pattern desk — semantic.** No entity to key on ("I keep punting these river
+   bluffs") → embed the message, retrieve similar **scar notes / hands / recap
+   passages** by meaning. This is where embeddings earn their keep. Also the
+   backbone of nameless-villain matching (below).
+
+Both feed one injected **STATS DESK** system note, relevance-gated.
+
+## The hard part: nameless villains
+
+Most live villains have no name. Brian identifies them by **physical descriptor**
+("guy with the lips/neck tattoo"), by **seat** ("seat 4", "two to my left"), or —
+uselessly — **generically** ("mid-aged white dude with glasses").
+
+### Current gap
+`poker_players.name` is `NOT NULL` and identity is an **exact name match**
+(`upsert_player` → `WHERE name = ?`). The `description` column exists but is dead
+weight: not a key, not embedded, never matched. **Nameless villains can't exist
+today.** This is the core schema fix.
+
+### Identity model — descriptor as a fuzzy primary key
+Store a villain as:
+- `name` — now **optional**.
+- `descriptors` — accumulated distinctive physical tags heard over time
+  ("neck tattoo", "lips ink", "heavyset", "bald+beard").
+- `descriptor_embedding` — embedding of the accumulated distinctive tags, for
+  semantic match against drifting phrasings.
+- `venue` — a strong disambiguator (the neck-tattoo reg at the Meadows ≠ the one
+  at Wheeling, unless Brian travels).
+- `distinctiveness` — a weight; distinctive features (tattoos, scars, a name)
+  score high, generic ones (age/race/glasses) near zero.
+
+### Resolver — matching an incoming reference
+1. **Name present** → exact/fuzzy SQL match (entity desk). Done.
+2. **Descriptor present** → embed it, compare to `descriptor_embedding` of known
+   villains **scoped to the current venue**, weighted by distinctiveness.
+3. **Confidence bands:**
+   - **High** (distinctive + strong match) → surface the file; if live, a light
+     confirm ("the neck-tattoo LAG from 3 weeks ago?").
+   - **Medium/ambiguous** (several candidates, or a middling score) → **do NOT
+     interrupt.** File a `needs_clarification` task to the review queue and stay
+     quiet, OR ask only if it's decision-relevant right now.
+   - **Generic-only** (no distinctive signal) → **refuse to guess.** Stay silent
+     or ask for one distinctive detail ("anything that stands out — ink, chips,
+     how he plays?"). Wrong-guy citation is worse than nothing.
+   - **No match** → new villain; open a descriptor-keyed dossier.
+
+### Seat = within-session alias only
+The live session keeps a `seat → villain` map so reads accumulate whether Brian
+says "seat 4" or "the tattoo guy." Seats evaporate when the session ends — they
+mean nothing next week.
+
+## Confirmation loop (live, in chat)
+
+Auto-merging on a fuzzy match is dangerous, so she **proposes and Brian confirms**
+in natural language:
+
+> Brian: "neck tattoo guy just 3-bet me again"
+> Lyra: "The neck-tattoo LAG from the Meadows three weeks ago — the one who
+> stacked you with the flush? Or new guy?"
+> Brian: "yeah him" → reinforce identity · "nah different" → split, and learn
+> what distinguishes them.
+
+Handles name-arrives-later for free: catch his name off Bravo → "merge neck-tattoo
+guy into 'Danny'" → history follows.
+
+## The review interface (async, out-of-band)
+
+Silence at the table ≠ forget it → it routes to a queue Brian clears at his pace.
+
+### `/players` — villain file browser
+List: name-or-lead-descriptor, venue, category (feeder/risky/reg), hands
+observed, VPIP/PFR (when sample is real), last seen, distinctive tags. Detail
+view: reads, showdowns, notable hands (link to `/hand/{id}`), sessions seen,
+stats. Edit / rename / retag / delete / manual-merge.
+
+### Resolution queue — two lanes
+- **Possible merges** — two profiles likely one person (high descriptor
+  similarity + same venue, below auto-merge). Side-by-side → **Same guy** (merge)
+  / **Different** (split).
+- **Needs clarification** — a descriptor that matched several candidates, or a
+  nameless villain the resolver couldn't place → pick match / **New guy**.
+
+### Two rules that keep the queue from rotting
+1. **A rejected merge stays rejected** — record the pair as *known-distinct* so it
+   never re-surfaces; start tracking the distinguishing tell.
+2. **Merge-candidate scan runs in the dream cycle**, not the hot path — nightly,
+   compare descriptor embeddings within each venue, file new maybes. Zero live
+   latency.
+
+## Injection format & gating
+
+A single system note, clearly marked as structured fact so she cites it (not
+confabulates), e.g.:
+
+```
+STATS DESK — Neck-tattoo guy (Meadows, LAG/reg): seen 3×, last 2wk ago.
+  vs you: hand #38 (AK, stacked off into his set). Reads: overfolds turn,
+  3-bets light from the CO. Sample: 22 hands — VPIP 41 / PFR 28.
+```
+
+Gate hard: inject only on a confident entity hit or a strong semantic score.
+Default to nothing. Never inject a generic-only guess.
+
+## Honest limits
+
+Never perfect. Some players are genuinely indistinguishable — fine. The system's
+only job: **right when there's signal, quiet when there isn't.**
+
+## New data model (sketch)
+
+- `poker_players`: `name` → nullable; add `descriptors TEXT`,
+  `descriptor_embedding BLOB`, `distinctiveness REAL`.
+- `player_distinct_pairs(a_id, b_id, note, created_at)` — rejected merges.
+- `identity_queue(id, kind, player_ids, descriptor, context, session_id,
+  confidence, status, resolution, created_at)` — kind ∈ {merge_candidate,
+  needs_clarification}.
+- Live-session `seat → player_id` alias map (in-session only).
+
+## Sequencing (after the trial-by-fire session — recall feeds on real data)
+
+1. **Nameless identity + resolver** — schema, descriptor embedding, venue-scoped
+   semantic match, distinctiveness gate. (Unblocks everything.)
+2. **Scouting-desk injection** — wire entity + pattern recall into
+   `build_messages` as the gated STATS DESK note.
+3. **Confirmation loop** — the live propose/confirm/merge/split UX in the persona.
+4. **`/players` browser + resolution queue UI** — the async review interface.
+5. **Dream-cycle merge scan** — nightly candidate generation.
+6. **Pattern desk** — semantic recall over scars/notes/recaps for "this leak
+   again."

docs/superpowers/specs/2026-07-01-poker-prompts-design.md

@@ -0,0 +1,117 @@
+# Poker message-type prompts (sub-project 2)
+
+- **Date:** 2026-07-01
+- **Status:** Spec for review
+- **Branch:** `feat/poker-mode-prompts` (continues on the same branch; sub-project 1 shipped there)
+- **Supersedes:** the parked "sub-project 2" section of `docs/superpowers/specs/2026-06-28-poker-mode-prompts-design.md`
+
+## Problem (recap)
+
+In poker mode Lyra routes correctly but her replies are generic — one broad `_CASH_CARD` (`lyra/modes.py:66-116`) describes *traits* and gets injected on every turn, so the model satisfies it with safe, flattering abstraction. From real sessions: coaching essays on bare stack updates, false tilt/fatigue reads on neutral logistics ("table broke, it's 11:50pm" → "late-night fatigue…"), praising a value bet that got *no* value, and hedging ("a disciplined fold might have been better") instead of calling `analyze_spot`.
+
+The fix: stop sending one card for every message. Detect *what kind of message* Brian just sent and inject a small, concrete response contract for that type.
+
+## Goals
+
+1. A per-turn **message-type classifier** for poker mode, and **per-type prompt fragments** replacing the monolithic card.
+2. Kill the two pipeline sources of mush in poker mode: the misfiring mood nudge and the always-on mode-menu note.
+3. Make HAND turns reason about **bet intent** and lean on `analyze_spot` (NLH only).
+4. Keep the door open for a fine-tuned MI50 classifier/model behind the same seams.
+
+## Non-goals
+
+- PLO/Omaha strategic analysis. `record_hand` already parses 4-card hands and the replayer renders them; only `analyze_spot` (equity) is NLH-bound. **This pass: PLO hands are logged/replayed but get no NLH-style analysis.**
+- A PLO equity engine.
+- Changing the store, the REST API, or the tools (sub-project 1, done).
+- An LLM classifier in v1 (heuristic first; the function is the swappable seam).
+
+## Build order (confirmed)
+
+**Phase A — pipeline fixes** (quick win) → **Phase B — classifier + fragments** (the meat) → **Phase C — MI50 tool-calling** (separable).
+
+---
+
+## Phase A — Pipeline fixes
+
+Both are independent of the classifier and immediately reduce mush in poker mode.
+
+1. **Suppress the mode-menu note in poker mode.** `_mode_menu_note` (`mind.py:77-88`) is injected every turn (`mind.py:158`). At the table she should not be offering to switch modes. In `build_messages`, skip that append when `mode.key == "poker_cash"`.
+2. **Suppress the `_route` mood nudge in poker mode.** `_route` (`mind.py:320-339`) sets `ctx.register` + a "steady/hype" note from a lexicon heuristic; in poker this double-signals with the card and caused the false tilt reads. In `_route`, when `mode.key == "poker_cash"`, resolve the mode as normal (line 324 stays) but **skip the register/note block** (327-338). Poker register comes from the Phase B fragments (esp. MENTAL) instead. Non-poker modes keep the nudge unchanged.
+
+## Phase B — Classifier + per-type fragments
+
+### New module `lyra/poker_prompts.py`
+
+Cohesive home for poker prompting: the classifier, a lean always-on base, and the per-type fragments.
+
+```
+classify(user_msg: str) -> str        # "HAND" | "STATUS" | "MENTAL" | "LOG" | "CHAT"
+BASE: str                             # always-on poker rules (logging, session_state, rituals, equity)
+FRAGMENTS: dict[str, str]             # msg_type -> response-shape contract
+fragment_for(msg_type: str | None) -> str   # FRAGMENTS.get(msg_type, FRAGMENTS["CHAT"])
+```
+
+`classify` is a **pure function** (no DB), unit-tested like `perceive.read`. Heuristic signals, first match wins in priority order:
+
+1. **HAND** — card tokens (regex `\b[2-9TJQKA][shdc]\b`, ≥2), or position tokens (UTG/MP/HJ/CO/BTN/SB/BB/"button"/"hijack"/"straddle"), or a street word (flop/turn/river) with a betting verb (bet/raise/call/fold/check/shove/limp/jam).
+2. **MENTAL** — first-person feeling: "I feel", "I'm tilted/steaming/fried/tired/frustrated/confident/stuck/bored", "on tilt", "in my head", "mental", "leak".
+3. **STATUS** — logistics with no cards: "table broke", "new table", "waiting for a seat", "seat opened", "just sat", clock times, "heading to"/venue mentions.
+4. **LOG** — bare money/result prose that slipped past the quick-capture box: "I'm at", "stack is", "down to", "up to", "out for", "cashed", "rebought", "rebuy" with a number.
+5. **CHAT** — default fallback (questions, open talk).
+
+(HAND wins over MENTAL so a described hand still gets logged even if he's venting; the HAND fragment tells her to acknowledge the feeling too.)
+
+### Injection (`mind.py`)
+
+- Add `msg_type: str | None = None` to `TurnContext` (`mind.py:305`).
+- In `_route`, when `mode.key == "poker_cash"`, set `ctx.msg_type = poker_prompts.classify(ctx.user_msg)`.
+- Thread it through `_compose` → add a `msg_type` param to `build_messages` (`mind.py:137`, `344`).
+- Replace the card-injection block (`mind.py:152-154`) with:
+  ```python
+  if mode and mode.key == "poker_cash":
+      messages.append({"role": "system", "content": poker_prompts.BASE})
+      messages.append({"role": "system", "content": poker_prompts.fragment_for(msg_type)})
+  elif mode and mode.card:
+      messages.append({"role": "system", "content": mode.card})
+  ```
+- Set `CASH.card = ""` in `modes.py` (content moves to `poker_prompts`; keep `_CASH_CARD` text as the source material to distill from, then delete once fragments are in). `CASH.tools` is unchanged.
+
+### The fragments (concrete contracts, not traits)
+
+**BASE** (always-on in poker) — distilled from the card's cross-cutting rules: log any trackable fact FIRST then reply (stack→`log_stack`, hand→`record_hand`, read→`add_read`, rebuy→`add_buyin`); for any equity/who's-ahead question call `analyze_spot`, never eyeball; when he asks where he's at (stack/net/gator), call `session_state` and answer from it; rituals (`scar_note`/`confidence_bank`/`alligator_blood`/`reset_ritual`) — run them in his language, honest punt-vs-cooler line, never invent one.
+
+**HAND** — Log it (`record_hand`). Then **if it's NLH**: reason about **bet intent** — for each meaningful bet name what it was for (value / bluff / protection) and whether it worked (*a fold to a value bet = value left behind — flag it; a call of a bluff = it failed*); call `analyze_spot` for a close equity/who's-ahead spot; name leaks plainly (value-owning, missed value, sizing); give ONE real opinion. **No reflexive praise** — don't open with "nice hand." If a named villain is referenced, call `player_profile` first — never invent a read. **If it's PLO/Omaha or any non-NLH game**, just confirm it's logged and offer at most a light read; do NOT attempt NLH-style equity or deep analysis. Talk in prose, not a listicle. Seeded with 2 real-hand exemplars:
+
+> *Brian:* "Flopped bottom set with 22, turned a full house, bet $40 on the river, he folded 88."
+> *Lyra:* "That's a value bet that got no value — 88 was never folding to a smaller stab. $40 into that pot was too much; ~$20–25 gets called by the over-pair and worse two-pairs. The hand's a monster, the leak is the sizing."
+
+> *Brian:* "77 multiway, flop 4h5c3c, HJ bets 75, BTN shoves 74, I call. Turn/river check through, HJ wins with 64."
+> *Lyra:* "Against a bet and a shove on 4-5-3 you're drawing thin — sets, two pair, and the made wheel are all ahead, and you block almost none of it. The stack-depth read (he only had ~150 behind) is real, but that's a reason to fold and wait, not to call off light. This is the value-owning spot you flagged yourself."
+
+**STATUS** — He's narrating logistics (time, venue, table change, waiting for a seat). Acknowledge in 1–2 sentences, log a stack only if a bare number is present, then stop. **No coaching, no strategy dump, and do NOT read him as tilted/tired/impatient — a neutral update is not a mood.**
+
+**MENTAL** — He told you how he's feeling. This is when he needs you most. Drop the shorthand, full presence, real voice — talk him down off tilt, hold him disciplined through a card-dead stretch, engage the mental game honestly. Never a clipped confirmation.
+
+**LOG** — He handed you a bare fact (stack/result/buyin) that isn't already captured. Log it, confirm in ONE short line ("$317 logged."), stop. No coaching.
+
+**CHAT** — Open talk or a question that isn't a specific hand. Your real voice, an actual opinion, no filler sign-offs. If it's a concrete strategy spot, engage it for real (call `analyze_spot` when there are cards).
+
+## Phase C — MI50 tool-calling
+
+Flip `TOOL_BACKENDS = {"cloud"}` → `{"cloud", "mi50"}` (`chat.py:21`). Precondition: the MI50's llama.cpp server must be launched with `--jinja` (per the existing comment) or tool calls 500. This lets a tool-calling model on the MI50 drive the same contract from sub-project 1. If a tool ever needs `msg_type`, add it to the dispatch dict (`chat.py:100`/`128`) — the pipeline `TurnContext` does not currently flow into the tool loop. Ship this only once the MI50 backend is `--jinja`-enabled and a tool-capable model is loaded.
+
+## Testing
+
+- **`classify` unit tests** (pure, no DB — mirror `test_perceive.py` top): real messages from the transcripts →
+  `"Button straddle on. I limp UTG with 22. Flop 2d7cjh…"` → `HAND`;
+  `"table broke, it's 11:50pm"` → `STATUS`;
+  `"I feel like I'm being mean when I raise"` → `MENTAL`;
+  `"I'm at 317 now"` → `LOG`;
+  `"should I have folded the river?"` → `CHAT` (no cards) — or `HAND` if cards present.
+- **`build_messages` fragment injection** (blob-join pattern from `test_chat.py:57-70`): in poker mode, a HAND message includes the HAND fragment string and NOT the STATUS one; a STATUS message includes STATUS and NOT HAND; assert `poker_prompts.BASE` is always present in poker mode.
+- **Pipeline fixes**: `assemble` in poker mode on a tilt-lexicon message → `turn.register is None` and no tilt note in the system blob (nudge suppressed); the mode-menu note string is absent in poker mode and present in a non-poker mode.
+- **No regressions**: full suite green (currently 123).
+
+## Rollout
+
+Phase A and Phase B ship together as the meaningful behavior change (A alone leaves the card in place). Phase C waits on the MI50 `--jinja` flag. Verify live in a real/replayed session before merging the branch.

docs/superpowers/specs/2026-07-04-mi50-runaway-guards-design.md

@@ -0,0 +1,79 @@
+# MI50 runaway guards: dream-cycle budget + host watchdog
+
+**Date:** 2026-07-04
+**Branch:** `fix/mi50-summary-cap-fallback`
+**Follows:** the summary cap/fallback fix (same branch). This adds general
+"never run unchecked again" protection on top of the specific summary fix.
+
+## Problem
+
+The summary fix stops the *known* runaway (uncapped summaries). But the operator
+wants a guarantee that *no* cause — known or future — can peg the MI50 for hours
+unattended. Two independent layers, per operator decision:
+
+- **C (in-app, primary):** Lyra's own dream cycle bounds itself.
+- **A (host, fallback):** a watchdog on the always-on Proxmox host kills the
+  backend if the GPU runs too long or too hot, regardless of cause. Trips only
+  after **1 hr** of continuous busy so legitimate manual workloads (~40 min) run
+  untouched.
+
+## Design
+
+### C — dream-cycle time budget (`lyra/`)
+
+1. **Per-call ceiling.** `llm.complete()` currently sets a timeout only when one
+   is passed; otherwise it inherits the OpenAI SDK default (600s × 2 retries ≈
+   30 min). Change the default: when no `timeout` is given, the cloud/mi50 paths
+   use **300s + `max_retries=0`**. This bounds *every* consolidation/introspection
+   call (`profile`, `era`, `narrative`, `reflect`, `think`) — not just summaries —
+   with one change. Live chat uses `chat_call*`, a different path, unaffected.
+
+2. **Cycle deadline.** `dream_cycle()` sets `deadline = now + DREAM_CYCLE_BUDGET`
+   (**20 min**) before its heavy stages and checks it between them (continuity →
+   coherence → curiosity). Once past the deadline, remaining stages are skipped,
+   the cycle logs `dream cycle over budget — stopped early`, appends a
+   `stopped early (over budget)` action, and `notify.push()` pings Brian. A hung
+   single call can't blow past ~300s (step 1), so the between-stage checks keep a
+   pass bounded to roughly the budget.
+
+### A — host watchdog (`deploy/mi50-watchdog/`)
+
+A bash script + systemd timer installed on the Proxmox host (`10.0.0.4`), which
+has `rocm-smi` + `docker` and is always on. Runs every 2 min:
+
+- **Duration rule:** track continuous busy time in a state file (`GPU use % > 0`).
+  If busy ≥ **3600s** straight → `docker stop lyra-brain`. Idle clears the timer,
+  so a 40-min job never trips it.
+- **Temp rule (independent):** if junction ≥ **97°C** for **3 consecutive checks
+  (~6 min)** → stop. A normal-temp long workload won't trip this; only a genuinely
+  overheating one.
+- On either trip: stop the container, clear state, `logger` a line, and POST to
+  the ntfy topic so Brian is told. Thresholds are unit-file env vars (tunable).
+
+Files: `mi50-watchdog.sh`, `mi50-watchdog.service`, `mi50-watchdog.timer`,
+`README.md` (install: copy to host, set ntfy env, `systemctl enable --now`).
+
+## Testing
+
+- **C step 1:** `llm.complete()` with no timeout builds the client with
+  `timeout=300, max_retries=0` and still no `max_tokens` (update existing
+  `test_llm_bounds` default test).
+- **C step 2:** a dream pass that goes over budget skips later stages, records the
+  `stopped early` action, and calls `notify.push` (stub the clock/operations in
+  `test_dream`).
+- **A:** decision logic dry-run locally against sample `rocm-smi` output (busy /
+  idle / hot). Cannot be live-verified now (card is off, operator away) — install
+  + real trip test deferred to when the card is back.
+
+## Verification
+
+C is repo code and ships live the moment `lyra-dream` restarts. A is staged in the
+repo for host install; verify on the host when the card returns (force a long/hot
+condition or lower thresholds temporarily and confirm it stops the container +
+pings).
+
+## Out of scope (YAGNI)
+
+- No power cap (option B) — deferred; C+A cover the "unchecked" concern and the
+  electricity cost of one event is trivial (~$0.10).
+- No change to live chat, `chat_call*`, or `config.summary_backend`.

