chat/CLAUDE.md

Roleplay Engine

Local-first roleplay chat app that treats fiction as a simulation, not a chat log. The LLM is a renderer for structured world state — it does not hold state.

See rp-engine-design.md for the architectural design and docs/plans/2026-04-26-v1-requirements-design.md for the v1 product requirements & behavioral spec. This file is the working summary.

Why this exists

Fixes three failure modes of conventional RP chatbots:

1. Memory loss — old context drops as history grows
2. Quality decay — bots get terse and generic over long conversations
3. Stale state pollution — bots fixate on past props (the "picnic basket" problem)

Hard scope constraints

• Single user, single machine (the user's Mac)
• Max 3 entities per scene: you + up to 2 bots (botA, botB)
• Chat-only — no voice, no real-time

The 3-entity cap is load-bearing: it makes the relationship graph fully enumerable (6 directed edges + 1 group node). Don't design for N entities.

Architecture

• Mac (always-on): web UI, orchestrator, persistence, event queue, retrieval, prompt construction, all state.
• Inference endpoint: stateless generate(prompt, params) -> text. Swap implementations behind one interface. The orchestrator never knows which.
• Streaming required for UX.

Runtime stack (locked for v1)

• Backend: Python 3.11+ with FastAPI.
• Frontend: server-rendered HTML + HTMX + minimal vanilla JS/CSS. No JS build chain.
• Live updates: SSE per chat. Per-chat asyncio.Queue pub/sub. Multi-tab sync is a Phase 1 requirement — two browser tabs on the same chat must mirror each other live (streamed tokens, drawer state, edge updates).
• Inference backend: Featherless (OpenAI-compatible API).
  • narrative_model = dphn/Dolphin-Mistral-24B-Venice-Edition (32K ctx, uncensored).
  • classifier_model = NousResearch/Hermes-3-Llama-3.1-8B (128K ctx, uncensored, structured-output reliable). Fallbacks: cognitivecomputations/dolphin-2.9.4-llama3-8b → mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated.
• Token budgets: narrative 8K hard / 6K soft; classifier 4K hard. Trim tiers must / should / nice — never trim must-include.
• OOC marker: ((double parens)) (configurable).
• Data layout: everything under <repo>/data/ — chat.db, backups/, snapshots/, exports/, config.toml. The whole tree is .gitignored. CHAT_DB_PATH env var honored as override.
• Auth: bind to 127.0.0.1 only in v1. No auth.

Behavioral defaults (locked in v1 brainstorm round 2)

• Significance scale: 0=Routine, 1=Notable, 2=Significant, 3=Pivotal. Score-3 turns auto-pin per witness. Drives retrieval ranking, compression, JSON exports.
• Edge updates: per-turn deltas (affinity_delta, trust_delta, knowledge_facts, last_interaction); per-scene-close summary rewrite. Every mutation goes through the event log as edge_update.
• Classifier failure handling: Pydantic-constrained → 1 retry with stricter reminder → schema-default fallback. 10s timeout. Never block the play loop. Refusals trigger fallback-model swap for that one call. Failures logged to classifier_failures table.
• Activity verbs: open string + classifier-extracted interruptible, required_attention, expected_duration. Attention is optional free-form; omit from prompt when empty.
• Containers: parse-and-extend. Per-chat scoped. Kickoff parse seeds initial; transitions create new.
• Pinning: soft cap 8 / bot. Pivotal (score 3) = auto-pin. Manual pins never auto-evicted.
• Snapshots: periodic every 100 events / 30 min; pre-rewind always. 5 periodic retained; pre-rewind retained 14 days.
• Streaming: Stop button on streaming row; mid-stream disconnect commits partial with truncated: true; Send disabled mid-stream; multi-tab streaming via per-chat SSE channel.
• Display: lightweight markdown; *action* italic; OOC ((parens)) shown dimmed/italic, never sent to bot.
• Multi-entity defaults (Phase 2): when chat.guest_bot_id is None, behavior matches Phase 1 single-bot 1:1. With a guest, all 3 entities are present in the prompt, witness writes, and state-update fan-out (6 directed pairs).
• Addressee detection: simple substring match (whole-word, case-insensitive) over the user turn's body. If both bot names match or neither does, the host gets the floor.
• Interjection: classifier-driven, conservative bias (default false on classifier failure / refusal / parse error). When the classifier returns true, the addressee speaks first, then the non-addressee may interject in a follow-up turn.
• Per-POV summaries (multi-entity): each present witness with a memory store gets their own per-POV summary on scene close. The summary differs per bot based on persona + their edge to "you". The group node summary is updated alongside.

Core concepts (vocabulary)

• Entity: you | botA | botB. Has identity (immutable), state (mood/goals/status), activity, per-POV memory.
• Container: anything with slots that holds entities (car, booth, room). Has properties (moving, public, audible range). Spatial grounding lives here, separate from the relationship graph.
• Activity record: per-entity live struct — position (container+slot), posture, current action (verb, duration, interruptible, required_attention), holding, attention, status. Always in the prompt as a small structured block.
• Relationship graph: 6 directed edges (asymmetric feelings matter — never collapse to a single shared field) + 1 group node. Edges hold affinity, trust, summary, knowledge-known-about-target, private moments, last-interaction.
• Scene configurations: exactly 4 — solo with botA, solo with botB, all three present, botA+botB without you ("meanwhile…"). Each has a fixed prompt-loading rule.
• Witnessed-by flag: every memory has a 3-bit [you, botA, botB] mask. A speaker only sees memories where their bit is set. This is the mechanism that prevents bots referencing things they can't know.
• Event: scoped lifecycle (planned | active | completed | cancelled | expired) with its own props, preconditions, on_start/on_complete hooks, significance. Solves the picnic-basket problem — props live and die with the event, only narrative gist promotes to memory.
• Active threads: unresolved plot tensions. Sticky in context until resolved/dropped. Cheap, anchor continuity across compressed scenes.
• Scene: closes when container changes meaningfully or significant time passes. Compression boundary.
• Per-POV summary: every witness gets their own record of a closed scene, written from their POV. Different details, different interpretations. This is what gives bots inner lives — never write omniscient narration into per-POV stores.
• Time skip: elision (skip the boring middle of an in-progress activity) vs jump (next morning, a week later). Skips run intervening events forward, compress, reset landing activity.

What promotes out of an event (and what doesn't)

• Object acquired → inventory
• Knowledge gained → edge knowledge field
• Relationship change → edge summary
• Everything else stays in the closed event record. The blanket, the basket, the specific sandwich do not become memories. This rule is the whole point — don't bypass it.

Persistence

• SQLite (single file) for everything structured. WAL mode, foreign keys on, each turn in a transaction.
• sqlite-vss or sqlite-vec for embeddings (same DB file). Decide at Phase 4.
• JSON for snapshots, character templates, scene exports.
• No Postgres, Redis, Pinecone, Docker. Single-user; don't over-engineer.

Schema is event-sourced. See design doc § "Persistence Layer" for the full sketch.

Event sourcing — non-negotiable

State is a projection of an append-only