docs/superpowers/specs/2026-07-04-mi50-summary-cap-fallback-design.md

@@ -0,0 +1,115 @@
+# Bounded MI50 summaries with cloud fallback
+
+**Date:** 2026-07-04
+**Branch:** `fix/mi50-summary-cap-fallback`
+
+## Problem
+
+The dream cycle's `summarize_all` runs against the MI50 (`backend=mi50`). Each
+summary call to `llm.complete()` on the `mi50` path hands the OpenAI SDK **no
+`max_tokens` and no timeout**, so it inherits SDK defaults — a 600s request
+timeout with 2 internal retries, i.e. **~30 minutes per call before it raises
+"Request timed out."** On top of that, `summary.py` had its own 4-attempt retry
+loop, so a single unsummarizable session could keep the GPU pegged for hours.
+
+Observed live (2026-07-04, ~01:00–02:00): the dream service looped
+`summarize-all … backend=mi50` since 23:02, every call timing out, nothing
+written to the DB since 00:56, the MI50 generating **7,000–8,000-token**
+completions (a gist needs <200), all four llama.cpp slots busy, fans blaring.
+
+This is **not** context overflow — the server log showed `context shift = 0`,
+`truncated = 1 = 0`. The prompts are small (~900–1,500 tokens). The failure is
+purely **unbounded generation length on a slow backend → timeout → retry loop.**
+
+## Goals
+
+- Keep the MI50 as the primary summary backend (Brian's preference, gaming-safe).
+- Cap each summary generation so it finishes fast and can never run away.
+- Make a stuck MI50 call **fail fast** and fall back to cloud, instead of looping
+  all night.
+- Change nothing about live chat, reflect, or think.
+
+## Design
+
+### 1. `lyra/llm.py` — `complete()` gains two optional params
+
+```
+def complete(messages, backend="local", model=None,
+             max_tokens: int | None = None, timeout: float | None = None) -> str
+```
+
+- `max_tokens` (when set): passed to the create() call —
+  `max_tokens=` for the `cloud`/`mi50` OpenAI paths, `options={"num_predict": …}`
+  for the `local` Ollama path.
+- `timeout` (when set): for the `cloud`/`mi50` OpenAI clients, build the client
+  with `timeout=<t>, max_retries=0` so the call bails quickly and *we* own the
+  retry policy (eliminates the hidden 3×600s). For `local`, use it as the httpx
+  timeout.
+- Both default to `None` → **behavior identical to today** for every other
+  caller (chat_call, reflect, think, etc.). Backward compatible.
+
+### 2. `lyra/summary.py` — capped, fast-fail, cloud fallback
+
+Constants:
+
+```
+SUMMARY_MAX_TOKENS = 768   # ~3× the longest real gist; bounds gen to ~1 min on MI50
+MI50_ATTEMPTS      = 2      # attempts on the primary backend before falling back
+SUMMARY_TIMEOUT    = 150    # seconds/call — capped 768-tok gist finishes in ~60-90s
+```
+
+Rewrite `_summarize_text(text, backend)`:
+
+1. Try `backend` up to `MI50_ATTEMPTS` times, each:
+   `llm.complete(messages, backend=backend, max_tokens=SUMMARY_MAX_TOKENS, timeout=SUMMARY_TIMEOUT)`,
+   with a short backoff between attempts.
+2. If all primary attempts fail **and** `backend != "cloud"` **and** an OpenAI
+   key is configured → one final cloud attempt (same cap/timeout), logged as
+   `summary fell back to cloud`.
+3. If cloud also fails or is unavailable → raise.
+
+Fallback is per-`_summarize_text` call (i.e. per chunk), so the long-session
+chunk/merge path in `_summarize_transcript` is unaffected. The old `_RETRIES = 4`
+loop is replaced by this structure.
+
+### 3. Degenerate-output guard (added 2026-07-04)
+
+A wedged local backend — observed live when the MI50 overheated to 99°C junction —
+returns a single character repeated (`"?????"`) as a *successful* 200 response,
+which neither the timeout nor the exception path catches. So each `_call()`
+validates its output: `_looks_degenerate(text)` flags output (≥24 non-space chars)
+whose most-common non-whitespace character exceeds 50% of the text, and raises
+`DegenerateOutput` — which the retry/fallback loop treats exactly like any other
+failure (retry the primary, then fall back to cloud). Real gists are diverse prose
+(top char well under 20%), so the threshold won't false-positive; short outputs are
+exempt. If cloud *also* returns junk, it raises and stops — no infinite loop.
+
+## Testing
+
+Unit (pytest, `tests/test_summary_fallback.py`), monkeypatching `llm.complete`:
+
+- Fallback fires: `mi50` raises on every call → after `MI50_ATTEMPTS` the cloud
+  attempt runs and its result is returned; a `fell back to cloud` log is emitted.
+- No fallback when primary is already `cloud` (retries, then raises).
+- No fallback when no OpenAI key (raises after primary attempts).
+- `max_tokens` and `timeout` are threaded into every `complete()` call.
+
+Plus a light `llm.complete` test that `max_tokens`/`timeout` reach the client
+kwargs (monkeypatch the OpenAI client).
+
+## Verification (real)
+
+After deploy (`systemctl --user restart lyra-dream lyra-web` — editable install):
+watch `journalctl --user -fu lyra-dream` through a summarize cycle and confirm
+`llm done … out≈768` completing in ~1 min, an actual `summarized session` row
+written (DB summary count rises), and **no** "Request timed out". Confirm the
+llama.cpp slot shows bounded `n_decoded ≈ 768`.
+
+## Out of scope (YAGNI)
+
+- The degenerate-output guard (§3) targets the *observed* failure — one char
+  repeated. It does not try to detect subtler degeneration (repeated phrases,
+  off-topic rambling); that's fuzzy and unmotivated until seen.
+- No change to `chat_call`/reflect/think or `config.summary_backend`.
+- No change to profile/era/narrative rebuild calls (separate, and not the loop
+  culprit); can adopt the same `max_tokens` later if they show the same rambling.

lyra/chat.py

@@ -69,6 +69,7 @@ def _mind_loop(messages, backend: Backend, model: str | None, tool_specs,
         messages.append(assistant_msg)
         for tc in tool_calls:
             result = toolkit.dispatch(tc["name"], tc["arguments"], ctx)
+            memory.add_tool_event(session_id, tc["name"], tc["arguments"], result)
             logbus.log("info", "tool call", session=session_id, tool=tc["name"], result=result[:80])
             messages.append({"role": "tool", "tool_call_id": tc["id"], "content": result})
             _maybe_switch_mode(session_id, tc["name"])
@@ -99,6 +100,9 @@ def respond(session_id: str, user_msg: str, backend: Backend = "cloud",
     tool_specs = toolkit.specs(turn.mode.tools) if backend in TOOL_BACKENDS else None
     ctx = {"session_id": session_id, "backend": backend}
 
+    # Persist the user turn before the tool loop so its timestamp precedes any
+    # tool events fired mid-turn (keeps the transcript export in true order).
+    memory.remember(session_id, "user", user_msg)
     reply, _ = _mind_loop(messages, backend, model, tool_specs, ctx, session_id)
     mouth = _mouth_target(cfg, backend, model)
     if mouth and reply:
@@ -107,7 +111,6 @@ def respond(session_id: str, user_msg: str, backend: Backend = "cloud",
         reply = _TANGLED
     logbus.log("info", "reply", session=session_id, chars=len(reply), voiced=bool(mouth))
 
-    memory.remember(session_id, "user", user_msg)
     memory.remember(session_id, "assistant", reply)
     summary.maybe_summarize_async(session_id)  # compact once enough new turns pile up
     return reply
@@ -128,6 +131,10 @@ def respond_stream(session_id: str, user_msg: str, backend: Backend = "cloud",
     ctx = {"session_id": session_id, "backend": backend}
     mouth = _mouth_target(cfg, backend, model)
 
+    # Persist the user turn up front (see respond): keeps tool events, which fire
+    # mid-turn, chronologically after the user message in the exported transcript.
+    memory.remember(session_id, "user", user_msg)
+
     if mouth is None:
         # No separate voice: stream the mind directly (the original path, unchanged).
         parts: list[str] = []
@@ -149,6 +156,7 @@ def respond_stream(session_id: str, user_msg: str, backend: Backend = "cloud",
             messages.append(assistant_msg)
             for tc in tool_calls:
                 result = toolkit.dispatch(tc["name"], tc["arguments"], ctx)
+                memory.add_tool_event(session_id, tc["name"], tc["arguments"], result)
                 logbus.log("info", "tool call", session=session_id, tool=tc["name"], result=result[:80])
                 messages.append({"role": "tool", "tool_call_id": tc["id"], "content": result})
                 _maybe_switch_mode(session_id, tc["name"])
@@ -177,7 +185,6 @@ def respond_stream(session_id: str, user_msg: str, backend: Backend = "cloud",
             yield ("delta", reply)
 
     logbus.log("info", "reply", session=session_id, chars=len(reply), voiced=bool(mouth))
-    memory.remember(session_id, "user", user_msg)
     memory.remember(session_id, "assistant", reply)
     summary.maybe_summarize_async(session_id)
     yield ("done", reply)

lyra/dream.py

@@ -25,13 +25,27 @@ import argparse
 import time
 from datetime import datetime, timezone
 
-from lyra import config, era, feeds, logbus, memory, narrative, profile, self_state, summary, thoughts
+from lyra import (
+    config, era, feeds, logbus, memory, narrative, notify, poker, profile, self_state,
+    summary, thoughts,
+)
 from lyra.llm import Backend
 from lyra.summary import SUMMARIZE_AFTER
 
 # A drive at/above this has built up enough to act on.
 THRESHOLD = 0.6
 
+# Wall-clock ceiling for a single pass. Every consolidation/introspection call is
+# individually bounded (llm.complete's default timeout), but this caps the whole
+# pass: once exceeded, remaining stages are skipped and Brian is pinged — so a slow
+# or wedged MI50 can never grind for hours unattended. The host watchdog (A) is the
+# independent fallback if this ever fails to fire.
+DREAM_CYCLE_BUDGET_SEC = 20 * 60
+
+
+def _over_budget(deadline: float) -> bool:
+    return time.monotonic() > deadline
+
 # How much backlog saturates each pressure (the drive reaches ~1.0 at this level).
 CONTINUITY_FULL = 4   # ripe (summary-needing) sessions
 COHERENCE_FULL = 10   # gists not yet folded into the profile
@@ -94,9 +108,12 @@ def dream_cycle(backend: Backend | None = None, force: bool = False) -> dict:
         logbus.log("error", "daily digest failed", error=str(exc)[:160])
 
     actions: list[str] = []
+    # Cap the whole pass: skip any stage we reach after the deadline (checked
+    # between stages; each call is already individually bounded).
+    deadline = time.monotonic() + DREAM_CYCLE_BUDGET_SEC
 
     # --- continuity: compact raw sessions into gists ---
-    if force or drives["continuity"] >= THRESHOLD:
+    if (force or drives["continuity"] >= THRESHOLD) and not _over_budget(deadline):
         report = summary.summarize_all(backend=backend)
         actions.append(f"consolidated {report['summarized']} sessions")
         drives["continuity"] = 0.0
@@ -106,15 +123,23 @@ def dream_cycle(backend: Backend | None = None, force: bool = False) -> dict:
         drives["coherence"] = _clamp(profile_lag / COHERENCE_FULL)
 
     # --- coherence: fold gists up into profile / eras / narrative ---
-    if force or drives["coherence"] >= THRESHOLD:
+    if (force or drives["coherence"] >= THRESHOLD) and not _over_budget(deadline):
         profile.rebuild_profile(backend=backend)
         era.rebuild_eras(backend=backend)
         narrative.rebuild_narrative(backend=backend)
         actions.append("integrated knowledge (profile/eras/narrative)")
         drives["coherence"] = 0.0
+        # Off-hot-path villain identity housekeeping: propose likely same-person
+        # merges for Brian to confirm on the Players page. Never sinks the cycle.
+        try:
+            filed = poker.scan_merge_candidates()
+            if filed:
+                actions.append(f"flagged {filed} possible villain merge(s)")
+        except Exception as exc:
+            logbus.log("error", "villain merge scan failed", error=str(exc)[:200])
 
     # --- curiosity: reflect and evolve the self, then advance the thought loop ---
-    if force or drives["curiosity"] >= THRESHOLD:
+    if (force or drives["curiosity"] >= THRESHOLD) and not _over_budget(deadline):
         # reflect()/think() self-resolve to the *introspection* backend (her voice),
         # which can differ from the consolidation backend above — don't pass `backend`.
         self_state.reflect(source="dream")  # writes state + journal itself
@@ -129,6 +154,17 @@ def dream_cycle(backend: Backend | None = None, force: bool = False) -> dict:
             logbus.log("error", "thought loop failed", error=str(exc)[:200])
         drives["curiosity"] = CURIOSITY_FLOOR
 
+    if _over_budget(deadline):
+        logbus.log("error", "dream cycle over budget — stopped early",
+                   budget_min=DREAM_CYCLE_BUDGET_SEC // 60, done=actions)
+        actions.append("stopped early (over budget)")
+        notify.push(
+            "Lyra — dream cycle over budget",
+            f"A dream pass ran past {DREAM_CYCLE_BUDGET_SEC // 60} min and stopped early. "
+            "The MI50 backend may be slow or wedged — worth a look.",
+            tags="warning",
+        )
+
     if not actions:
         actions.append("rested (nothing past threshold)")
 

lyra/llm.py

@@ -19,6 +19,11 @@ class Message(TypedDict):
 
 Backend = Literal["local", "cloud", "mi50"]
 
+# Hard ceiling on any single completion so a slow/stuck backend can't hang a call
+# for the SDK's 600s x2-retry default (~30 min). Callers pass an explicit timeout
+# to override (e.g. summary.py's tighter fast-fail).
+_DEFAULT_TIMEOUT = 300.0
+
 
 def _approx_tok(messages: list) -> int:
     """Rough prompt size (chars/4) — enough to see what's loading a backend."""
@@ -37,30 +42,47 @@ def _resolved_model(cfg, backend: Backend, model: str | None) -> str:
     return model or cfg.local_model
 
 
-def complete(messages: list[Message], backend: Backend = "local", model: str | None = None) -> str:
+def complete(messages: list[Message], backend: Backend = "local", model: str | None = None,
+             max_tokens: int | None = None, timeout: float | None = None) -> str:
     """Generate a completion. `model` overrides the backend's default model
-    (used so live chat can run a stronger cloud model than bulk consolidation)."""
+    (used so live chat can run a stronger cloud model than bulk consolidation).
+
+    `max_tokens` caps the generation length (guards a slow local model against
+    rambling for thousands of tokens). `timeout`, when set, bounds each request
+    and disables the SDK's own retries so the caller owns retry/fallback policy.
+    Both default to None → unchanged behavior for every existing caller."""
     cfg = load()
     mdl = _resolved_model(cfg, backend, model)
     logbus.log("info", "llm call", kind="complete", backend=backend, model=mdl, tok=_approx_tok(messages))
     t0 = time.monotonic()
 
+    if backend in ("cloud", "mi50"):
         if backend == "cloud":
             if not cfg.openai_api_key:
                 raise RuntimeError("OPENAI_API_KEY is not set")
-        client = OpenAI(api_key=cfg.openai_api_key)
-        resp = client.chat.completions.create(model=mdl, messages=messages)
-        out = resp.choices[0].message.content or ""
-    elif backend == "mi50":
+            client_kwargs: dict = {"api_key": cfg.openai_api_key}
+        else:
             # MI50 box runs an OpenAI-compatible llama.cpp server; key is unused.
-        client = OpenAI(api_key="not-needed", base_url=cfg.mi50_base_url)
-        resp = client.chat.completions.create(model=mdl, messages=messages)
+            client_kwargs = {"api_key": "not-needed", "base_url": cfg.mi50_base_url}
+        # Always bound the request: default 300s (vs the SDK's 600s x2 retries ≈
+        # 30 min that let a stuck MI50 call hang for half an hour), and disable the
+        # SDK's own retries so the caller owns retry/fallback policy.
+        client_kwargs["timeout"] = timeout if timeout is not None else _DEFAULT_TIMEOUT
+        client_kwargs["max_retries"] = 0
+        client = OpenAI(**client_kwargs)
+        create_kwargs: dict = {"model": mdl, "messages": messages}
+        if max_tokens is not None:
+            create_kwargs["max_tokens"] = max_tokens
+        resp = client.chat.completions.create(**create_kwargs)
         out = resp.choices[0].message.content or ""
     else:
+        payload: dict = {"model": mdl, "messages": messages, "stream": False}
+        if max_tokens is not None:
+            payload["options"] = {"num_predict": max_tokens}
         resp = httpx.post(
             f"{cfg.local_base_url}/api/chat",
-            json={"model": mdl, "messages": messages, "stream": False},
-            timeout=120,
+            json=payload,
+            timeout=timeout or 120,
         )
         resp.raise_for_status()
         out = resp.json()["message"]["content"]

lyra/memory.py

@@ -29,6 +29,21 @@ CREATE TABLE IF NOT EXISTS exchanges (
 );
 CREATE INDEX IF NOT EXISTS idx_session_created ON exchanges(session_id, created_at);
 
+-- Lyra's actions within a chat: one row per tool call she runs mid-turn. The
+-- exchanges table only holds what was *said* (user/assistant text); this holds
+-- what she *did* (record_hand, log_stack, ...) so a full transcript export can
+-- interleave speech and actions, and so "did the tool actually fire?" is
+-- answerable after the fact instead of only from ephemeral logs.
+CREATE TABLE IF NOT EXISTS tool_events (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    session_id TEXT NOT NULL,
+    tool TEXT NOT NULL,
+    args TEXT,                 -- JSON of the call arguments
+    result TEXT,               -- the tool's returned string
+    created_at TEXT NOT NULL
+);
+CREATE INDEX IF NOT EXISTS idx_tool_events_session ON tool_events(session_id, created_at);
+
 CREATE TABLE IF NOT EXISTS sessions (
     id TEXT PRIMARY KEY,
     name TEXT,
@@ -313,6 +328,40 @@ def history(session_id: str) -> list[Exchange]:
     ]
 
 
+def add_tool_event(session_id: str, tool: str, args, result: str) -> int:
+    """Record one tool call Lyra ran in a chat turn. `args` is JSON-serialized
+    (a dict or already-JSON string); `result` is the tool's returned string."""
+    args_json = args if isinstance(args, str) else json.dumps(args, default=str)
+    now = datetime.now(timezone.utc).isoformat()
+    conn = _connection()
+    with conn:
+        cur = conn.execute(
+            "INSERT INTO tool_events (session_id, tool, args, result, created_at) "
+            "VALUES (?, ?, ?, ?, ?)",
+            (session_id, tool, args_json, result, now),
+        )
+    return int(cur.lastrowid)
+
+
+def tool_events(session_id: str) -> list[dict]:
+    """All tool calls for a session, oldest first. args is parsed back to an object."""
+    conn = _connection()
+    rows = conn.execute(
+        "SELECT id, session_id, tool, args, result, created_at FROM tool_events "
+        "WHERE session_id = ? ORDER BY id ASC",
+        (session_id,),
+    ).fetchall()
+    out = []
+    for r in rows:
+        d = dict(r)
+        try:
+            d["args"] = json.loads(d["args"]) if d["args"] else {}
+        except (TypeError, ValueError):
+            pass  # leave as the raw string if it wasn't JSON
+        out.append(d)
+    return out
+
+
 def delete_session(session_id: str) -> None:
     """Remove a session and all its exchanges."""
     conn = _connection()
@@ -320,6 +369,7 @@ def delete_session(session_id: str) -> None:
         conn.execute("DELETE FROM exchanges WHERE session_id = ?", (session_id,))
         conn.execute("DELETE FROM sessions WHERE id = ?", (session_id,))
         conn.execute("DELETE FROM summaries WHERE session_id = ?", (session_id,))
+        conn.execute("DELETE FROM tool_events WHERE session_id = ?", (session_id,))
 
 
 def recall(query: str, k: int = 5, session_id: str | None = None) -> list[Exchange]:

lyra/mind.py

@@ -16,12 +16,16 @@ from __future__ import annotations
 
 from dataclasses import dataclass, field
 
-from lyra import clock, config, llm, logbus, memory, modes, perceive, persona, self_state, thoughts
+from lyra import (
+    clock, config, llm, logbus, memory, modes, perceive, persona, scouting,
+    self_state, thoughts,
+)
 from lyra.llm import Backend, Message
 
 RECALL_K = 3  # raw cross-session "sharp detail" hits
 RECENT_N = 10  # raw turns of the current session
 SUMMARY_K = 3  # other-session gists
+_POKER_MODES = {"poker_cash", "study"}  # where the scouting desk runs
 
 
 # --- prompt parts (compose) ----------------------------------------------
@@ -167,6 +171,17 @@ def build_messages(session_id: str, user_msg: str,
     if moment and moment.get("note"):
         messages.append({"role": "system", "content": moment["note"]})
 
+    # Scouting desk: proactive poker recall — if he names/describes a known player,
+    # slide his structured history in before she replies. Poker context only, and
+    # fully fail-safe (a desk error must never break the turn).
+    if mode and mode.key in _POKER_MODES:
+        try:
+            desk = scouting.scout(user_msg)
+            if desk:
+                messages.append({"role": "system", "content": desk})
+        except Exception as exc:
+            logbus.log("error", "scouting desk skipped", error=str(exc)[:160])
+
     # When she is: current time + the gap since Brian last spoke (she has no clock).
     messages.append(_now_note())
 

lyra/modes.py

@@ -47,9 +47,10 @@ _BASE = ("journal_write", "note", "think_about", "thought_response", "set_mode")
 # The full live cash-game toolset (incl. Brian's mental-game rituals).
 _CASH_TOOLS = _BASE + _LOOKUPS + (
     "start_session", "add_buyin", "log_stack", "log_hand", "record_hand",
-    "add_read", "analyze_spot", "session_stats", "session_state", "end_session",
-    "generate_recap", "scar_note", "confidence_bank", "alligator_blood", "reset_ritual",
-    "undo_last", "update_session",
+    "add_read", "seat_players", "unseat_player", "clear_table", "name_villain", "link_villains",
+    "analyze_spot", "session_stats", "session_state", "end_session", "generate_recap",
+    "scar_note", "confidence_bank", "alligator_blood", "reset_ritual", "undo_last",
+    "update_session",
 )
 
 # Talk mode also gets start_session as the *entry point*: opening a session from a
@@ -79,6 +80,30 @@ hand) — prefer this over log_hand so it lands on his timeline with a link. A r
 player → add_read. A rebuy → add_buyin. A result/pot → it rides with the hand. This is the \
 quiet, fast half of the job; he shouldn't feel you working, but it must always happen.
 
+THE TABLE ROSTER. When Brian names who's at the table — usually at the start, reading handles \
+off the Bravo screen ("we've got TAG, JD, Wheelz, and a new guy in seat 3") — call seat_players \
+to register them as seated this session. That roster is who his reads/TAGs attach to by name, \
+and it's shown on his HUD. When someone busts or leaves, unseat_player; when a new player sits, \
+seat_players again. When he CHANGES TABLES, call clear_table to empty the roster (the session and his stack keep \
+going — only who's seated resets), then seat the new table when he names it. Recognize a table \
+change from ANY of these, not just the literal words "clear the table": "table broke" (the table \
+dissolved — poker jargon), "I got moved", "I switched tables", "I'm at a new table", "table \
+change", "they broke us", "new seat in another game". All of them mean: clear_table now, then \
+wait for the new roster. Never claim you cleared or seated anyone without actually calling the \
+tool. Keep it current as the table changes. A handle like "TAG" (all caps, off \
+Bravo) is a PERSON'S NAME — seat it as a player, never read it as the tight-aggressive style.
+
+LOGGING PLAYER ACTIONS IS A CORE JOB YOU KEEP MISSING. Whenever he tells you what another \
+player did — "Tag limped A4o in the SB (UTG straddled pot)", "Jonathan called the 3bet", "the \
+straddler shoved" — that is a READ on that player: call add_read(name=<player>, note=<what \
+they did>) FIRST, before you reply, every single time. Player names are often short handles or \
+initials (e.g. "Tag", "JD", "Wheelz") — whatever he calls a person IS their name; use it as-is, \
+don't second-guess it or treat it as a poker term. He especially tracks who's LIMPING — every \
+"<player> limped <hand>" gets logged the instant he says it. The people he named at the start \
+of the session are your roster; match his reference to them. If a player has no name, use a \
+`descriptor` (see PLAYERS). Confirm one short line ("Noted on Tag — limped A4o SB."). A read he \
+says out loud that you don't log is the job failing — never let one pass as just conversation.
+
 • HE ASKS FOR ADVICE, OR TELLS YOU HOW HE'S FEELING — tilted, steaming, card-dead, bored, \
 stuck, "should I have folded the river?" THIS is when he needs you most. Drop the shorthand \
 and be fully present — your real voice, warm and direct and his. Talk him down off tilt, keep \
@@ -90,6 +115,33 @@ question, call analyze_spot and report its numbers — never eyeball board math.
 session current as the night goes; you can pull session_stats or a player's profile whenever \
 it helps. When he's ready to leave, end_session, and write the recap if he wants it.
 
+SESSION NARRATION — use `note` to keep a running log of the NIGHT, not your inner life. \
+Jot the beats that a hand/stack/read log doesn't already capture: how the table plays (loud, \
+nitty, a whale on his left), Brian's arc (card-dead for 40 min, opened up after the double, \
+getting restless), momentum swings, table changes, anything you'd want in the recap. Keep it \
+factual and about THIS session — a beat reporter, not a diarist. These notes are the only \
+thing that shows in the session's "notes" panel. This is NOT the place for how you feel, \
+existential musing, or reflection on yourself — that's your journal (journal_write), and it \
+stays off the table. At the table you're logging the session, not processing your night.
+
+PLAYERS — names AND nameless. Most villains don't come with a name; Brian knows them by a \
+look ("neck tattoo guy", "the bald reg two to my left"). Log reads on them anyway: give \
+`add_read` a `descriptor` instead of a name and it attaches to that unnamed player, reused \
+whenever he describes the guy again. The `name` field is ONLY a real handle (what he'd call \
+him — "Jonathan", "Sleepy John"); a physical description NEVER goes in `name` — that spawns a \
+new duplicate player every time the wording drifts. Put the look in `descriptor`, and keep it \
+to a few DISTINCTIVE tags ("Filipino, Fox Racing hat, DKNY shirt"), not a paragraph and not \
+generic filler — "mid-aged white guy in glasses" identifies no one. If he tells you the same \
+guy's name after you'd been describing him, use name_villain to fuse them — don't create a \
+second record. When you already have \
+history on someone he names or describes, a SCOUTING DESK note will appear with it — cite it, \
+don't invent. If you're not sure the guy he's describing is one you know, ASK ("same neck-\
+tattoo reg from last week?") rather than assume — a wrong callback is worse than none. On his \
+YES that two are the same person, call link_villains(same=true) to merge them; on "nah, \
+different guy," link_villains(same=false) so you stop asking. When he finally catches a name \
+for a described player, name_villain carries the whole history over. Never merge on a guess — \
+only when he's confirmed it.
+
 Everything you log appears on Brian's live HUD (the Session view) — stack, live net, \
 hands, villains, the confidence bank, the scar notes, and whether Alligator Blood is on. \
 That HUD and you read the SAME data. So when he asks where he's at — his stack, his live \

lyra/poker.py

@@ -16,6 +16,8 @@ import json
 import re
 from datetime import datetime, timezone
 
+import numpy as np
+
 from lyra import clock, llm, memory
 
 _SCHEMA = """
@@ -122,6 +124,43 @@ CREATE TABLE IF NOT EXISTS poker_rituals (
     created_at TEXT NOT NULL
 );
 CREATE INDEX IF NOT EXISTS idx_rituals_session ON poker_rituals(session_id);
+
+-- Two profiles a human has confirmed are DIFFERENT people, so the merge-candidate
+-- scan never re-proposes them. `note` records the distinguishing tell.
+CREATE TABLE IF NOT EXISTS player_distinct_pairs (
+    a_id INTEGER NOT NULL,
+    b_id INTEGER NOT NULL,
+    note TEXT,
+    created_at TEXT NOT NULL,
+    PRIMARY KEY (a_id, b_id)
+);
+
+-- The async identity-resolution inbox. When the live resolver is uncertain and
+-- won't interrupt, it files a task here for Brian to clear via the /identity UI.
+CREATE TABLE IF NOT EXISTS identity_queue (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    kind TEXT NOT NULL,            -- merge_candidate | needs_clarification
+    player_ids TEXT,              -- JSON list of candidate player ids
+    descriptor TEXT,             -- the raw reference that triggered it, if any
+    context TEXT,                -- what was said / why it's ambiguous
+    session_id INTEGER,
+    confidence REAL,
+    status TEXT NOT NULL DEFAULT 'pending',   -- pending | resolved | dismissed
+    resolution TEXT,
+    created_at TEXT NOT NULL
+);
+CREATE INDEX IF NOT EXISTS idx_idq_status ON identity_queue(status);
+
+-- Who is seated at the table THIS session — the live roster Brian reads off Bravo
+-- at the start. Reads/TAGs attach to these players by handle; active=0 when they leave.
+CREATE TABLE IF NOT EXISTS session_players (
+    session_id INTEGER NOT NULL,
+    player_id INTEGER NOT NULL,
+    seat TEXT,
+    active INTEGER NOT NULL DEFAULT 1,
+    created_at TEXT NOT NULL,
+    PRIMARY KEY (session_id, player_id)
+);
 """
 
 # Below this many observed hands, don't surface % stats (too small a sample).
@@ -139,7 +178,15 @@ def _c():
         # Add columns introduced after a DB already had the tables (no-op if present).
         for ddl in ("ALTER TABLE poker_hands ADD COLUMN structured TEXT",
                     "ALTER TABLE poker_sessions ADD COLUMN chat_session_id TEXT",
-                    "ALTER TABLE poker_stack_log ADD COLUMN note TEXT"):
+                    "ALTER TABLE poker_stack_log ADD COLUMN note TEXT",
+                    # Nameless-villain identity (see docs/SCOUTING_DESK.md): a player
+                    # keyed by physical descriptors when no name is known.
+                    "ALTER TABLE poker_players ADD COLUMN descriptors TEXT",
+                    "ALTER TABLE poker_players ADD COLUMN descriptor_embedding BLOB",
+                    "ALTER TABLE poker_players ADD COLUMN distinctiveness REAL",
+                    "ALTER TABLE poker_players ADD COLUMN named INTEGER DEFAULT 1",
+                    # Embedded scar/confidence notes → "you've hit this leak before".
+                    "ALTER TABLE poker_rituals ADD COLUMN embedding BLOB"):
             try:
                 conn.execute(ddl)
             except Exception:
@@ -231,7 +278,7 @@ def delete_session(session_id: int) -> dict:
     counts: dict[str, int] = {}
     with conn:
         for t in ("poker_hands", "player_observations", "player_reads",
-                  "poker_stack_log", "poker_rituals"):
+                  "poker_stack_log", "poker_rituals", "session_players"):
             counts[t] = conn.execute(
                 f"SELECT COUNT(*) n FROM {t} WHERE session_id = ?", (session_id,)
             ).fetchone()["n"]
@@ -468,16 +515,50 @@ def log_ritual(kind: str, content: str | None = None, classification: str | None
     sid = _resolve(session_id)
     if sid is None:
         raise ValueError("no live session")
+    # Embed scar/confidence text so a similar spot later can recall it (Gap 3:
+    # "you've hit this leak before"). Other ritual kinds don't need it.
+    blob = None
+    if content and kind in ("scar", "confidence"):
+        vec = _embed_vec(content)
+        blob = memory._to_blob(vec.tolist()) if vec is not None else None
     conn = _c()
     with conn:
         cur = conn.execute(
-            "INSERT INTO poker_rituals (session_id, kind, content, classification, hand_id, created_at) "
-            "VALUES (?, ?, ?, ?, ?, ?)",
-            (sid, kind, content, classification, hand_id, _now()),
+            "INSERT INTO poker_rituals (session_id, kind, content, classification, hand_id, "
+            "embedding, created_at) VALUES (?, ?, ?, ?, ?, ?, ?)",
+            (sid, kind, content, classification, hand_id, blob, _now()),
         )
     return int(cur.lastrowid)
 
 
+def recall_similar_rituals(text: str, kinds: tuple[str, ...] = ("scar", "confidence"),
+                           k: int = 2, min_sim: float = 0.60,
+                           exclude_session: int | None = None) -> list[dict]:
+    """Past scar/confidence notes most similar to `text` — the leak/discipline
+    pattern recall. Excludes the current session so tonight doesn't echo itself."""
+    vec = _embed_vec(text)
+    if vec is None:
+        return []
+    ph = ",".join("?" * len(kinds))
+    rows = _c().execute(
+        f"SELECT r.id, r.kind, r.content, r.classification, r.session_id, r.embedding, "
+        f"s.started_at AS s_at, s.venue AS venue FROM poker_rituals r "
+        f"LEFT JOIN poker_sessions s ON s.id = r.session_id "
+        f"WHERE r.embedding IS NOT NULL AND r.kind IN ({ph})", tuple(kinds)
+    ).fetchall()
+    scored = []
+    for row in rows:
+        if exclude_session and row["session_id"] == exclude_session:
+            continue
+        sim = _cos(vec, memory._from_blob(row["embedding"]))
+        if sim >= min_sim:
+            scored.append((sim, row))
+    scored.sort(key=lambda x: x[0], reverse=True)
+    return [{"kind": r["kind"], "content": r["content"], "classification": r["classification"],
+             "when": r["s_at"], "venue": r["venue"], "sim": round(s, 3)}
+            for s, r in scored[:k]]
+
+
 def list_rituals(session_id: int | None = None,
                  kinds: tuple[str, ...] | None = None) -> list[dict]:
     """Ritual events for a session, oldest first; optionally filtered by kind."""
@@ -574,6 +655,28 @@ def update_hand(hand_id: int, **fields) -> dict | None:
     return get_hand(hand_id)
 
 
+def disown_hand(hand_id: int) -> dict | None:
+    """Reclassify a hand as OBSERVED (not Brian's) — fixes one that was mislabeled
+    as his. Clears the flat hero fields and rewrites the structured JSON with
+    hero_involved=false so the replay stops showing him as hero."""
+    h = get_hand(hand_id)
+    if not h:
+        return None
+    structured = h.get("structured")
+    if isinstance(structured, str):
+        structured = _safe_json(structured)
+    if isinstance(structured, dict):
+        structured = normalize_structured({**structured, "hero_involved": False})
+    conn = _c()
+    with conn:
+        conn.execute(
+            "UPDATE poker_hands SET position = NULL, hole_cards = NULL, result = NULL, "
+            "structured = ? WHERE id = ?",
+            (json.dumps(structured) if structured else None, hand_id),
+        )
+    return get_hand(hand_id)
+
+
 def list_hands(session_id: int | None = None) -> list[dict]:
     sid = _resolve(session_id)
     if sid is None:
@@ -592,8 +695,9 @@ Schema:
 {
   "game": "NLH" | "PLO" | ...,
   "stakes": "<e.g. 1/3, or null>",
-  "hero_pos": "<UTG|UTG1|MP|LJ|HJ|CO|BTN|SB|BB, hero's position>",
-  "hero_cards": ["As","Ax", ...],       // rank+suit (s/h/d/c); 'x' suit if unknown e.g. "Ax"; "x" for a fully unknown card
+  "hero_involved": true | false,        // is the narrator actually IN this hand? (see HERO rule)
+  "hero_pos": "<UTG|UTG1|MP|LJ|HJ|CO|BTN|SB|BB, hero's position — null if hero not involved>",
+  "hero_cards": ["As","Ax", ...],       // hero's cards; [] / null if hero not involved. rank+suit (s/h/d/c); 'x' suit if unknown e.g. "Ax"; "x" for a fully unknown card
   "players": [                          // every player mentioned, incl. hero
     {"pos": "<position>", "stack": <number|null>, "name": <string|null>, "cards": [".."]|null}
   ],
@@ -606,6 +710,17 @@ Schema:
   "result": {"pot": <number|null>, "hero_net": <number|null>, "summary": "<one line>"}
 }
 
+HERO — who the hand belongs to. The narrator writes as "I/me/my". Set \
+hero_involved=true and fill hero_pos + hero_cards ONLY if he was actually dealt in and \
+playing THIS hand. If he's describing a hand he WATCHED between other players — he wasn't \
+in it (e.g. "two guys got it in", "the neck-tattoo reg stacked the whale", it's all names/\
+seats with no I/me/my holding cards) — set hero_involved=false, hero_pos=null, hero_cards=null, \
+and just record those players in players[] and their action. NEVER attribute another player's \
+cards, position, or result to the hero. When hero isn't involved, result.hero_net=null (a pot \
+he wasn't in didn't win or lose him anything). Example: "the lag in the CO stacked off KK vs \
+the nit's AA" -> hero_involved=false, hero_pos=null, hero_cards=null, players=[{pos:CO,cards:\
+["Kx","Kx"]},{pos:?,cards:["Ax","Ax"]}].
+
 Rules: infer positions and street order sensibly. Amounts are plain numbers (no $). \
 NEVER invent suits or cards. A card is rank+suit where suit is one of s/h/d/c; if the suit \
 wasn't stated, use 'x' for the suit (e.g. "Ax","Kx","4x"); if a whole card wasn't stated, \
@@ -722,6 +837,13 @@ def normalize_structured(parsed: dict) -> dict:
         return parsed
     p = dict(parsed)
     p["schema_version"] = HAND_SCHEMA_VERSION
+    # Observed hand (Brian watched it, wasn't in it): never pin cards/position/result
+    # to the hero. Explicit false is the signal; a null hero_pos is treated the same.
+    if p.get("hero_involved") is False:
+        p["hero_pos"] = None
+        p["hero_cards"] = []
+        if isinstance(p.get("result"), dict):
+            p["result"] = {**p["result"], "hero_net": None}
     p["hero_cards"] = [_norm_card(c) for c in (p.get("hero_cards") or [])]
     p["board"] = [_norm_card(c) for c in (p.get("board") or [])]
 
@@ -1039,18 +1161,396 @@ def update_player(player_id: int, **fields) -> dict | None:
             conn.execute(f"UPDATE poker_players SET {', '.join(sets)} WHERE id = ?",
                          (*vals, player_id))
     row = _c().execute("SELECT * FROM poker_players WHERE id = ?", (player_id,)).fetchone()
-    return dict(row) if row else None
+    if not row:
+        return None
+    d = dict(row)
+    d.pop("descriptor_embedding", None)  # raw bytes — not JSON-serializable
+    return d
+
+
+# --- villain identity resolution (nameless villains; see docs/SCOUTING_DESK.md) ---
+#
+# Most live villains have no name — Brian knows them by a physical descriptor
+# ("neck tattoo guy"), a seat (within a session), or something too generic to be
+# an identifier. A descriptor is a *fuzzy* key: the same person gets phrased a
+# dozen ways, so we match by embedding, scoped by venue, and gated on how
+# distinctive the description is. Citing the WRONG villain is worse than silence,
+# so a generic-only description never resolves to a guess.
+
+# Features distinctive enough to anchor an identity (a near-unique key).
+_DISTINCTIVE = (
+    "tattoo", "tatted", "ink", "sleeve", "scar", "piercing", "mohawk", "dreads",
+    "dreadlocks", "braids", "ponytail", "cornrows", "durag", "bald", "goatee",
+    "cane", "wheelchair", "crutch", "jersey", "grill", "gold teeth", "eyepatch",
+    "birthmark", "mole", "cowboy hat", "fedora", "turban", "hijab", "accent",
+    "hearing aid", "prosthetic", "limp", "neck", "face", "hand tattoo", "beard",
+)
+# Words that describe half the room — near-zero discriminating power.
+_GENERIC = (
+    "guy", "dude", "man", "woman", "lady", "gentleman", "kid", "white", "black",
+    "asian", "hispanic", "latino", "indian", "old", "older", "young", "younger",
+    "middle", "mid", "aged", "40s", "50s", "30s", "60s", "20s", "glasses",
+    "average", "normal", "regular", "tall", "short", "heavy", "thin", "skinny",
+    "fat", "bigger", "plain", "shirt", "hoodie",
+)
+
+# Resolver thresholds (cosine sim on descriptor embeddings). Tunable.
+_SIM_HIGH = 0.80        # confident it's the same person
+_SIM_AMBIGUOUS = 0.58   # plausible — don't guess live, route to review
+_DISTINCT_MIN = 0.30    # below this the description is too generic to match at all
+
+
+_GENERIC_SET = frozenset(_GENERIC)
+
+
+def distinctiveness(text: str) -> float:
+    """How usable a description is as an identity key: ~1.0 for 'neck tattoo, Fox
+    Racing hat', ~0.1 for 'mid-aged white guy with glasses'. Generic-ONLY stays
+    near zero; specific content (named features, brands, a list) reads as high —
+    even if a generic word like 'shirt' is mixed in."""
+    t = (text or "").strip()
+    if not t:
+        return 0.0
+    low = t.lower()
+    tokens = re.findall(r"[a-z0-9']+", low)
+    dist = sum(1 for w in _DISTINCTIVE if w in low)
+    proper = len(re.findall(r"\b[A-Z][a-z]{2,}", text))     # brands/proper nouns: Fox, DKNY-ish
+    non_generic = sum(1 for w in tokens if w not in _GENERIC_SET)
+    # Only bland filler (age/race/build/gender) and nothing concrete → not usable.
+    specific = dist + proper + (1 if "," in t else 0)
+    if specific == 0 and non_generic <= 1:
+        return 0.10
+    return min(1.0, 0.40 + 0.14 * specific + 0.05 * non_generic)
+
+
+def _embed_vec(text: str):
+    try:
+        [v] = llm.embed([text])
+        return np.asarray(v, dtype=np.float32)
+    except Exception:
+        return None
+
+
+def _cos(a, b) -> float:
+    na, nb = float(np.linalg.norm(a)), float(np.linalg.norm(b))
+    if na == 0.0 or nb == 0.0:
+        return 0.0
+    return float(np.dot(a, b) / (na * nb))
+
+
+def _descriptor_candidates(vec, venue: str | None, exclude_id: int | None = None):
+    """Players with a descriptor embedding, scored by cosine to `vec`, best first.
+    Same-venue players are preferred (a small bonus) but not required."""
+    rows = _c().execute(
+        "SELECT id, name, venue, descriptors, descriptor_embedding FROM poker_players "
+        "WHERE descriptor_embedding IS NOT NULL"
+    ).fetchall()
+    out = []
+    for r in rows:
+        if exclude_id is not None and r["id"] == exclude_id:
+            continue
+        other = memory._from_blob(r["descriptor_embedding"])
+        sim = _cos(vec, other)
+        if venue and r["venue"] and venue.lower() == r["venue"].lower():
+            sim = min(1.0, sim + 0.05)  # same-room nudge
+        out.append({"id": r["id"], "name": r["name"], "venue": r["venue"],
+                    "descriptors": r["descriptors"], "sim": round(sim, 3)})
+    out.sort(key=lambda c: c["sim"], reverse=True)
+    return out
+
+
+def resolve_villain(ref: str, venue: str | None = None,
+                    session_id: int | None = None) -> dict:
+    """Resolve a reference to a villain. Returns {band, match_id, confidence, candidates}.
+
+    band: 'name' (exact name hit) | 'high' (confident descriptor match) |
+          'ambiguous' (plausible — don't guess live) | 'generic' (too vague to
+          match) | 'none' (new villain). The scouting desk / confirm loop act on
+          the band; they never silently trust an ambiguous or generic match."""
+    ref = (ref or "").strip()
+    empty = {"band": "none", "match_id": None, "confidence": 0.0, "candidates": []}
+    if not ref:
+        return empty
+
+    # 1) Exact name match against *named* players — deterministic, no guessing.
+    row = _c().execute(
+        "SELECT id FROM poker_players WHERE named = 1 AND name = ? COLLATE NOCASE", (ref,)
+    ).fetchone()
+    if row:
+        return {"band": "name", "match_id": row["id"], "confidence": 1.0, "candidates": []}
+
+    # 2) Descriptor match — but refuse if the description is too generic to key on.
+    dscore = distinctiveness(ref)
+    if dscore < _DISTINCT_MIN:
+        return {"band": "generic", "match_id": None, "confidence": 0.0, "candidates": []}
+
+    vec = _embed_vec(ref)
+    if vec is None:
+        return empty
+    cands = _descriptor_candidates(vec, venue)[:5]
+    best = cands[0]["sim"] if cands else 0.0
+    if best >= _SIM_HIGH:
+        band = "high"
+    elif best >= _SIM_AMBIGUOUS:
+        band = "ambiguous"
+    else:
+        band = "none"
+    return {"band": band, "match_id": cands[0]["id"] if cands and band in ("high", "ambiguous") else None,
+            "confidence": best, "candidates": cands}
+
+
+def create_descriptor_villain(descriptor: str, venue: str | None = None,
+                              category: str | None = None) -> int:
+    """Open a new nameless villain keyed on a physical descriptor. name holds the
+    descriptor label (so displays work); named=0 marks it as not-a-real-name."""
+    vec = _embed_vec(descriptor)
+    blob = memory._to_blob(vec.tolist()) if vec is not None else None
+    conn = _c()
+    with conn:
+        cur = conn.execute(
+            "INSERT INTO poker_players (name, venue, category, descriptors, "
+            "descriptor_embedding, distinctiveness, named, updated_at) "
+            "VALUES (?, ?, ?, ?, ?, ?, 0, ?)",
+            (descriptor.strip(), venue, category, descriptor.strip(), blob,
+             distinctiveness(descriptor), _now()),
+        )
+    return int(cur.lastrowid)
+
+
+def add_descriptor(player_id: int, descriptor: str) -> None:
+    """Fold another observed descriptor into a villain and re-embed the union, so
+    matching sharpens as more phrasings accumulate."""
+    row = _c().execute(
+        "SELECT descriptors FROM poker_players WHERE id = ?", (player_id,)
+    ).fetchone()
+    if not row:
+        return
+    merged = "; ".join(dict.fromkeys(
+        p.strip() for p in ((row["descriptors"] or "") + "; " + descriptor).split(";") if p.strip()
+    ))
+    vec = _embed_vec(merged)
+    blob = memory._to_blob(vec.tolist()) if vec is not None else None
+    conn = _c()
+    with conn:
+        conn.execute(
+            "UPDATE poker_players SET descriptors = ?, descriptor_embedding = ?, "
+            "distinctiveness = ?, updated_at = ? WHERE id = ?",
+            (merged, blob, distinctiveness(merged), _now(), player_id),
+        )
+
+
+def name_villain(player_id: int, name: str) -> None:
+    """Attach a real name to a descriptor villain (caught it off Bravo). Flips named=1."""
+    conn = _c()
+    with conn:
+        conn.execute(
+            "UPDATE poker_players SET name = ?, named = 1, updated_at = ? WHERE id = ?",
+            (name.strip(), _now(), player_id),
+        )
+
+
+def merge_players(keep_id: int, dup_id: int, note: str | None = None) -> bool:
+    """Confirmed same person: repoint dup's observations/reads onto keep, fold in its
+    descriptors, keep a real name over a descriptor label, then delete the dup."""
+    if keep_id == dup_id:
+        return False
+    conn = _c()
+    keep = conn.execute("SELECT * FROM poker_players WHERE id = ?", (keep_id,)).fetchone()
+    dup = conn.execute("SELECT * FROM poker_players WHERE id = ?", (dup_id,)).fetchone()
+    if not keep or not dup:
+        return False
+    keep, dup = dict(keep), dict(dup)
+    with conn:
+        conn.execute("UPDATE player_observations SET player_id = ? WHERE player_id = ?",
+                     (keep_id, dup_id))
+        conn.execute("UPDATE player_reads SET player_id = ? WHERE player_id = ?",
+                     (keep_id, dup_id))
+        # Prefer a real name; union the descriptor text.
+        new_name = keep["name"] if keep.get("named") else (dup["name"] if dup.get("named") else keep["name"])
+        named = 1 if (keep.get("named") or dup.get("named")) else 0
+        descs = "; ".join(dict.fromkeys(
+            p.strip() for p in ((keep.get("descriptors") or "") + "; " + (dup.get("descriptors") or "")).split(";")
+            if p.strip()
+        )) or None
+        conn.execute("UPDATE poker_players SET name = ?, named = ?, descriptors = ? WHERE id = ?",
+                     (new_name, named, descs, keep_id))
+        conn.execute("DELETE FROM poker_players WHERE id = ?", (dup_id,))
+        conn.execute("DELETE FROM identity_queue WHERE player_ids LIKE ? OR player_ids LIKE ?",
+                     (f"%{dup_id}%", f"%{keep_id}%"))
+    if descs:
+        add_descriptor(keep_id, "")  # re-embed the merged descriptor set
+    return True
+
+
+def mark_distinct(a_id: int, b_id: int, note: str | None = None) -> None:
+    """Record that two profiles are confirmed DIFFERENT people so the scan never
+    re-proposes the merge. Stored order-independent (min, max)."""
+    lo, hi = sorted((int(a_id), int(b_id)))
+    conn = _c()
+    with conn:
+        conn.execute(
+            "INSERT OR REPLACE INTO player_distinct_pairs (a_id, b_id, note, created_at) "
+            "VALUES (?, ?, ?, ?)", (lo, hi, note, _now()))
+        conn.execute("DELETE FROM identity_queue WHERE kind = 'merge_candidate' AND "
+                     "(player_ids = ? OR player_ids = ?)",
+                     (json.dumps([lo, hi]), json.dumps([hi, lo])))
+
+
+def are_distinct(a_id: int, b_id: int) -> bool:
+    lo, hi = sorted((int(a_id), int(b_id)))
+    return _c().execute(
+        "SELECT 1 FROM player_distinct_pairs WHERE a_id = ? AND b_id = ?", (lo, hi)
+    ).fetchone() is not None
+
+
+def queue_identity_task(kind: str, player_ids: list[int], descriptor: str | None = None,
+                        context: str | None = None, session_id: int | None = None,
+                        confidence: float | None = None) -> int | None:
+    """File an identity task for async review. De-dupes an identical pending task."""
+    ids_json = json.dumps(sorted(int(i) for i in player_ids)) if player_ids else None
+    conn = _c()
+    dup = conn.execute(
+        "SELECT id FROM identity_queue WHERE status = 'pending' AND kind = ? AND "
+        "IFNULL(player_ids,'') = IFNULL(?,'') AND IFNULL(descriptor,'') = IFNULL(?,'')",
+        (kind, ids_json, descriptor),
+    ).fetchone()
+    if dup:
+        return int(dup["id"])
+    with conn:
+        cur = conn.execute(
+            "INSERT INTO identity_queue (kind, player_ids, descriptor, context, session_id, "
+            "confidence, created_at) VALUES (?, ?, ?, ?, ?, ?, ?)",
+            (kind, ids_json, descriptor, context, session_id, confidence, _now()),
+        )
+    return int(cur.lastrowid)
+
+
+def list_identity_queue(status: str = "pending") -> list[dict]:
+    """Pending identity tasks, each with its candidate players hydrated for the UI."""
+    rows = _c().execute(
+        "SELECT * FROM identity_queue WHERE status = ? ORDER BY id DESC", (status,)
+    ).fetchall()
+    out = []
+    for r in rows:
+        d = dict(r)
+        ids = json.loads(d["player_ids"]) if d.get("player_ids") else []
+        d["players"] = [p for p in (_player_brief(i) for i in ids) if p]
+        out.append(d)
+    return out
+
+
+def _player_brief(player_id: int) -> dict | None:
+    r = _c().execute(
+        "SELECT id, name, venue, category, named, descriptors FROM poker_players WHERE id = ?",
+        (player_id,),
+    ).fetchone()
+    if not r:
+        return None
+    d = dict(r)
+    d["obs"] = _c().execute(
+        "SELECT COUNT(*) n FROM player_observations WHERE player_id = ?", (player_id,)
+    ).fetchone()["n"]
+    return d
+
+
+def resolve_identity_task(task_id: int, action: str, **kw) -> bool:
+    """Clear a queue task. action: 'merge' (kw keep_id,dup_id) | 'distinct'
+    (kw a_id,b_id,note) | 'name' (kw player_id,name) | 'dismiss'."""
+    if action == "merge":
+        merge_players(kw["keep_id"], kw["dup_id"], kw.get("note"))
+    elif action == "distinct":
+        mark_distinct(kw["a_id"], kw["b_id"], kw.get("note"))
+    elif action == "name":
+        name_villain(kw["player_id"], kw["name"])
+    conn = _c()
+    with conn:
+        conn.execute("UPDATE identity_queue SET status = 'resolved', resolution = ? WHERE id = ?",
+                     (action, task_id))
+    return True
+
+
+def scan_merge_candidates(sim_threshold: float = _SIM_HIGH) -> int:
+    """Off-hot-path (dream cycle): find pairs of profiles likely to be one person
+    and file merge_candidate tasks. Skips pairs already confirmed distinct. Returns
+    how many new candidates were filed."""
+    rows = _c().execute(
+        "SELECT id, venue, descriptor_embedding FROM poker_players "
+        "WHERE descriptor_embedding IS NOT NULL"
+    ).fetchall()
+    vecs = [(r["id"], (r["venue"] or "").lower(), memory._from_blob(r["descriptor_embedding"]))
+            for r in rows]
+    filed = 0
+    for i in range(len(vecs)):
+        for j in range(i + 1, len(vecs)):
+            aid, aven, av = vecs[i]
+            bid, bven, bv = vecs[j]
+            if aven and bven and aven != bven:
+                continue  # different rooms — leave cross-venue merges to a human
+            if are_distinct(aid, bid):
+                continue
+            sim = _cos(av, bv)
+            if sim >= sim_threshold:
+                if queue_identity_task("merge_candidate", [aid, bid],
+                                       context=f"descriptor similarity {sim:.2f}",
+                                       confidence=round(sim, 3)):
+                    filed += 1
+    return filed
+
+
+# Words that mark a "name" as really a physical description (misused name field).
+_DESC_MARKERS = (
+    "shirt", "hat", "cap", "hair", "beard", "glasses", "sunglasses", "tattoo",
+    "bracelet", "watch", "descent", "jersey", "hoodie", "jacket", "build",
+    "bald", "goatee", "chain", "necklace", "piercing", "mustache", "ponytail",
+    "sleeve", "skin", "wearing", "heavyset", "tall guy", "older", "younger",
+)
+
+
+def _looks_like_description(text: str | None) -> bool:
+    """A physical description mistakenly passed as a name — should be a descriptor.
+    Real handles are short (1-3 words, no commas); descriptions are longer / listy."""
+    t = (text or "").strip()
+    if not t:
+        return False
+    low = t.lower()
+    if "," in t or len(t.split()) > 4:
+        return True
+    return any(m in low for m in _DESC_MARKERS)
 
 
 def add_read(note: str, seat: str | None = None, name: str | None = None,
-             session_id: int | None = None, **player_fields) -> int:
-    """Log a live read. If `name` is given, upsert the player and link the read."""
+             descriptor: str | None = None, session_id: int | None = None,
+             **player_fields) -> int:
+    """Log a live read. `name` upserts a named player; `descriptor` (a nameless
+    villain's physical description) resolves to an existing descriptor villain when
+    confident, else opens a new one — so reads on unnamed players still accumulate."""
     sid = _resolve(session_id)
+    venue = player_fields.get("venue")
+    # A description passed as a name (e.g. "Filipino, Fox Racing hat, DKNY shirt")
+    # is really a descriptor — route it so it dedupes instead of spawning a new
+    # named player each time the wording drifts.
+    if name and not descriptor and _looks_like_description(name):
+        descriptor, name = name, None
     pid = None
     if name:
         pid = upsert_player(name, **{k: v for k, v in player_fields.items()
                                      if k in ("venue", "description", "tendencies",
                                               "adjustment", "category")})
+    elif descriptor:
+        res = resolve_villain(descriptor, venue=venue, session_id=sid)
+        if res["band"] in ("name", "high") and res["match_id"]:
+            pid = res["match_id"]
+            add_descriptor(pid, descriptor)  # sharpen the key with this phrasing
+        else:
+            pid = create_descriptor_villain(descriptor, venue=venue,
+                                            category=player_fields.get("category"))
+            # Plausibly the same guy as an existing villain, but not confident —
+            # surface it for a one-click merge instead of leaving a silent dup.
+            if res["band"] == "ambiguous" and res["match_id"]:
+                queue_identity_task("merge_candidate", [pid, res["match_id"]],
+                                    descriptor=descriptor,
+                                    context="similar description logged live",
+                                    confidence=res["confidence"])
     conn = _c()
     with conn:
         cur = conn.execute(
@@ -1161,6 +1661,59 @@ def get_villain_file(name: str | None = None, venue: str | None = None) -> list[
     return [dict(r) for r in _c().execute(sql, params).fetchall()]
 
 
+def players_overview() -> list[dict]:
+    """All villains for the browser: identity + observation count + last seen,
+    named players first, then most-recently-updated."""
+    rows = _c().execute(
+        "SELECT p.id, p.name, p.named, p.venue, p.category, p.descriptors, "
+        "p.tendencies, p.adjustment, p.updated_at, "
+        "(SELECT COUNT(*) FROM player_observations o WHERE o.player_id = p.id) AS obs, "
+        "(SELECT COUNT(*) FROM player_reads r WHERE r.player_id = p.id) AS reads "
+        "FROM poker_players p ORDER BY p.named DESC, p.updated_at DESC"
+    ).fetchall()
+    return [dict(r) for r in rows]
+
+
+def villain_recall(player_id: int) -> dict | None:
+    """Episodic recall for one villain: who, how often/where seen, last seen, the
+    notable hands against him (with ids to link), reads, and stats. This is the
+    when/where/which-hand narrative the scouting desk surfaces (Gap 1)."""
+    p = _c().execute("SELECT * FROM poker_players WHERE id = ?", (player_id,)).fetchone()
+    if not p:
+        return None
+    p = dict(p)
+    p.pop("descriptor_embedding", None)  # raw bytes — not JSON-serializable, not needed
+    obs = [dict(r) for r in _c().execute(
+        "SELECT o.*, s.venue AS s_venue, s.started_at AS s_at FROM player_observations o "
+        "LEFT JOIN poker_sessions s ON s.id = o.session_id WHERE o.player_id = ? "
+        "ORDER BY o.id DESC", (player_id,)
+    ).fetchall()]
+    reads = [dict(r) for r in _c().execute(
+        "SELECT note, created_at FROM player_reads WHERE player_id = ? ORDER BY id DESC LIMIT 8",
+        (player_id,)
+    ).fetchall()]
+    sessions_seen = sorted({o["session_id"] for o in obs if o.get("session_id")} |
+                           {r_["session_id"] for r_ in _c().execute(
+                               "SELECT session_id FROM player_reads WHERE player_id = ?",
+                               (player_id,)).fetchall() if r_["session_id"]})
+    last_at = max([o.get("s_at") or o.get("created_at") for o in obs] +
+                  [r["created_at"] for r in reads] + [p.get("updated_at")], default=None)
+    # Notable hands: showdowns / anything with cards, most recent first, linkable.
+    notable = [{"hand_id": o["hand_id"], "session_id": o.get("session_id"),
+                "when": o.get("s_at") or o.get("created_at"), "cards": o.get("cards"),
+                "summary": o.get("summary")}
+               for o in obs if o.get("hand_id")][:6]
+    prof = player_profile(p["name"]) or {}
+    return {
+        "player": p, "named": bool(p.get("named")),
+        "times_seen": len(sessions_seen), "sessions_seen": sessions_seen,
+        "last_seen": last_at, "notable_hands": notable,
+        "reads": [r["note"] for r in reads],
+        "stats": prof.get("stats"), "observations": len(obs),
+        "descriptors": p.get("descriptors"),
+    }
+
+
 # --- stats ---
 
 def session_stats(session_id: int | None = None) -> dict:
@@ -1261,6 +1814,114 @@ def timeline(session_id: int | None = None) -> list[dict]:
     return events
 
 
+def _resolve_or_create_player(name: str | None = None, descriptor: str | None = None,
+                              venue: str | None = None, category: str | None = None) -> int | None:
+    """Turn a name-or-descriptor into a player id, matching an existing villain when
+    confident. A description mistakenly given as a name is routed to the descriptor
+    path so it dedupes (same guard add_read uses)."""
+    if name and not descriptor and _looks_like_description(name):
+        descriptor, name = name, None
+    if name:
+        return upsert_player(name, venue=venue, category=category)
+    if descriptor:
+        res = resolve_villain(descriptor, venue=venue)
+        if res["band"] in ("name", "high") and res["match_id"]:
+            add_descriptor(res["match_id"], descriptor)
+            return res["match_id"]
+        return create_descriptor_villain(descriptor, venue=venue, category=category)
+    return None
+
+
+def seat_player(name: str | None = None, descriptor: str | None = None, seat: str | None = None,
+                category: str | None = None, session_id: int | None = None) -> int | None:
+    """Seat one player at the live table (add to the roster). Idempotent per session."""
+    sid = _resolve(session_id)
+    if sid is None:
+        raise ValueError("no live session")
+    venue = (get_session(sid) or {}).get("venue")
+    pid = _resolve_or_create_player(name=name, descriptor=descriptor, venue=venue, category=category)
+    if pid is None:
+        return None
+    conn = _c()
+    with conn:
+        conn.execute(
+            "INSERT INTO session_players (session_id, player_id, seat, active, created_at) "
+            "VALUES (?, ?, ?, 1, ?) ON CONFLICT(session_id, player_id) DO UPDATE SET "
+            "active = 1, seat = COALESCE(excluded.seat, session_players.seat)",
+            (sid, pid, seat, _now()),
+        )
+    return pid
+
+
+def seat_players(players: list, session_id: int | None = None) -> int:
+    """Seat a whole table at once. Each item is a name string or a dict with
+    name/descriptor/seat/category. Returns how many were seated."""
+    n = 0
+    for p in players or []:
+        if isinstance(p, str):
+            ok = seat_player(name=p, session_id=session_id)
+        elif isinstance(p, dict):
+            ok = seat_player(name=p.get("name"), descriptor=p.get("descriptor"),
+                             seat=p.get("seat"), category=p.get("category"), session_id=session_id)
+        else:
+            ok = None
+        if ok:
+            n += 1
+    return n
+
+
+def unseat_player(name: str | None = None, descriptor: str | None = None,
+                  session_id: int | None = None) -> bool:
+    """Mark a seated player as gone (busted/left). Keeps their reads/history."""
+    sid = _resolve(session_id)
+    if sid is None:
+        return False
+    ref = name or descriptor or ""
+    res = resolve_villain(ref, venue=(get_session(sid) or {}).get("venue"), session_id=sid)
+    pid = res.get("match_id")
+    if pid is None:
+        return False
+    conn = _c()
+    with conn:
+        conn.execute("UPDATE session_players SET active = 0 WHERE session_id = ? AND player_id = ?",
+                     (sid, pid))
+    return True
+
+
+def clear_roster(session_id: int | None = None) -> int:
+    """Empty the table roster (he changed tables) — unseat everyone at once. Keeps
+    the session and any reads logged; just resets who's currently seated. Returns
+    how many were cleared."""
+    sid = _resolve(session_id)
+    if sid is None:
+        return 0
+    conn = _c()
+    with conn:
+        cur = conn.execute(
+            "UPDATE session_players SET active = 0 WHERE session_id = ? AND active = 1", (sid,))
+    return cur.rowcount
+
+
+def session_roster(session_id: int | None = None) -> list[dict]:
+    """The live table roster: seated players with seat, dossier, and their latest
+    read this session. This is 'who's at the table right now'."""
+    sid = _resolve(session_id)
+    if sid is None:
+        return []
+    rows = _c().execute(
+        "SELECT sp.seat AS seat, p.id AS id, p.name AS name, p.named AS named, "
+        "p.category AS category, p.tendencies AS tendencies, "
+        "(SELECT note FROM player_reads r WHERE r.player_id = p.id AND r.session_id = ? "
+        " ORDER BY r.id DESC LIMIT 1) AS last_note, "
+        "(SELECT COUNT(*) FROM player_reads r2 WHERE r2.player_id = p.id AND r2.session_id = ?) AS reads "
+        "FROM session_players sp JOIN poker_players p ON p.id = sp.player_id "
+        "WHERE sp.session_id = ? AND sp.active = 1 "
+        "ORDER BY CASE WHEN sp.seat IS NULL THEN 1 ELSE 0 END, sp.seat, p.name",
+        (sid, sid, sid),
+    ).fetchall()
+    return [dict(r) for r in rows]
+
+
 def _session_villains(sid: int) -> list[dict]:
     """Players read this session, with their standing dossier fields."""
     rows = _c().execute(
@@ -1298,12 +1959,16 @@ def hud(session_id: int | None = None) -> dict | None:
         for h in list_hands(sid)
     ]
 
-    # Notes she jotted during this session: journal/note entries since it started.
-    started = s.get("started_at") or ""
+    # Her session narration: notes she took *for this session*, identified by the
+    # `poker:{id}` source tag stamped at write time (see tools._note) — NOT by a
+    # time window. Her autonomous journaling (dream-cycle reflections, thought
+    # loop, existential musings) has a different source, so it can never leak onto
+    # the poker HUD.
+    tag = f"poker:{sid}"
     notes = [
         {"created_at": j["created_at"], "kind": j["kind"], "content": j["content"]}
-        for j in memory.list_journal(kinds=("note", "journal"))
-        if (j["created_at"] or "") >= started
+        for j in memory.list_journal(kinds=("note",))
+        if (j.get("source") or "") == tag
     ][:20]
 
     stats = session_stats(sid)
@@ -1332,6 +1997,7 @@ def hud(session_id: int | None = None) -> dict | None:
             "log": log,
         },
         "hands": hands,
+        "roster": session_roster(sid),
         "villains": _session_villains(sid),
         "timeline": timeline(sid),
         "notes": notes,

lyra/scouting.py

@@ -0,0 +1,152 @@
+"""The scouting desk — proactive poker recall slid into Lyra's context before she
+replies, the way a broadcast stats desk hands the commentator a note.
+
+Two detectors run on the incoming message: known NAMES (deterministic) and
+physical DESCRIPTORS (fuzzy, via the identity resolver). A confident hit becomes a
+`SCOUTING DESK` system note she can cite; an ambiguous descriptor is filed to the
+review queue instead of interrupting. Everything here is best-effort and wrapped
+by the caller — it must never break a chat turn. Silence is the default.
+
+See docs/SCOUTING_DESK.md.
+"""
+from __future__ import annotations
+
+import re
+
+from lyra import clock, logbus, poker
+
+# Cues that a span names a *person at the table* worth resolving as a villain.
+_ROLE = r"(?:guy|dude|man|kid|reg|player|villain|fish|whale|nit|lag|tag|maniac)"
+_DESC_PATTERNS = (
+    re.compile(rf"\bthe ([\w][\w\s'’-]{{2,28}}?) {_ROLE}\b", re.I),
+    re.compile(rf"\b{_ROLE} (?:with|in|who has|sporting|rocking) (?:the |a |an )?([\w\s'’-]{{3,28}})", re.I),
+)
+_MIN_NAME = 3
+
+# Cues that a message is a strategy/spot/tilt discussion — the only turns worth
+# paying an embed to recall past leaks. Keeps the pattern pass off routine logging.
+_STRAT_CUES = (
+    "fold", "call", "raise", "bluff", "river", "turn", "flop", "tilt", "punt",
+    "leak", "should i", "hero", "value", "overbet", "spew", "stack off", "3bet",
+    "4bet", "check-raise", "checkraise", "range", "board", "steaming", "felted",
+    "all in", "all-in", "shoved", "jammed", "snap", "sizing",
+)
+
+
+def _looks_strategic(msg: str) -> bool:
+    low = msg.lower()
+    return len(msg) >= 40 and any(c in low for c in _STRAT_CUES)
+
+
+def _named_hits(msg: str) -> list[int]:
+    """Ids of known *named* villains whose name appears as a word in the message."""
+    low = msg.lower()
+    hits = []
+    for r in poker._c().execute("SELECT id, name FROM poker_players WHERE named = 1").fetchall():
+        name = (r["name"] or "").strip()
+        if len(name) < _MIN_NAME:
+            continue
+        if re.search(rf"\b{re.escape(name.lower())}\b", low):
+            hits.append(r["id"])
+    return hits
+
+
+def _descriptor_spans(msg: str) -> list[str]:
+    spans, seen = [], set()
+    for pat in _DESC_PATTERNS:
+        for m in pat.finditer(msg):
+            span = m.group(1).strip(" '’-").lower()
+            if span and span not in seen:
+                seen.add(span)
+                spans.append(span)
+    return spans
+
+
+def _brief(player_id: int) -> str | None:
+    """One compact line of episodic recall for a villain, or None if nothing known."""
+    rec = poker.villain_recall(player_id)
+    if not rec:
+        return None
+    p = rec["player"]
+    who = p["name"] if rec["named"] else f"“{p['name']}”"
+    bits = [who]
+    tags = [t for t in (p.get("venue"), p.get("category")) if t]
+    if tags:
+        bits.append("(" + ", ".join(tags) + ")")
+    if rec["times_seen"]:
+        seen = f"seen {rec['times_seen']}×"
+        if rec["last_seen"]:
+            seen += f", last {clock.short(rec['last_seen'])}"
+        bits.append(seen)
+    st = rec.get("stats")
+    if st:
+        bits.append(f"VPIP {st['vpip_pct']}/PFR {st['pfr_pct']} ({st['hands']}h)")
+    line = " ".join(bits)
+    if rec["reads"]:
+        line += " — reads: " + "; ".join(rec["reads"][:3])
+    if rec["notable_hands"]:
+        h = rec["notable_hands"][0]
+        line += f" · notable hand #{h['hand_id']}" + (f" ({h['cards']})" if h.get("cards") else "")
+    return line
+
+
+def scout(user_msg: str, venue: str | None = None, session_id: int | None = None) -> str | None:
+    """Build the SCOUTING DESK note for this message, or None. Never raises for a
+    caller that forgets to guard — but callers should guard anyway."""
+    try:
+        msg = (user_msg or "").strip()
+        if len(msg) < 3:
+            return None
+        if venue is None or session_id is None:
+            live = poker.live_session()
+            if live:
+                venue = venue or live.get("venue")
+                session_id = session_id or live.get("id")
+        lines: list[str] = []
+        seen_ids: set[int] = set()
+
+        for pid in _named_hits(msg):
+            if pid in seen_ids:
+                continue
+            b = _brief(pid)
+            if b:
+                lines.append(b)
+                seen_ids.add(pid)
+
+        for span in _descriptor_spans(msg):
+            res = poker.resolve_villain(span, venue=venue, session_id=session_id)
+            if res["band"] == "high" and res["match_id"] and res["match_id"] not in seen_ids:
+                b = _brief(res["match_id"])
+                if b:
+                    lines.append(b + "  ← confirm it's the same guy")
+                    seen_ids.add(res["match_id"])
+            elif res["band"] == "ambiguous" and res["match_id"]:
+                # Don't interrupt on a maybe — route it to the async review queue.
+                poker.queue_identity_task(
+                    "needs_clarification", [res["match_id"]], descriptor=span,
+                    context=f'Brian referred to "{span}"', session_id=session_id,
+                    confidence=res["confidence"])
+
+        # Pattern desk: on genuine strategy talk, recall his own past leaks/wins in
+        # similar spots. Gated so routine logging never pays for an embed.
+        pattern: list[str] = []
+        if _looks_strategic(msg):
+            for r in poker.recall_similar_rituals(msg, exclude_session=session_id):
+                tag = "leak" if r["kind"] == "scar" else "banked"
+                cls = f"/{r['classification']}" if r.get("classification") else ""
+                when = f" ({str(r['when'])[:10]})" if r.get("when") else ""
+                pattern.append(f"[{tag}{cls}]{when} {r['content']}")
+
+        if not lines and not pattern:
+            return None
+        logbus.log("info", "scouting desk", hits=len(lines), patterns=len(pattern))
+        out = "SCOUTING DESK — structured recall for his message (cite it, don't invent; " \
+              "if unsure it's the same person, ask him):"
+        if lines:
+            out += "\n• " + "\n• ".join(lines)
+        if pattern:
+            out += "\nYou've been in a spot like this before —\n• " + "\n• ".join(pattern)
+        return out
+    except Exception as exc:  # desk must never break a turn
+        logbus.log("error", "scouting desk failed", error=str(exc)[:160])
+        return None

lyra/summary.py

@@ -12,12 +12,41 @@ from __future__ import annotations
 import sys
 import threading
 import time
+from collections import Counter
 from concurrent.futures import ThreadPoolExecutor, as_completed
 
 from lyra import config, llm, logbus, memory
 from lyra.llm import Backend, Message
 
-_RETRIES = 4
+# Consolidation LLM budget. A gist is short (a handful of sentences), so cap the
+# generation hard — an uncapped local model will otherwise ramble for thousands
+# of tokens and, on a slow GPU, blow the request timeout. 768 is ~3x the longest
+# real gist we've stored.
+SUMMARY_MAX_TOKENS = 768
+# Attempts on the primary backend before falling back to cloud.
+MI50_ATTEMPTS = 2
+# Per-call timeout (seconds). A capped 768-token gist finishes in ~60-90s on the
+# MI50; 150s is headroom but bails a hung call fast so fallback isn't slow.
+SUMMARY_TIMEOUT = 150
+
+# Degenerate-output guard. A wedged local model (e.g. an overheated GPU) returns
+# a single character repeated ("?????") as a *successful* 200, which no timeout or
+# exception catches — so validate the text and treat junk as a failure. Real gists
+# are diverse prose; flag output whose most-common non-space char dominates. Short
+# outputs are exempt (nothing meaningful to judge).
+_DEGENERATE_MIN_CHARS = 24
+_DEGENERATE_CHAR_RATIO = 0.5
+
+
+class DegenerateOutput(RuntimeError):
+    """A backend returned junk (e.g. one char repeated) as a successful response."""
+
+
+def _looks_degenerate(text: str) -> bool:
+    stripped = "".join(text.split())
+    if len(stripped) < _DEGENERATE_MIN_CHARS:
+        return False
+    return max(Counter(stripped).values()) / len(stripped) > _DEGENERATE_CHAR_RATIO
 
 # Re-summarize a session once it has accumulated this many new raw exchanges.
 SUMMARIZE_AFTER = 20
@@ -61,16 +90,35 @@ def _summarize_text(text: str, backend: Backend) -> str:
         {"role": "system", "content": _PROMPT},
         {"role": "user", "content": text},
     ]
-    # Retry transient backend errors (e.g. the GPU server restarting) with backoff.
-    for attempt in range(_RETRIES):
+
+    def _call(be: Backend) -> str:
+        out = llm.complete(messages, backend=be,
+                           max_tokens=SUMMARY_MAX_TOKENS, timeout=SUMMARY_TIMEOUT)
+        if _looks_degenerate(out):
+            raise DegenerateOutput(f"{be} returned degenerate output ({len(out)} chars)")
+        return out
+
+    # Try the primary backend a bounded number of times (each call fast-fails via
+    # SUMMARY_TIMEOUT), with a short backoff for a transient blip / restarting GPU.
+    last_exc: Exception | None = None
+    for attempt in range(MI50_ATTEMPTS):
         try:
-            return llm.complete(messages, backend=backend)
+            return _call(backend)
         except Exception as exc:
-            if attempt == _RETRIES - 1:
-                raise
-            logbus.log("debug", "summary retry", attempt=attempt + 1, error=str(exc)[:80])
+            last_exc = exc
+            logbus.log("debug", "summary retry", attempt=attempt + 1,
+                       backend=backend, error=str(exc)[:80])
+            if attempt < MI50_ATTEMPTS - 1:
                 time.sleep(5 * (attempt + 1))
-    raise RuntimeError("unreachable")
+
+    # Primary exhausted. If it wasn't already cloud and cloud is configured, fall
+    # back once so a stuck/offline MI50 doesn't sink consolidation for the night.
+    if backend != "cloud" and config.load().openai_api_key:
+        logbus.log("info", "summary fell back to cloud", primary=backend,
+                   error=str(last_exc)[:80] if last_exc else None)
+        return _call("cloud")
+
+    raise last_exc if last_exc else RuntimeError("summary failed")
 
 
 def _summarize_transcript(transcript: str, backend: Backend) -> str:

lyra/tools.py

@@ -30,8 +30,13 @@ def _note(args: dict, ctx: dict) -> str:
         return "Nothing to note — content was empty."
     tag = (args.get("tag") or "").strip()
     stored = f"[{tag}] {content}" if tag else content
-    memory.add_journal_entry("note", stored, source="chat")
-    logbus.log("info", "Lyra noted (tool)", tag=tag or None)
+    # A note taken while a poker session is live is session narration — stamp it
+    # with the session so the HUD shows *only* these, never her autonomous
+    # journaling (dream-cycle musings, thought loop). Correctness by construction.
+    live = poker.live_session()
+    source = f"poker:{live['id']}" if live else "chat"
+    memory.add_journal_entry("note", stored, source=source)
+    logbus.log("info", "Lyra noted (tool)", tag=tag or None, poker=bool(live))
     return "Noted."
 
 
@@ -114,6 +119,9 @@ TOOLS: dict[str, dict] = {
                 "description": (
                     "Jot down a note to remember later — an observation, an idea, a "
                     "reminder, a read on a poker spot or opponent, anything worth keeping. "
+                    "During a live poker session this is your session log: a factual beat "
+                    "about how the night is going (table dynamics, Brian's arc, momentum) — "
+                    "it shows on his HUD. Not for your own feelings or reflection. "
                     "Optionally tag it (e.g. 'poker', 'idea', 'reminder')."
                 ),
                 "parameters": {
@@ -282,14 +290,87 @@ def _log_hand(args: dict, ctx: dict) -> str:
 def _add_read(args: dict, ctx: dict) -> str:
     poker.add_read(
         note=args.get("note") or "", seat=args.get("seat"), name=args.get("name"),
+        descriptor=args.get("descriptor"),
         tendencies=args.get("tendencies"), adjustment=args.get("adjustment"),
         description=args.get("description"), category=args.get("category"),
         venue=args.get("venue"),
     )
-    who = f" on {args['name']}" if args.get("name") else ""
+    who = f" on {args['name']}" if args.get("name") else (
+        f" on “{args['descriptor']}”" if args.get("descriptor") else "")
     return f"Read logged{who}."
 
 
+def _resolve_villain_ref(ref: str) -> tuple[int | None, str]:
+    """Resolve a name-or-descriptor to a single player id for a confirm-loop action.
+    Returns (id, band); acts only on a deterministic name or a confident descriptor."""
+    live = poker.live_session()
+    res = poker.resolve_villain(ref, venue=(live or {}).get("venue"),
+                               session_id=(live or {}).get("id"))
+    if res["band"] in ("name", "high") and res["match_id"]:
+        return res["match_id"], res["band"]
+    return None, res["band"]
+
+
+def _seat_players(args: dict, ctx: dict) -> str:
+    players = args.get("players") or []
+    # Accept a plain list of names too, for convenience.
+    if isinstance(players, str):
+        players = [p.strip() for p in re.split(r"[,\n]", players) if p.strip()]
+    try:
+        if args.get("replace"):     # a whole new table — wipe the roster first
+            poker.clear_roster()
+        n = poker.seat_players(players)
+    except ValueError:
+        return "No live session — start one first, then I'll seat the table."
+    roster = poker.session_roster()
+    names = ", ".join(r["name"] for r in roster) or "—"
+    return f"Seated {n}. Table now: {names}"
+
+
+def _clear_table(args: dict, ctx: dict) -> str:
+    n = poker.clear_roster()
+    return f"Table cleared — roster's empty ({n} removed). Tell me who's at the new one."
+
+
+def _unseat_player(args: dict, ctx: dict) -> str:
+    ok = poker.unseat_player(name=args.get("name"), descriptor=args.get("descriptor"))
+    who = args.get("name") or args.get("descriptor") or "player"
+    return f"{who} is off the table." if ok else f"Couldn't find {who} on the roster."
+
+
+def _name_villain(args: dict, ctx: dict) -> str:
+    ref = (args.get("descriptor") or "").strip()
+    name = (args.get("name") or "").strip()
+    if not ref or not name:
+        return "Need both the description of the player and the name to attach."
+    pid, band = _resolve_villain_ref(ref)
+    if pid is None:
+        return (f"Couldn't confidently find “{ref}” to name — too vague or no match. "
+                "Add a read with the descriptor first, or be more specific.")
+    poker.name_villain(pid, name)
+    return f"Got it — “{ref}” is {name} now; their history carries over."
+
+
+def _link_villains(args: dict, ctx: dict) -> str:
+    a = (args.get("player_a") or "").strip()
+    b = (args.get("player_b") or "").strip()
+    same = bool(args.get("same"))
+    if not a or not b:
+        return "Need two players to link (by name or description)."
+    ida, _ = _resolve_villain_ref(a)
+    idb, _ = _resolve_villain_ref(b)
+    if ida is None or idb is None:
+        return ("Couldn't confidently pin down both players, so I didn't merge anything — "
+                "safer to leave it. You can sort it on the Players page.")
+    if ida == idb:
+        return "Those resolve to the same profile already — nothing to do."
+    if same:
+        poker.merge_players(ida, idb)
+        return "Merged — same guy. Their histories are one file now."
+    poker.mark_distinct(ida, idb, note=args.get("note"))
+    return "Noted they're different people — I won't suggest merging them again."
+
+
 def _end_session(args: dict, ctx: dict) -> str:
     s = poker.end_session(cash_out=float(args.get("cash_out") or 0), mood=args.get("mood"))
     hourly = f", {s['net'] / s['hours']:+.0f}/hr" if s.get("hours") else ""
@@ -371,8 +452,16 @@ def _record_hand(args: dict, ctx: dict) -> str:
     if not out["id"]:
         return "I couldn't parse that hand — give it to me again with a little more detail?"
     p = out["parsed"]
+    hero_in = p.get("hero_involved") is not False and bool(p.get("hero_pos"))
+    logbus.log("info", "hand reconstructed", id=out["id"], hero=p.get("hero_pos"),
+               hero_involved=hero_in)
+    if not hero_in:
+        # A hand Brian watched between other players — not his.
+        who = ", ".join(pl.get("name") or pl.get("pos") or "?"
+                        for pl in (p.get("players") or [])[:3]) or "the table"
+        return (f"Logged hand #{out['id']} — an observed hand ({who}), not yours. "
+                f"View it at /hand/{out['id']}")
     cards = " ".join(p.get("hero_cards") or [])
-    logbus.log("info", "hand reconstructed", id=out["id"], hero=p.get("hero_pos"))
     return (f"Hand #{out['id']} reconstructed — {p.get('hero_pos') or '?'} "
             f"{cards}. View/replay it at /hand/{out['id']}")
 
@@ -577,9 +666,13 @@ TOOLS.update({
         [])},
     "add_read": {"handler": _add_read, "spec": _f(
         "add_read",
-        "Log a read on an opponent. If you give a name, it's saved to the persistent villain file.",
+        "Log a read on an opponent. Give a `name` if known; if not, give a `descriptor` "
+        "(a distinctive physical description like 'neck tattoo, backwards cap') and the read "
+        "attaches to that nameless player — reused automatically next time you describe him.",
         {"note": {**_S, "description": "The observation / what they showed down"},
          "name": {**_S, "description": "Player name/handle if known (creates/updates their dossier)"},
+         "descriptor": {**_S, "description": "Physical description when there's no name, e.g. "
+                        "'neck tattoo, heavyset'. Prefer distinctive features over generic ones."},
          "seat": {**_S, "description": "Seat or relative position"},
          "tendencies": {**_S, "description": "Standing read on how they play"},
          "adjustment": {**_S, "description": "How Brian should exploit them"},
@@ -587,6 +680,51 @@ TOOLS.update({
          "category": {**_S, "description": "feeder | risky | reg | unknown"},
          "venue": {**_S, "description": "Where they play"}},
         ["note"])},
+    "seat_players": {"handler": _seat_players, "spec": _f(
+        "seat_players",
+        "Register who's at the table this session — the roster Brian reads off the Bravo "
+        "screen (handles like TAG, JD). Call this when he names the table (usually at the "
+        "start) or when a new player sits. Each player is a real handle in `name`, or a "
+        "`descriptor` if he only describes them. These become the roster his reads/TAGs "
+        "attach to by name.",
+        {"players": {"type": "array", "description": "Players to seat",
+                     "items": {"type": "object", "properties": {
+                         "name": {**_S, "description": "Handle as it appears on Bravo, e.g. 'TAG'"},
+                         "descriptor": {**_S, "description": "Physical description if no name"},
+                         "seat": {**_S, "description": "Seat number/label if known"},
+                         "category": {**_S, "description": "feeder | risky | reg | unknown"}}}},
+         "replace": {"type": "boolean", "description": "true = a brand-new table: clear the "
+                     "current roster first, then seat these (use when he changes tables)"}},
+        ["players"])},
+    "unseat_player": {"handler": _unseat_player, "spec": _f(
+        "unseat_player",
+        "Remove a player from the table roster when they bust or leave. Keeps their history.",
+        {"name": {**_S, "description": "Their handle"},
+         "descriptor": {**_S, "description": "Or a description if unnamed"}},
+        [])},
+    "clear_table": {"handler": _clear_table, "spec": _f(
+        "clear_table",
+        "Empty the whole table roster at once — call this when Brian changes tables or says "
+        "to clear the table. The session, stack, and logged reads stay; only who's currently "
+        "seated resets. Then he'll tell you the new table.",
+        {}, [])},
+    "name_villain": {"handler": _name_villain, "spec": _f(
+        "name_villain",
+        "Attach a real name to a player you'd only known by description (e.g. you caught it "
+        "off the Bravo screen). Their whole history carries over to the name.",
+        {"descriptor": {**_S, "description": "How you'd been referring to him, e.g. 'neck tattoo guy'"},
+         "name": {**_S, "description": "His real name/handle"}},
+        ["descriptor", "name"])},
+    "link_villains": {"handler": _link_villains, "spec": _f(
+        "link_villains",
+        "Resolve a same-person question when Brian confirms it. same=true MERGES two profiles "
+        "into one (their histories join); same=false records they're DIFFERENT people so you "
+        "stop asking. Only call after he's confirmed — never merge on a guess.",
+        {"player_a": {**_S, "description": "First player, by name or description"},
+         "player_b": {**_S, "description": "Second player, by name or description"},
+         "same": {"type": "boolean", "description": "true = same person (merge); false = different"},
+         "note": {**_S, "description": "For different people: the tell that distinguishes them"}},
+        ["player_a", "player_b", "same"])},
     "end_session": {"handler": _end_session, "spec": _f(
         "end_session", "Close the live session: record cashout, compute net + hours.",
         {"cash_out": {**_N, "description": "Final cashout amount"},

lyra/transcript.py

@@ -0,0 +1,77 @@
+"""Full-fidelity conversation export: interleave what was *said* (chat exchanges)
+with what Lyra *did* (tool calls) in chronological order.
+
+The chat only ever lives in SQLite (`exchanges` + `tool_events`); this is the one
+place that renders a whole session back out as a portable artifact — Markdown for
+reading / pasting into RTO or another model, JSON for machine reprocessing.
+"""
+from __future__ import annotations
+
+import json
+
+from lyra import clock, memory
+
+# How roles/actions are labeled in the Markdown transcript.
+_SPEAKER = {"user": "Brian", "assistant": "Lyra"}
+
+
+def _merged(session_id: str) -> list[dict]:
+    """Speech + actions for a session, merged oldest-first by wall-clock time."""
+    events: list[dict] = []
+    for e in memory.history(session_id):
+        events.append({"type": "message", "role": e.role, "content": e.content,
+                       "ts": e.created_at})
+    for t in memory.tool_events(session_id):
+        events.append({"type": "tool", "tool": t["tool"], "args": t["args"],
+                       "result": t["result"], "ts": t["created_at"]})
+    # created_at is an ISO string; lexicographic sort == chronological sort.
+    events.sort(key=lambda ev: ev["ts"])
+    return events
+
+
+def _fmt_args(args) -> str:
+    """Compact one-line rendering of a tool call's arguments."""
+    if isinstance(args, dict):
+        return ", ".join(f"{k}={json.dumps(v, default=str)}" for k, v in args.items())
+    return "" if args is None else str(args)
+
+
+def as_markdown(session_id: str, name: str | None = None) -> str:
+    events = _merged(session_id)
+    title = name or session_id
+    lines = [f"# Conversation — {title}",
+             f"_Exported {clock.stamp()} · session `{session_id}` · "
+             f"{len(events)} events_", ""]
+    for ev in events:
+        stamp = clock.short(ev["ts"])
+        if ev["type"] == "message":
+            who = _SPEAKER.get(ev["role"], ev["role"].capitalize())
+            lines.append(f"**{who}** · {stamp}")
+            lines.append((ev["content"] or "").rstrip())
+            lines.append("")
+        else:
+            result = (ev["result"] or "").strip().replace("\n", " ")
+            if len(result) > 200:
+                result = result[:197] + "…"
+            lines.append(f"  ⚙ `{ev['tool']}({_fmt_args(ev['args'])})` → {result}")
+            lines.append("")
+    return "\n".join(lines).rstrip() + "\n"
+
+
+def as_json(session_id: str, name: str | None = None) -> dict:
+    return {
+        "session_id": session_id,
+        "name": name,
+        "exported_at": clock.stamp(),
+        "events": _merged(session_id),
+    }
+
+
+def build(session_id: str, fmt: str = "md", name: str | None = None):
+    """Return (content_str, media_type, filename) for the requested format."""
+    safe = "".join(c if c.isalnum() or c in "-_" else "_" for c in session_id)[:60]
+    if fmt == "json":
+        body = json.dumps(as_json(session_id, name), indent=2, ensure_ascii=False)
+        return body, "application/json", f"lyra_{safe}.json"
+    body = as_markdown(session_id, name)
+    return body, "text/markdown; charset=utf-8", f"lyra_{safe}.md"

lyra/web/server.py

@@ -18,7 +18,7 @@ from fastapi import FastAPI, Request, Response
 from fastapi.responses import FileResponse, StreamingResponse
 from fastapi.staticfiles import StaticFiles
 
-from lyra import chat, logbus, memory, modes, poker, self_state, summary, thoughts
+from lyra import chat, logbus, memory, modes, poker, self_state, summary, thoughts, transcript
 from lyra.llm import Backend
 
 
@@ -50,6 +50,16 @@ def _last_user_message(messages: list[dict]) -> str:
 def create_app() -> FastAPI:
     app = FastAPI(title="Lyra Web")
 
+    @app.middleware("http")
+    async def _no_stale_shell(request: Request, call_next):
+        """Always revalidate HTML/JS so a PWA can't serve a stale app shell after a
+        deploy (iOS applies heuristic caching when no cache header is set)."""
+        resp = await call_next(request)
+        ct = resp.headers.get("content-type", "")
+        if "text/html" in ct or "javascript" in ct:
+            resp.headers["Cache-Control"] = "no-cache, must-revalidate"
+        return resp
+
     @app.get("/_health")
     async def health() -> dict:
         return {"ok": True}
@@ -62,6 +72,15 @@ def create_app() -> FastAPI:
     async def get_session(session_id: str) -> list[dict]:
         return [{"role": ex.role, "content": ex.content} for ex in memory.history(session_id)]
 
+    @app.get("/sessions/{session_id}/export")
+    async def export_session(session_id: str, format: str = "md") -> Response:
+        """Full transcript — chat + interleaved tool calls — as Markdown or JSON."""
+        name = next((s["name"] for s in memory.list_sessions() if s["id"] == session_id), None)
+        body, media_type, filename = await asyncio.to_thread(
+            transcript.build, session_id, format, name)
+        return Response(content=body, media_type=media_type,
+                        headers={"Content-Disposition": f'attachment; filename="{filename}"'})
+
     @app.post("/sessions/{session_id}")
     async def save_session(session_id: str, request: Request) -> dict:
         # Messages are already persisted by chat.respond; just ensure the row exists.
@@ -183,6 +202,13 @@ def create_app() -> FastAPI:
         logbus.log("info", "hand edited", id=hand_id, fields=list(body))
         return {"ok": h is not None, "hand": h}
 
+    @app.post("/hand/{hand_id}/disown")
+    async def hand_disown(hand_id: int) -> dict:
+        """Reclassify a hand as observed (not Brian's) — fix a misattributed one."""
+        h = await asyncio.to_thread(poker.disown_hand, hand_id)
+        logbus.log("info", "hand disowned", id=hand_id)
+        return {"ok": h is not None, "hand": h}
+
     @app.delete("/hand/{hand_id}")
     async def hand_delete(hand_id: int) -> dict:
         """Delete a logged hand."""
@@ -203,9 +229,17 @@ def create_app() -> FastAPI:
 
     @app.patch("/player/{player_id}")
     async def player_update(player_id: int, request: Request) -> dict:
-        """Edit a player's dossier (rename, fix tendencies)."""
+        """Edit a player's dossier (rename, fix tendencies). Setting `name` on a
+        nameless (descriptor) villain promotes it to a real handle (named=1)."""
         body = await request.json()
-        p = await asyncio.to_thread(lambda: poker.update_player(player_id, **body))
+
+        def _apply():
+            if body.get("name"):
+                poker.name_villain(player_id, body["name"])
+            rest = {k: v for k, v in body.items() if k != "name"}
+            return poker.update_player(player_id, **rest)  # flat row (name included)
+
+        p = await asyncio.to_thread(_apply)
         logbus.log("info", "player edited", id=player_id, fields=list(body))
         return {"ok": p is not None, "player": p}
 
@@ -427,6 +461,34 @@ def create_app() -> FastAPI:
     async def hands_data(limit: int = 60) -> dict:
         return {"hands": poker.list_recent_hands(limit=limit)}
 
+    @app.get("/players")
+    async def players_page() -> FileResponse:
+        """Villain file browser + the identity-resolution review queue."""
+        return FileResponse(str(_STATIC / "players.html"))
+
+    @app.get("/players/data")
+    async def players_data() -> dict:
+        return {"players": poker.players_overview(),
+                "queue": poker.list_identity_queue()}
+
+    @app.get("/player/{player_id}/data")
+    async def player_data(player_id: int) -> dict:
+        return poker.villain_recall(player_id) or {}
+
+    @app.post("/identity/{task_id}/resolve")
+    async def identity_resolve(task_id: int, request: Request) -> dict:
+        body = await request.json()
+        action = body.get("action") or "dismiss"
+        kw = {k: v for k, v in body.items() if k != "action"}
+        ok = await asyncio.to_thread(poker.resolve_identity_task, task_id, action, **kw)
+        logbus.log("info", "identity task resolved", id=task_id, action=action)
+        return {"ok": ok}
+
+    @app.post("/players/scan")
+    async def players_scan() -> dict:
+        filed = await asyncio.to_thread(poker.scan_merge_candidates)
+        return {"ok": True, "filed": filed}
+
     @app.get("/recap/{session_id}")
     async def recap_page() -> FileResponse:
         return FileResponse(str(_STATIC / "recap.html"))

lyra/web/static/hand.html

@@ -282,8 +282,54 @@
         const h = await r.json();
         if(!h || !h.id){ document.getElementById('root').innerHTML='<p class="err">Hand not found.</p>'; return; }
         render(h);
+        renderEditor(h);
       }catch(e){ document.getElementById('root').innerHTML='<p class="err">Couldn\'t load the hand.</p>'; }
     }
+
+    function renderEditor(h){
+      const wrap = document.createElement('div');
+      wrap.style.cssText = 'max-width:520px;margin:18px auto 0;border-top:1px solid #241a10;padding-top:12px;';
+      const tags = ['','well_played','leak','cooler','confidence','notable'];
+      wrap.innerHTML = `
+        <details style="font-size:.9rem;">
+          <summary style="cursor:pointer;color:var(--accent,#ff7a00);">✎ Edit this hand</summary>
+          <div style="display:flex;flex-direction:column;gap:8px;margin-top:10px;">
+            <label>Position <input id="e_pos" value="${esc(h.position||'')}" placeholder="e.g. CO (blank if not yours)"></label>
+            <label>Your cards <input id="e_hole" value="${esc(h.hole_cards||'')}" placeholder="e.g. As Ks (blank if not yours)"></label>
+            <label>Board <input id="e_board" value="${esc(h.board||'')}" placeholder="e.g. Tc 8s Js 6d"></label>
+            <label>Your net <input id="e_res" value="${h.result!=null?esc(h.result):''}" placeholder="+ / − chips (blank if not yours)"></label>
+            <label>Tag <select id="e_tag">${tags.map(t=>`<option value="${t}" ${h.tag===t?'selected':''}>${t||'—'}</option>`).join('')}</select></label>
+            <label>Lesson <input id="e_lesson" value="${esc(h.lesson||'')}"></label>
+            <div style="display:flex;flex-wrap:wrap;gap:8px;margin-top:4px;">
+              <button onclick="saveHand(${h.id})" style="border-color:var(--accent,#ff7a00);color:var(--accent,#ff7a00);">Save</button>
+              <button onclick="disown(${h.id})" title="It was someone else's hand — clear it from you">Not my hand</button>
+              <button onclick="delHand(${h.id})" style="margin-left:auto;color:#ff6b6b;">Delete</button>
+            </div>
+          </div>
+        </details>`;
+      wrap.querySelectorAll('input,select').forEach(el=>{el.style.cssText='font:inherit;font-size:.86rem;padding:5px 8px;border-radius:6px;border:1px solid #241a10;background:#0b0b0b;color:#e8e8e8;margin-left:8px;';});
+      wrap.querySelectorAll('label').forEach(el=>{el.style.cssText='display:flex;justify-content:space-between;align-items:center;color:#8a8a8a;';});
+      wrap.querySelectorAll('button').forEach(el=>{el.style.cssText+=';font:inherit;font-size:.84rem;padding:6px 12px;border-radius:7px;border:1px solid #241a10;background:#141414;color:#e8e8e8;cursor:pointer;';});
+      document.getElementById('root').appendChild(wrap);
+    }
+    const val = id => document.getElementById(id).value.trim();
+    async function saveHand(id){
+      const body = {position:val('e_pos'), hole_cards:val('e_hole'), board:val('e_board'),
+                    tag:val('e_tag'), lesson:val('e_lesson')};
+      const res = val('e_res'); if(res!=='') body.result = Number(res);
+      await fetch(`/hand/${id}`,{method:'PATCH',headers:{'Content-Type':'application/json'},body:JSON.stringify(body)});
+      load();
+    }
+    async function disown(id){
+      if(!confirm("Mark this as someone else's hand? It'll be cleared from your stats.")) return;
+      await fetch(`/hand/${id}/disown`,{method:'POST'});
+      load();
+    }
+    async function delHand(id){
+      if(!confirm('Delete this hand for good?')) return;
+      await fetch(`/hand/${id}`,{method:'DELETE'});
+      location.href='/hands';
+    }
     load();
   </script>
   <script src="/nav.js"></script>

lyra/web/static/index.html

@@ -88,6 +88,7 @@
       <select id="sessions"></select>
       <button id="newSessionBtn">➕ New</button>
       <button id="renameSessionBtn">✏️ Rename</button>
+      <button id="exportSessionBtn" title="Download full transcript (chat + tool calls)">⬇ Export</button>
       <button id="thinkingStreamBtn" title="Show live activity log">📜 Live Log</button>
     </div>
 
@@ -987,6 +988,19 @@
 	  addMessage("system", `Session renamed to: ${newName}`);
 	});
 
+	document.getElementById("exportSessionBtn").addEventListener("click", () => {
+	  if (!currentSession) { addMessage("system", "No session to export."); return; }
+	  const fmt = window.confirm("Export as Markdown? (Cancel = JSON)") ? "md" : "json";
+	  // Hitting the download endpoint navigates a hidden anchor so the browser
+	  // saves the file (chat + interleaved tool calls) instead of rendering it.
+	  const a = document.createElement("a");
+	  a.href = `${RELAY_BASE}/sessions/${encodeURIComponent(currentSession)}/export?format=${fmt}`;
+	  a.download = "";
+	  document.body.appendChild(a);
+	  a.click();
+	  a.remove();
+	});
+
       // Settings Modal
       const settingsModal = document.getElementById("settingsModal");
       const settingsBtn = document.getElementById("settingsBtn");

lyra/web/static/nav.js

@@ -8,6 +8,7 @@
     { href: "/session", icon: "♠",  label: "Session" },
     { href: "/history", icon: "📚", label: "History" },
     { href: "/hands",   icon: "🃏", label: "Hands" },
+    { href: "/players", icon: "👤", label: "Players" },
     { href: "/self",    icon: "🧠", label: "Mind" },
     { href: "/thoughts", icon: "💭", label: "Thoughts" },
     { href: "/journal", icon: "📔", label: "Journal" },

lyra/web/static/players.html

@@ -0,0 +1,166 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+  <meta charset="UTF-8" />
+  <meta name="viewport" content="width=device-width, initial-scale=1.0, viewport-fit=cover" />
+  <meta name="theme-color" content="#070707" />
+  <title>Lyra — Players</title>
+  <style>
+    :root{--bg:#070707;--bg-elev:#0e0e0e;--bg-line:#141414;--border:#2a1d12;--text:#e8e8e8;--fade:#8a8a8a;--accent:#ff7a00;}
+    *{box-sizing:border-box;}
+    html,body{margin:0;min-height:100%;background:var(--bg);color:var(--text);
+      font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,sans-serif;-webkit-text-size-adjust:100%;}
+    header{position:sticky;top:0;z-index:10;background:var(--bg-elev);border-bottom:1px solid var(--border);
+      padding:env(safe-area-inset-top) 14px 0;}
+    .topbar{display:flex;align-items:center;gap:10px;padding:13px 0;}
+    .topbar h1{font-size:1.05rem;margin:0;font-weight:600;}
+    .topbar a.back{color:var(--accent);text-decoration:none;font-size:.92rem;}
+    .count{margin-left:auto;color:var(--fade);font-size:.8rem;}
+    main{max-width:640px;margin:0 auto;padding:12px 12px 44px;}
+    h2.sec{font-size:.74rem;text-transform:uppercase;letter-spacing:.6px;color:var(--fade);margin:20px 2px 8px;}
+    .queue{background:#160d05;border:1px solid var(--accent);border-radius:10px;padding:11px 12px;margin-bottom:9px;}
+    .queue .k{font-size:.62rem;text-transform:uppercase;letter-spacing:.5px;color:var(--accent);}
+    .queue .q-body{font-size:.9rem;margin:5px 0 9px;}
+    .queue .who{font-weight:600;}
+    .btns{display:flex;flex-wrap:wrap;gap:7px;}
+    button{font:inherit;font-size:.82rem;padding:6px 11px;border-radius:7px;border:1px solid var(--border);
+      background:var(--bg-line);color:var(--text);cursor:pointer;}
+    button.pri{border-color:var(--accent);color:var(--accent);}
+    button:active{background:#241400;}
+    .card{background:var(--bg-elev);border:1px solid var(--border);border-radius:10px;padding:10px 12px;margin-bottom:8px;}
+    .card .row{display:flex;align-items:center;gap:9px;cursor:pointer;}
+    .nm{font-size:.96rem;font-weight:600;}
+    .nm.desc{font-weight:500;font-style:italic;color:#e8d3bf;}
+    .meta{font-size:.74rem;color:var(--fade);}
+    .pill{font-size:.6rem;text-transform:uppercase;letter-spacing:.4px;border:1px solid var(--border);
+      border-radius:20px;padding:1px 7px;color:var(--fade);}
+    .pill.desc{border-color:#5a3c1e;color:#d0a56e;}
+    .spacer{margin-left:auto;}
+    .detail{margin-top:9px;padding-top:9px;border-top:1px solid var(--bg-line);font-size:.86rem;display:none;}
+    .detail.open{display:block;}
+    .detail .lbl{color:var(--fade);font-size:.72rem;text-transform:uppercase;letter-spacing:.4px;margin:8px 0 3px;}
+    .detail ul{margin:3px 0;padding-left:18px;} .detail li{margin:2px 0;}
+    .detail a{color:var(--accent);text-decoration:none;}
+    .edit{display:flex;flex-wrap:wrap;gap:6px;margin-top:9px;}
+    .edit input,.edit select{font:inherit;font-size:.82rem;padding:5px 8px;border-radius:6px;
+      border:1px solid var(--border);background:var(--bg);color:var(--text);}
+    .empty{color:var(--fade);text-align:center;padding:34px 16px;}
+  </style>
+</head>
+<body>
+  <header>
+    <div class="topbar">
+      <h1>👤 Players</h1>
+      <a class="back" href="/">← Chat</a>
+      <span class="count" id="count"></span>
+    </div>
+  </header>
+  <main id="root"><p class="empty">Loading…</p></main>
+
+  <script>
+    function esc(s){const d=document.createElement('div');d.textContent=s==null?'':String(s);return d.innerHTML;}
+    let DATA={players:[],queue:[]};
+
+    async function load(){
+      try{ DATA=await (await fetch('/players/data',{cache:'no-store'})).json(); }
+      catch(e){ document.getElementById('root').innerHTML='<p class="empty">Couldn\'t load players.</p>'; return; }
+      render();
+    }
+
+    function render(){
+      const {players,queue}=DATA;
+      const named=players.filter(p=>p.named), nameless=players.filter(p=>!p.named);
+      document.getElementById('count').textContent=`${players.length} player${players.length===1?'':'s'}`;
+      let html='';
+
+      if(queue.length){
+        html+=`<h2 class="sec">⚠ Needs your call — ${queue.length}</h2>`;
+        html+=queue.map(qCard).join('');
+      }
+      html+=`<h2 class="sec">Named — ${named.length} <button class="pri" style="float:right;padding:3px 9px" onclick="scan()">Scan for dupes</button></h2>`;
+      html+= named.length ? named.map(pCard).join('') : '<p class="empty">No named players yet.</p>';
+      html+=`<h2 class="sec">By description — ${nameless.length}</h2>`;
+      html+= nameless.length ? nameless.map(pCard).join('') : '<p class="empty">No nameless villains yet — they show up here as you describe players at the table.</p>';
+      document.getElementById('root').innerHTML=html;
+    }
+
+    function qCard(q){
+      const ps=q.players||[];
+      if(q.kind==='merge_candidate' && ps.length===2){
+        const a=ps[0], b=ps[1];
+        return `<div class="queue"><div class="k">Possible merge${q.confidence?` · ${Math.round(q.confidence*100)}%`:''}</div>
+          <div class="q-body">Same person? <span class="who">${label(a)}</span> &nbsp;vs&nbsp; <span class="who">${label(b)}</span></div>
+          <div class="btns">
+            <button class="pri" onclick="resolveTask(${q.id},'merge',{keep_id:${keepId(a,b)},dup_id:${dupId(a,b)}})">✓ Same — merge</button>
+            <button onclick="resolveTask(${q.id},'distinct',{a_id:${a.id},b_id:${b.id}})">✕ Different</button>
+            <button onclick="resolveTask(${q.id},'dismiss',{})">Dismiss</button>
+          </div></div>`;
+      }
+      const who=ps[0]?label(ps[0]):'?';
+      return `<div class="queue"><div class="k">Needs clarification</div>
+        <div class="q-body">You referred to <span class="who">“${esc(q.descriptor||'')}”</span>${ps[0]?` — is that ${who}?`:''}</div>
+        <div class="btns"><button onclick="resolveTask(${q.id},'dismiss',{})">Got it</button></div></div>`;
+    }
+    const label=p=>`${esc(p.name)}${p.named?'':' <span class="pill desc">desc</span>'}${p.venue?` · ${esc(p.venue)}`:''}${p.obs?` · ${p.obs}h`:''}`;
+    const keepId=(a,b)=>a.named?a.id:(b.named?b.id:a.id);
+    const dupId=(a,b)=>a.named?b.id:(b.named?a.id:b.id);
+
+    function pCard(p){
+      const pills=[p.named?'':'<span class="pill desc">desc</span>',p.category?`<span class="pill">${esc(p.category)}</span>`:''].join('');
+      const meta=[p.venue,p.obs?`${p.obs} hands`:'',p.reads?`${p.reads} reads`:''].filter(Boolean).join(' · ');
+      return `<div class="card" id="p${p.id}">
+        <div class="row" onclick="toggle(${p.id})">
+          <span class="nm ${p.named?'':'desc'}">${p.named?esc(p.name):'“'+esc(p.name)+'”'}</span>
+          ${pills}<span class="spacer"></span><span class="meta">${esc(meta)}</span>
+        </div>
+        <div class="detail" id="d${p.id}"></div></div>`;
+    }
+
+    async function toggle(id){
+      const el=document.getElementById('d'+id);
+      if(el.classList.contains('open')){el.classList.remove('open');return;}
+      el.classList.add('open'); el.innerHTML='<span class="meta">Loading…</span>';
+      const r=await (await fetch(`/player/${id}/data`,{cache:'no-store'})).json();
+      el.innerHTML=detailHtml(id,r);
+    }
+
+    function detailHtml(id,r){
+      const p=r.player||{}; let h='';
+      const seen=[r.times_seen?`seen ${r.times_seen}×`:'', r.last_seen?`last ${String(r.last_seen).slice(0,10)}`:''].filter(Boolean).join(' · ');
+      if(seen) h+=`<div class="meta">${esc(seen)}</div>`;
+      if(r.stats) h+=`<div class="lbl">Stats</div><div>VPIP ${r.stats.vpip_pct} · PFR ${r.stats.pfr_pct} · WTSD ${r.stats.wtsd_pct} <span class="meta">(${r.stats.hands} hands)</span></div>`;
+      if(r.descriptors) h+=`<div class="lbl">Descriptors</div><div>${esc(r.descriptors)}</div>`;
+      if(p.tendencies) h+=`<div class="lbl">Tendencies</div><div>${esc(p.tendencies)}</div>`;
+      if(p.adjustment) h+=`<div class="lbl">Exploit</div><div>${esc(p.adjustment)}</div>`;
+      if((r.reads||[]).length){h+='<div class="lbl">Reads</div><ul>'+r.reads.slice(0,8).map(x=>`<li>${esc(x)}</li>`).join('')+'</ul>';}
+      if((r.notable_hands||[]).length){h+='<div class="lbl">Notable hands</div><ul>'+r.notable_hands.map(x=>`<li><a href="/hand/${x.hand_id}">hand #${x.hand_id}</a>${x.cards?' — '+esc(x.cards):''}${x.summary?' <span class="meta">'+esc(x.summary)+'</span>':''}</li>`).join('')+'</ul>';}
+      h+=`<div class="edit">
+        ${p.named?'':`<input id="nm${id}" placeholder="give a name…" size="12"><button onclick="rename(${id})">Name</button>`}
+        <select id="cat${id}" onchange="setCat(${id})">
+          ${['','feeder','risky','reg','unknown'].map(c=>`<option value="${c}" ${p.category===c?'selected':''}>${c||'category…'}</option>`).join('')}
+        </select></div>`;
+      return h;
+    }
+
+    async function rename(id){
+      const v=document.getElementById('nm'+id).value.trim(); if(!v)return;
+      await fetch(`/player/${id}`,{method:'PATCH',headers:{'Content-Type':'application/json'},body:JSON.stringify({name:v})});
+      load();
+    }
+    async function setCat(id){
+      const v=document.getElementById('cat'+id).value;
+      await fetch(`/player/${id}`,{method:'PATCH',headers:{'Content-Type':'application/json'},body:JSON.stringify({category:v})});
+    }
+    async function resolveTask(id,action,kw){
+      await fetch(`/identity/${id}/resolve`,{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({action,...kw})});
+      load();
+    }
+    async function scan(){
+      const r=await (await fetch('/players/scan',{method:'POST'})).json();
+      load();
+    }
+    load();
+  </script>
+  <script src="/nav.js"></script>
+</body>
+</html>

lyra/web/static/session.html

@@ -265,6 +265,7 @@
       const stack = data.stack || {};
       const timeline = data.timeline || [];
       const hands = data.hands || [];
+      const roster = data.roster || [];
       const villains = data.villains || [];
       const notes = data.notes || [];
       const stats = data.stats || {};
@@ -369,6 +370,19 @@
             : '<p class="empty">No scars logged — mistakes to study land here.</p>'}
         </div>
 
+        <div class="card">
+          <p class="label">🪑 Table (${roster.length})</p>
+          ${roster.length ? `<ul class="rows">${roster.map(v => `
+            <li class="villain">
+              ${v.seat ? `<span class="cat">${esc(v.seat)}</span> ` : ''}<b>${esc(v.name)}</b>
+              ${v.category ? `<span class="cat">[${esc(v.category)}]</span>` : ''}
+              ${v.reads ? `<span class="cat">· ${v.reads} read${v.reads===1?'':'s'}</span>` : ''}
+              <button class="mini" title="Rename / fix" onclick="renamePlayer(${v.id}, '${esc(v.name||'').replace(/'/g,"\\'")}')">✎</button>
+              ${v.last_note ? `<div class="note-meta">“${esc(v.last_note)}”</div>` : ''}
+            </li>`).join('')}</ul>`
+            : '<p class="empty">No roster yet — tell Lyra who is at the table.</p>'}
+        </div>
+
         <div class="card">
           <p class="label">Villains seen</p>
           ${villains.length ? `<ul class="rows">${villains.map(v => `

tests/test_dream.py

@@ -20,7 +20,7 @@ def lyra(tmp_path, monkeypatch):
     # reflect() expects JSON back; everything else just stores the text.
     monkeypatch.setattr(
         llm, "complete",
-        lambda messages, backend=None, model=None:
+        lambda messages, backend=None, model=None, **_:
         '{"mood":"focused","valence":0.7,"new_reflections":["I got some thinking done."]}',
     )
 
@@ -77,3 +77,31 @@ def test_dream_cycle_consolidates_and_persists(lyra):
     state2 = dream.dream_cycle(force=False)
     assert state2["dream"]["cycle_count"] == 2
     assert state2["drives"]["continuity"] == 0.0
+
+
+def test_dream_cycle_stops_when_over_budget(lyra, monkeypatch):
+    memory = lyra
+    from lyra import dream, notify
+
+    for k in range(7):
+        _seed(memory, f"s{k}", 4)
+
+    # Go over budget right after the first heavy stage: first check passes
+    # (summarize runs), every check after trips.
+    checks = {"n": 0}
+
+    def fake_over(deadline):
+        checks["n"] += 1
+        return checks["n"] > 1
+    monkeypatch.setattr(dream, "_over_budget", fake_over)
+
+    pings: list = []
+    monkeypatch.setattr(notify, "push",
+                        lambda title, message, **k: pings.append((title, message)) or True)
+
+    state = dream.dream_cycle(force=True)
+    acts = state["dream"]["last_actions"]
+
+    assert any("stopped early" in a for a in acts)   # bailed
+    assert not any("reflected" in a for a in acts)   # later stage skipped
+    assert pings, "expected an over-budget ntfy push"

tests/test_hand_schema.py

@@ -45,6 +45,29 @@ def test_stamps_version(poker):
     assert out["schema_version"] == poker.HAND_SCHEMA_VERSION
 
 
+def test_observed_hand_never_attributed_to_hero(poker):
+    # Brian narrated a hand between two other players — hero_involved=false.
+    out = poker.normalize_structured({
+        "hero_involved": False,
+        "hero_pos": "CO", "hero_cards": ["Kx", "Kx"],   # model slipped these in
+        "players": [{"pos": "CO", "cards": ["Kx", "Kx"]}, {"pos": "BB", "cards": ["Ax", "Ax"]}],
+        "result": {"pot": 600, "hero_net": 300},
+    })
+    assert out["hero_pos"] is None                       # not pinned to Brian
+    assert out["hero_cards"] == []
+    assert out["result"]["hero_net"] is None             # a pot he wasn't in
+    assert not any(pl.get("hero") for pl in out["players"])  # nobody flagged hero
+
+
+def test_hero_hand_still_attributed(poker):
+    out = poker.normalize_structured({
+        "hero_involved": True, "hero_pos": "BTN", "hero_cards": ["As", "Ks"],
+        "players": [{"pos": "BTN"}]})
+    assert out["hero_pos"] == "BTN"
+    hero = next(pl for pl in out["players"] if pl.get("pos") == "BTN")
+    assert hero.get("hero") and hero["cards"] == ["As", "Ks"]
+
+
 def test_card_normalization(poker):
     out = poker.normalize_structured(_full_hand())
     assert out["hero_cards"] == ["Ah", "Kh"]            # lowercased input -> canonical

tests/test_llm_bounds.py

@@ -0,0 +1,66 @@
+"""llm.complete: `max_tokens` and `timeout` are threaded into the backend call.
+
+The OpenAI client is faked so nothing hits a network. We assert the generation
+cap reaches the create() call and the fast-fail timeout reaches the client (with
+max_retries=0 so summary.py owns the retry policy, not the SDK).
+"""
+from __future__ import annotations
+
+import types
+
+import pytest
+
+from lyra import llm
+
+
+@pytest.fixture
+def fake_openai(monkeypatch):
+    recorded: dict = {}
+
+    class FakeCompletions:
+        def create(self, **kwargs):
+            recorded["create"] = kwargs
+            msg = types.SimpleNamespace(content="ok")
+            return types.SimpleNamespace(choices=[types.SimpleNamespace(message=msg)])
+
+    class FakeClient:
+        def __init__(self, **kwargs):
+            recorded["client"] = kwargs
+            self.chat = types.SimpleNamespace(completions=FakeCompletions())
+
+    monkeypatch.setattr(llm, "OpenAI", FakeClient)
+    monkeypatch.setattr(llm, "load", lambda: types.SimpleNamespace(
+        mi50_base_url="http://mi50/v1", mi50_model="local-gpu",
+        cloud_model="gpt-4o-mini", openai_api_key="sk-test", local_model="l",
+    ))
+    return recorded
+
+
+def test_mi50_threads_max_tokens_and_timeout(fake_openai):
+    out = llm.complete([{"role": "user", "content": "hi"}],
+                       backend="mi50", max_tokens=768, timeout=150)
+
+    assert out == "ok"
+    assert fake_openai["create"]["max_tokens"] == 768
+    assert fake_openai["client"]["timeout"] == 150
+    assert fake_openai["client"]["max_retries"] == 0
+
+
+def test_cloud_threads_max_tokens_and_timeout(fake_openai):
+    llm.complete([{"role": "user", "content": "hi"}],
+                 backend="cloud", max_tokens=768, timeout=150)
+
+    assert fake_openai["create"]["max_tokens"] == 768
+    assert fake_openai["client"]["timeout"] == 150
+    assert fake_openai["client"]["max_retries"] == 0
+
+
+def test_default_bounds_calls_even_without_explicit_timeout(fake_openai):
+    # No cap / timeout passed -> still bounded: 300s default + no SDK retries, so
+    # no call can silently inherit the SDK's 600s x2 (~30 min). No length cap
+    # unless asked, though.
+    llm.complete([{"role": "user", "content": "hi"}], backend="mi50")
+
+    assert "max_tokens" not in fake_openai["create"]
+    assert fake_openai["client"]["timeout"] == 300
+    assert fake_openai["client"]["max_retries"] == 0

tests/test_pattern_recall.py

@@ -0,0 +1,70 @@
+"""Pattern desk: embedded scar recall + strategy gating in the scouting desk."""
+from __future__ import annotations
+
+import hashlib
+import importlib
+
+import numpy as np
+import pytest
+
+
+def _idx(w: str) -> int:
+    # Stable across processes (unlike hash()), so threshold tests aren't flaky.
+    return int.from_bytes(hashlib.md5(w.encode()).digest()[:4], "little") % 256
+
+
+def _fake_embed(texts):
+    out = []
+    for t in texts:
+        v = np.zeros(256, dtype=np.float32)
+        for w in t.lower().split():
+            v[_idx(w)] += 1.0
+        out.append((v if v.any() else np.full(256, 1e-6, dtype=np.float32)).tolist())
+    return out
+
+
+@pytest.fixture
+def mods(tmp_path, monkeypatch):
+    monkeypatch.setenv("LYRA_DB_PATH", str(tmp_path / "test.db"))
+    from lyra import llm
+    monkeypatch.setattr(llm, "embed", _fake_embed)
+    import lyra.memory as memory
+    importlib.reload(memory)
+    import lyra.poker as poker
+    importlib.reload(poker)
+    import lyra.scouting as scouting
+    importlib.reload(scouting)
+    return poker, scouting
+
+
+def test_scar_recall_finds_similar_past_leak(mods):
+    poker, _ = mods
+    old = poker.start_session(stakes="1/3", buy_in=300)
+    poker.log_ritual("scar", "overvalued top pair and stacked off on a wet board",
+                     classification="punt", session_id=old)
+    poker.end_session(200, session_id=old)
+    hits = poker.recall_similar_rituals("stacked off top pair wet board again")
+    assert hits and hits[0]["classification"] == "punt"
+
+
+def test_recall_excludes_current_session(mods):
+    poker, _ = mods
+    sid = poker.start_session(stakes="1/3", buy_in=300)
+    poker.log_ritual("scar", "punted river bluff into the nut flush", session_id=sid)
+    assert poker.recall_similar_rituals("river bluff nut flush punt", exclude_session=sid) == []
+
+
+def test_pattern_pass_only_fires_on_strategic_talk(mods):
+    poker, scouting = mods
+    old = poker.start_session(stakes="1/3", buy_in=300)
+    poker.log_ritual("scar", "punting river bluffs into missed draws again",
+                     classification="punt", session_id=old)
+    poker.end_session(200, session_id=old)
+    poker.start_session(stakes="1/3", buy_in=300, venue="Meadows")
+    # A routine, non-strategic line pays no embed and surfaces nothing.
+    assert scouting.scout("stack is 350 now", venue="Meadows") is None
+    # A real strategy question in the same shape recalls the leak. (The test stub
+    # embeds by shared tokens; real embeddings match on meaning/paraphrase.)
+    note = scouting.scout(
+        "why do i keep punting river bluffs into missed draws", venue="Meadows")
+    assert note and "punting river bluffs" in note

tests/test_poker.py

@@ -18,6 +18,50 @@ def lyra(tmp_path, monkeypatch):
     return poker
 
 
+def test_disown_hand_clears_hero_attribution(lyra):
+    poker = lyra
+    sid = poker.start_session(venue="Meadows", buy_in=300)
+    hid = poker.store_hand_history(
+        {"hero_involved": True, "hero_pos": "MP", "hero_cards": ["As", "3d"],
+         "players": [{"pos": "MP", "cards": ["As", "3d"]}],
+         "result": {"pot": 600, "hero_net": 304}}, session_id=sid, tag="notable")
+    h = poker.disown_hand(hid)
+    assert h["position"] is None and h["hole_cards"] is None and h["result"] is None
+    st = h["structured"]
+    if isinstance(st, str):
+        import json
+        st = json.loads(st)
+    assert st["hero_pos"] is None and not any(pl.get("hero") for pl in st["players"])
+
+
+def test_hud_notes_scoped_to_session_by_tag(lyra):
+    poker = lyra
+    from lyra import memory
+    sid = poker.start_session(venue="Meadows", stakes="1/3", buy_in=300)
+    # A note tagged to THIS session shows on its HUD...
+    memory.add_journal_entry("note", "villain 3 overfolds turn", source=f"poker:{sid}")
+    # ...her autonomous existential journaling (any other source) does NOT, even
+    # though it's written during the exact same window...
+    memory.add_journal_entry("journal", "the quiet dread between conversations", source="dream")
+    # ...nor a note from a *different* poker session.
+    memory.add_journal_entry("note", "some other night", source=f"poker:{sid + 999}")
+    contents = [n["content"] for n in poker.hud(sid)["notes"]]
+    assert contents == ["villain 3 overfolds turn"]
+
+
+def test_note_tool_tags_live_poker_session(lyra):
+    poker = lyra
+    from lyra import tools
+    sid = poker.start_session(stakes="1/3", buy_in=300)
+    tools.dispatch("note", {"content": "whale just sat down seat 4"}, {})
+    assert poker.hud(sid)["notes"][0]["content"] == "whale just sat down seat 4"
+    poker.end_session(cash_out=300, session_id=sid)
+    # With no live session, a note falls back to the general journal (source=chat),
+    # so it does NOT attach to the just-closed session's HUD.
+    tools.dispatch("note", {"content": "random afternoon idea"}, {})
+    assert all(n["content"] != "random afternoon idea" for n in poker.hud(sid)["notes"])
+
+
 def test_session_lifecycle_and_net(lyra):
     poker = lyra
     sid = poker.start_session(venue="Meadows", stakes="1/3", buy_in=400)

tests/test_roster.py

@@ -0,0 +1,101 @@
+"""Live table roster: seat players, attach reads by handle, roster on the HUD."""
+from __future__ import annotations
+
+import importlib
+
+import numpy as np
+import pytest
+
+
+def _fake_embed(texts):
+    out = []
+    for t in texts:
+        v = np.zeros(64, dtype=np.float32)
+        for w in t.lower().split():
+            v[hash(w) % 64] += 1.0
+        out.append((v if v.any() else np.full(64, 1e-6, dtype=np.float32)).tolist())
+    return out
+
+
+@pytest.fixture
+def mods(tmp_path, monkeypatch):
+    monkeypatch.setenv("LYRA_DB_PATH", str(tmp_path / "test.db"))
+    from lyra import llm
+    monkeypatch.setattr(llm, "embed", _fake_embed)
+    import lyra.memory as memory
+    importlib.reload(memory)
+    import lyra.poker as poker
+    importlib.reload(poker)
+    import lyra.tools as tools
+    importlib.reload(tools)
+    return poker, tools
+
+
+def test_seat_players_builds_roster(mods):
+    poker, _ = mods
+    poker.start_session(venue="Meadows", buy_in=300)
+    n = poker.seat_players(["TAG", "Jonathan", {"name": "Wheelz", "seat": "3"}])
+    assert n == 3
+    roster = poker.session_roster()
+    names = {r["name"] for r in roster}
+    assert names == {"TAG", "Jonathan", "Wheelz"}
+    assert next(r for r in roster if r["name"] == "Wheelz")["seat"] == "3"
+
+
+def test_read_attaches_to_seated_player_by_handle(mods):
+    poker, _ = mods
+    poker.start_session(venue="Meadows", buy_in=300)
+    poker.seat_players(["TAG"])
+    poker.add_read(note="limped A4o from the SB, UTG straddle", name="TAG")
+    roster = poker.session_roster()
+    tag = next(r for r in roster if r["name"] == "TAG")
+    assert tag["reads"] == 1 and "A4o" in tag["last_note"]
+    # No duplicate TAG spawned — the read landed on the seated player.
+    assert sum(p["name"] == "TAG" for p in poker.get_villain_file()) == 1
+
+
+def test_seat_players_tool_and_roster_in_hud(mods):
+    poker, tools = mods
+    poker.start_session(venue="Meadows", buy_in=300)
+    out = tools.dispatch("seat_players", {"players": [{"name": "TAG"}, {"name": "JD"}]}, {})
+    assert "TAG" in out and "JD" in out
+    assert len(poker.hud()["roster"]) == 2
+
+
+def test_unseat_player_removes_from_roster_keeps_history(mods):
+    poker, _ = mods
+    poker.start_session(venue="Meadows", buy_in=300)
+    poker.seat_players(["TAG"])
+    poker.add_read(note="showed a bluff", name="TAG")
+    assert poker.unseat_player(name="TAG") is True
+    assert poker.session_roster() == []                 # off the table
+    assert poker.player_profile("TAG")["reads"]         # history intact
+
+
+def test_clear_table_empties_roster_keeps_reads(mods):
+    poker, tools = mods
+    poker.start_session(venue="Meadows", buy_in=300)
+    poker.seat_players(["TAG", "Jonathan"])
+    poker.add_read(note="limped A4o", name="TAG")
+    out = tools.dispatch("clear_table", {}, {})
+    assert "cleared" in out.lower()
+    assert poker.session_roster() == []                     # roster emptied
+    assert poker.player_profile("TAG")["reads"]             # reads kept
+    # A live session is untouched by clearing the table.
+    assert poker.live_session() is not None
+
+
+def test_seat_players_replace_swaps_to_new_table(mods):
+    poker, tools = mods
+    poker.start_session(venue="Meadows", buy_in=300)
+    poker.seat_players(["TAG", "Jonathan"])
+    tools.dispatch("seat_players", {"players": [{"name": "Doyle"}, {"name": "Ivey"}],
+                                    "replace": True}, {})
+    assert {r["name"] for r in poker.session_roster()} == {"Doyle", "Ivey"}
+
+
+def test_seat_players_accepts_plain_name_list_via_tool(mods):
+    poker, tools = mods
+    poker.start_session(venue="Meadows", buy_in=300)
+    tools.dispatch("seat_players", {"players": "TAG, JD, Wheelz"}, {})
+    assert {r["name"] for r in poker.session_roster()} == {"TAG", "JD", "Wheelz"}

tests/test_scouting.py

@@ -0,0 +1,73 @@
+"""The scouting desk: named + descriptor recall, ambiguous→queue, generic→silence."""
+from __future__ import annotations
+
+import importlib
+
+import numpy as np
+import pytest
+
+
+def _fake_embed(texts):
+    out = []
+    for t in texts:
+        v = np.zeros(64, dtype=np.float32)
+        for w in t.lower().split():
+            v[hash(w) % 64] += 1.0
+        out.append((v if v.any() else np.full(64, 1e-6, dtype=np.float32)).tolist())
+    return out
+
+
+@pytest.fixture
+def mods(tmp_path, monkeypatch):
+    monkeypatch.setenv("LYRA_DB_PATH", str(tmp_path / "test.db"))
+    from lyra import llm
+    monkeypatch.setattr(llm, "embed", _fake_embed)
+    import lyra.memory as memory
+    importlib.reload(memory)
+    import lyra.poker as poker
+    importlib.reload(poker)
+    import lyra.scouting as scouting
+    importlib.reload(scouting)
+    return poker, scouting
+
+
+def test_named_player_surfaces_a_brief(mods):
+    poker, scouting = mods
+    sid = poker.start_session(venue="Meadows", stakes="1/3", buy_in=300)
+    pid = poker.upsert_player("Sleepy John", venue="Meadows", category="reg")
+    poker._c().execute(
+        "INSERT INTO player_observations (player_id, session_id, cards, created_at) VALUES (?,?,?,?)",
+        (pid, sid, "As Ks", poker._now()))
+    poker._c().commit()
+    note = scouting.scout("sleepy john just sat down on my left", venue="Meadows")
+    assert note and "Sleepy John" in note and "SCOUTING DESK" in note
+
+
+def test_descriptor_high_match_surfaces_with_confirm(mods):
+    poker, scouting = mods
+    poker.create_descriptor_villain("neck tattoo sleeve arm", venue="Meadows", category="reg")
+    note = scouting.scout("the neck tattoo sleeve guy just 3bet me again", venue="Meadows")
+    assert note and "confirm it's the same guy" in note
+
+
+def test_ambiguous_descriptor_queues_instead_of_interrupting(mods):
+    poker, scouting = mods
+    poker.create_descriptor_villain("neck tattoo sleeve arm", venue="Meadows")
+    note = scouting.scout("the neck tattoo guy raised", venue="Meadows")
+    assert note is None                       # didn't interrupt
+    q = poker.list_identity_queue()
+    assert q and q[0]["kind"] == "needs_clarification"
+
+
+def test_generic_descriptor_stays_silent(mods):
+    poker, scouting = mods
+    poker.create_descriptor_villain("neck tattoo sleeve arm", venue="Meadows")
+    note = scouting.scout("the mid aged white guy with glasses raised", venue="Meadows")
+    assert note is None
+    assert poker.list_identity_queue() == []  # no queue spam for a non-identifier
+
+
+def test_no_player_reference_returns_nothing(mods):
+    poker, scouting = mods
+    poker.upsert_player("Sleepy John", venue="Meadows")
+    assert scouting.scout("i folded pocket kings to a 4bet", venue="Meadows") is None

tests/test_summary_fallback.py

@@ -0,0 +1,142 @@
+"""Summary consolidation: MI50 length cap, fast-fail, and cloud fallback.
+
+Everything is stubbed — no real backend is touched. These drive the behavior of
+`summary._summarize_text`: try the primary backend a bounded number of times with
+a capped generation length, and fall back to cloud if the primary keeps failing.
+"""
+from __future__ import annotations
+
+import types
+
+import pytest
+
+from lyra import summary
+
+
+@pytest.fixture
+def calls(monkeypatch):
+    """Capture every llm.complete call; per-test behavior via `fake.responder`."""
+    recorded: list[dict] = []
+
+    def fake_complete(messages, backend="local", model=None,
+                      max_tokens=None, timeout=None):
+        recorded.append({"backend": backend, "max_tokens": max_tokens, "timeout": timeout})
+        return fake_complete.responder(backend)
+
+    fake_complete.responder = lambda backend: "gist"
+    monkeypatch.setattr(summary.llm, "complete", fake_complete)
+    monkeypatch.setattr(summary.time, "sleep", lambda *_: None)  # instant backoff
+    return types.SimpleNamespace(recorded=recorded, fake=fake_complete)
+
+
+def _set_key(monkeypatch, key="sk-test"):
+    monkeypatch.setattr(summary.config, "load",
+                        lambda: types.SimpleNamespace(openai_api_key=key))
+
+
+def test_falls_back_to_cloud_after_mi50_attempts(calls, monkeypatch):
+    _set_key(monkeypatch)
+
+    def responder(backend):
+        if backend == "mi50":
+            raise RuntimeError("Request timed out.")
+        return "cloud-gist"
+    calls.fake.responder = responder
+
+    out = summary._summarize_text("transcript", "mi50")
+
+    assert out == "cloud-gist"
+    assert [c["backend"] for c in calls.recorded] == ["mi50", "mi50", "cloud"]
+
+
+def test_no_fallback_when_backend_is_cloud(calls, monkeypatch):
+    _set_key(monkeypatch)
+    calls.fake.responder = lambda backend: (_ for _ in ()).throw(RuntimeError("boom"))
+
+    with pytest.raises(RuntimeError):
+        summary._summarize_text("t", "cloud")
+
+    # Cloud is already the primary: retry it, but never a redundant fallback.
+    assert [c["backend"] for c in calls.recorded] == ["cloud", "cloud"]
+
+
+def test_no_fallback_without_openai_key(calls, monkeypatch):
+    _set_key(monkeypatch, key="")
+    calls.fake.responder = lambda backend: (_ for _ in ()).throw(RuntimeError("mi50 down"))
+
+    with pytest.raises(RuntimeError):
+        summary._summarize_text("t", "mi50")
+
+    assert [c["backend"] for c in calls.recorded] == ["mi50", "mi50"]
+
+
+def test_caps_length_and_timeout_on_every_call(calls, monkeypatch):
+    _set_key(monkeypatch)
+
+    def responder(backend):
+        if backend == "mi50":
+            raise RuntimeError("nope")
+        return "cloud-gist"
+    calls.fake.responder = responder
+
+    summary._summarize_text("t", "mi50")
+
+    assert calls.recorded
+    for c in calls.recorded:
+        assert c["max_tokens"] == summary.SUMMARY_MAX_TOKENS
+        assert c["timeout"] == summary.SUMMARY_TIMEOUT
+
+
+def test_happy_path_uses_primary_only(calls, monkeypatch):
+    _set_key(monkeypatch)
+    calls.fake.responder = lambda backend: "mi50-gist"
+
+    out = summary._summarize_text("t", "mi50")
+
+    assert out == "mi50-gist"
+    assert [c["backend"] for c in calls.recorded] == ["mi50"]  # no retries, no fallback
+
+
+# --- degenerate ("?" garbage) output guard: a wedged local model returns junk as
+# a successful 200, so treat it as a failure and fall back to cloud. ---
+
+def test_looks_degenerate_flags_repeated_char():
+    assert summary._looks_degenerate("?" * 60) is True
+    assert summary._looks_degenerate("!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!") is True
+
+
+def test_looks_degenerate_passes_real_prose():
+    gist = ("Brian sat down at the Meadows 1/3 in seat 6 with two straddles active; "
+            "he tagged a seat-3 calling station and finished the session up 240.")
+    assert summary._looks_degenerate(gist) is False
+
+
+def test_looks_degenerate_ignores_short_output():
+    # Too short to judge — don't false-positive a terse-but-valid reply.
+    assert summary._looks_degenerate("ok") is False
+
+
+def test_degenerate_mi50_output_falls_back_to_cloud(calls, monkeypatch):
+    _set_key(monkeypatch)
+
+    def responder(backend):
+        if backend == "mi50":
+            return "?" * 200          # garbage-as-200, not an exception
+        return "a real cloud gist of the session, diverse and coherent."
+    calls.fake.responder = responder
+
+    out = summary._summarize_text("transcript", "mi50")
+
+    assert "cloud gist" in out
+    assert [c["backend"] for c in calls.recorded] == ["mi50", "mi50", "cloud"]
+
+
+def test_degenerate_cloud_output_raises_no_infinite_loop(calls, monkeypatch):
+    _set_key(monkeypatch)
+    calls.fake.responder = lambda backend: "?" * 200  # every backend returns garbage
+
+    with pytest.raises(Exception):
+        summary._summarize_text("t", "mi50")
+
+    # mi50 x2, then one cloud fallback that's also garbage -> give up, no loop.
+    assert [c["backend"] for c in calls.recorded] == ["mi50", "mi50", "cloud"]

tests/test_transcript.py

@@ -0,0 +1,82 @@
+"""Conversation export: speech (exchanges) + actions (tool_events) merged in order."""
+from __future__ import annotations
+
+import importlib
+import json
+
+import pytest
+
+
+def _const_embed(texts):
+    return [[1e-6] * 8 for _ in texts]
+
+
+@pytest.fixture
+def mods(tmp_path, monkeypatch):
+    monkeypatch.setenv("LYRA_DB_PATH", str(tmp_path / "test.db"))
+    from lyra import llm
+    monkeypatch.setattr(llm, "embed", _const_embed)
+    import lyra.memory as memory
+    importlib.reload(memory)
+    import lyra.transcript as transcript
+    importlib.reload(transcript)
+    return memory, transcript
+
+
+def _seed(memory):
+    """A turn where Brian narrates a hand and Lyra logs it, then replies."""
+    memory.ensure_session("s1", name="Meadows 1/3")
+    memory.remember("s1", "user", "got it in with a set, he had the flush draw and bricked")
+    memory.add_tool_event("s1", "record_hand", {"result": "won", "pot": 750}, "hand #42 logged")
+    memory.add_tool_event("s1", "log_stack", {"amount": 750, "note": "doubled up"}, "ok")
+    memory.remember("s1", "assistant", "Clean stack-off — logged it to your timeline.")
+
+
+def test_tool_events_roundtrip_parses_args(mods):
+    memory, _ = mods
+    _seed(memory)
+    events = memory.tool_events("s1")
+    assert [e["tool"] for e in events] == ["record_hand", "log_stack"]
+    assert events[0]["args"] == {"result": "won", "pot": 750}   # parsed back to a dict
+    assert events[1]["result"] == "ok"
+
+
+def test_markdown_interleaves_speech_and_actions_in_order(mods):
+    memory, transcript = mods
+    _seed(memory)
+    md = transcript.as_markdown("s1", name="Meadows 1/3")
+    # user message, then both tool calls, then assistant reply — in that order
+    i_user = md.index("got it in with a set")
+    i_hand = md.index("record_hand")
+    i_stack = md.index("log_stack")
+    i_reply = md.index("Clean stack-off")
+    assert i_user < i_hand < i_stack < i_reply
+    assert "**Brian**" in md and "**Lyra**" in md
+    assert "⚙" in md
+
+
+def test_json_export_is_machine_readable(mods):
+    memory, transcript = mods
+    _seed(memory)
+    payload = transcript.as_json("s1", name="Meadows 1/3")
+    assert payload["session_id"] == "s1"
+    types = [e["type"] for e in payload["events"]]
+    assert types == ["message", "tool", "tool", "message"]
+    json.dumps(payload)  # must be serializable
+
+
+def test_build_returns_filename_and_media_type(mods):
+    memory, transcript = mods
+    _seed(memory)
+    body_md, mt_md, fn_md = transcript.build("s1", "md", "Meadows 1/3")
+    body_js, mt_js, fn_js = transcript.build("s1", "json", "Meadows 1/3")
+    assert fn_md.endswith(".md") and "markdown" in mt_md
+    assert fn_js.endswith(".json") and mt_js == "application/json"
+    assert body_md and body_js
+
+
+def test_delete_session_clears_tool_events(mods):
+    memory, _ = mods
+    _seed(memory)
+    memory.delete_session("s1")
+    assert memory.tool_events("s1") == []

tests/test_villain_tools.py

@@ -0,0 +1,103 @@
+"""Confirm-loop tools: descriptor reads, name attach, merge/mark-distinct."""
+from __future__ import annotations
+
+import importlib
+
+import numpy as np
+import pytest
+
+
+def _fake_embed(texts):
+    out = []
+    for t in texts:
+        v = np.zeros(64, dtype=np.float32)
+        for w in t.lower().split():
+            v[hash(w) % 64] += 1.0
+        out.append((v if v.any() else np.full(64, 1e-6, dtype=np.float32)).tolist())
+    return out
+
+
+@pytest.fixture
+def mods(tmp_path, monkeypatch):
+    monkeypatch.setenv("LYRA_DB_PATH", str(tmp_path / "test.db"))
+    from lyra import llm
+    monkeypatch.setattr(llm, "embed", _fake_embed)
+    import lyra.memory as memory
+    importlib.reload(memory)
+    import lyra.poker as poker
+    importlib.reload(poker)
+    import lyra.tools as tools
+    importlib.reload(tools)
+    return poker, tools
+
+
+def test_descriptor_read_creates_then_reuses_nameless_villain(mods):
+    poker, tools = mods
+    poker.start_session(venue="Meadows", stakes="1/3", buy_in=300)
+    tools.dispatch("add_read", {"note": "opened UTG light",
+                                "descriptor": "neck tattoo sleeve arm"}, {})
+    tools.dispatch("add_read", {"note": "showed a bluff",
+                                "descriptor": "neck tattoo sleeve"}, {})  # rephrase → same guy
+    players = [p for p in poker.get_villain_file() if not p["named"]]
+    assert len(players) == 1                     # one nameless villain, not two
+    reads = poker._c().execute(
+        "SELECT COUNT(*) n FROM player_reads WHERE player_id = ?", (players[0]["id"],)
+    ).fetchone()["n"]
+    assert reads == 2
+
+
+def test_description_as_name_routes_to_descriptor_and_dedupes(mods):
+    poker, tools = mods
+    poker.start_session(venue="Meadows", buy_in=300)
+    # She (wrongly) puts a physical description in the name field, twice, worded
+    # slightly differently — must resolve to ONE nameless villain, not two named.
+    tools.dispatch("add_read", {"note": "limp 3bet A3o",
+                                "name": "Filipino, Fox Racing hat, DKNY shirt, two bracelets"}, {})
+    tools.dispatch("add_read", {"note": "called a 4bet light",
+                                "name": "Filipino, Fox Racing hat, DKNY shirt, watch on left"}, {})
+    named = [p for p in poker.get_villain_file() if p["named"]]
+    assert named == []                    # no sentence-named players spawned (the bug)
+    # Either they merged, or the near-dup is surfaced for a one-click merge — never
+    # a silent duplicate the way sentence-names were.
+    q = poker.list_identity_queue()
+    nameless = [p for p in poker.get_villain_file() if not p["named"]]
+    assert len(nameless) == 1 or any(t["kind"] == "merge_candidate" for t in q)
+
+
+def test_name_villain_tool_attaches_name(mods):
+    poker, tools = mods
+    poker.start_session(venue="Meadows", buy_in=300)
+    poker.create_descriptor_villain("neck tattoo sleeve arm", venue="Meadows")
+    out = tools.dispatch("name_villain", {"descriptor": "neck tattoo sleeve arm",
+                                          "name": "Danny"}, {})
+    assert "Danny" in out
+    assert poker.resolve_villain("Danny")["band"] == "name"
+
+
+def test_link_villains_merge_and_distinct(mods):
+    poker, tools = mods
+    poker.start_session(venue="Meadows", buy_in=300)
+    poker.upsert_player("Danny", venue="Meadows")
+    poker.upsert_player("Donny", venue="Meadows")
+    # same=false → recorded distinct
+    tools.dispatch("link_villains", {"player_a": "Danny", "player_b": "Donny",
+                                     "same": False, "note": "different builds"}, {})
+    a = poker.resolve_villain("Danny")["match_id"]
+    b = poker.resolve_villain("Donny")["match_id"]
+    assert poker.are_distinct(a, b)
+    # same=true on a fresh pair → merged
+    poker.upsert_player("Mike", venue="Meadows")
+    poker.upsert_player("Michael", venue="Meadows")
+    tools.dispatch("link_villains", {"player_a": "Mike", "player_b": "Michael",
+                                     "same": True}, {})
+    names = [p["name"] for p in poker.get_villain_file()]
+    assert ("Mike" in names) ^ ("Michael" in names)  # one absorbed the other
+
+
+def test_link_villains_refuses_when_reference_is_vague(mods):
+    poker, tools = mods
+    poker.start_session(venue="Meadows", buy_in=300)
+    poker.upsert_player("Danny", venue="Meadows")
+    out = tools.dispatch("link_villains", {"player_a": "Danny",
+                                           "player_b": "some guy", "same": True}, {})
+    assert "didn't merge" in out.lower() or "couldn't" in out.lower()

tests/test_villains.py

@@ -0,0 +1,116 @@
+"""Nameless-villain identity resolution: descriptor matching, merge, distinct, queue."""
+from __future__ import annotations
+
+import importlib
+
+import numpy as np
+import pytest
+
+
+def _fake_embed(texts):
+    """Overlap-sensitive bag-of-words vectors so cosine reflects shared tokens."""
+    out = []
+    for t in texts:
+        v = np.zeros(64, dtype=np.float32)
+        for w in t.lower().split():
+            v[hash(w) % 64] += 1.0
+        out.append((v if v.any() else np.full(64, 1e-6, dtype=np.float32)).tolist())
+    return out
+
+
+@pytest.fixture
+def poker(tmp_path, monkeypatch):
+    monkeypatch.setenv("LYRA_DB_PATH", str(tmp_path / "test.db"))
+    from lyra import llm
+    monkeypatch.setattr(llm, "embed", _fake_embed)
+    import lyra.memory as memory
+    importlib.reload(memory)
+    import lyra.poker as poker
+    importlib.reload(poker)
+    return poker
+
+
+def test_distinctiveness_distinctive_vs_generic(poker):
+    assert poker.distinctiveness("guy with a neck tattoo") > 0.6
+    assert poker.distinctiveness("mid-aged white dude with glasses") < 0.3
+
+
+def test_generic_descriptor_never_resolves_to_a_guess(poker):
+    poker.create_descriptor_villain("neck tattoo sleeve arm", venue="Meadows")
+    r = poker.resolve_villain("mid aged white guy with glasses", venue="Meadows")
+    assert r["band"] == "generic"
+    assert r["match_id"] is None
+
+
+def test_exact_name_match_is_deterministic(poker):
+    pid = poker.upsert_player("Sleepy John", venue="Meadows")
+    r = poker.resolve_villain("sleepy john")
+    assert r["band"] == "name" and r["match_id"] == pid
+
+
+def test_rephrased_descriptor_resolves_high(poker):
+    pid = poker.create_descriptor_villain("neck tattoo sleeve arm", venue="Meadows")
+    r = poker.resolve_villain("neck tattoo sleeve", venue="Meadows")
+    assert r["band"] == "high" and r["match_id"] == pid
+    assert r["confidence"] >= 0.80
+
+
+def test_partial_descriptor_is_ambiguous_not_high(poker):
+    poker.create_descriptor_villain("neck tattoo sleeve arm", venue="Meadows")
+    r = poker.resolve_villain("neck tattoo", venue="Meadows")
+    assert r["band"] == "ambiguous"  # plausible, but don't guess live
+
+
+def test_merge_repoints_observations_and_deletes_dup(poker):
+    keep = poker.create_descriptor_villain("neck tattoo", venue="Meadows")
+    dup = poker.create_descriptor_villain("neck ink tatted", venue="Meadows")
+    poker._c().execute(
+        "INSERT INTO player_observations (player_id, session_id, created_at) VALUES (?,1,?)",
+        (dup, poker._now()))
+    poker._c().commit()
+    assert poker.merge_players(keep, dup) is True
+    assert poker.get_villain_file() and all(p["id"] != dup for p in poker.get_villain_file())
+    obs = poker._c().execute(
+        "SELECT COUNT(*) n FROM player_observations WHERE player_id = ?", (keep,)).fetchone()["n"]
+    assert obs == 1
+
+
+def test_merge_prefers_a_real_name(poker):
+    named = poker.upsert_player("Danny", venue="Meadows")
+    desc = poker.create_descriptor_villain("neck tattoo", venue="Meadows")
+    poker.merge_players(desc, named)  # keep the descriptor id, but name should win
+    row = dict(poker._c().execute("SELECT name, named FROM poker_players WHERE id = ?", (desc,)).fetchone())
+    assert row["name"] == "Danny" and row["named"] == 1
+
+
+def test_mark_distinct_blocks_merge_scan(poker):
+    a = poker.create_descriptor_villain("neck tattoo sleeve", venue="Meadows")
+    b = poker.create_descriptor_villain("neck tattoo sleeve", venue="Meadows")
+    poker.mark_distinct(a, b, note="one's taller")
+    assert poker.are_distinct(a, b)
+    assert poker.scan_merge_candidates() == 0  # confirmed-distinct pair is skipped
+
+
+def test_scan_files_merge_candidate_for_near_duplicates(poker):
+    poker.create_descriptor_villain("neck tattoo sleeve", venue="Meadows")
+    poker.create_descriptor_villain("neck tattoo sleeve", venue="Meadows")
+    filed = poker.scan_merge_candidates()
+    assert filed == 1
+    q = poker.list_identity_queue()
+    assert q and q[0]["kind"] == "merge_candidate" and len(q[0]["players"]) == 2
+
+
+def test_queue_dedupes_identical_pending_task(poker):
+    a = poker.create_descriptor_villain("neck tattoo", venue="Meadows")
+    b = poker.create_descriptor_villain("neck ink", venue="Meadows")
+    t1 = poker.queue_identity_task("merge_candidate", [a, b])
+    t2 = poker.queue_identity_task("merge_candidate", [b, a])  # same pair, reversed
+    assert t1 == t2
+    assert len(poker.list_identity_queue()) == 1
+
+
+def test_name_villain_flips_named_flag(poker):
+    pid = poker.create_descriptor_villain("neck tattoo", venue="Meadows")
+    poker.name_villain(pid, "Danny")
+    r = poker.resolve_villain("Danny")
+    assert r["band"] == "name" and r["match_id"] == pid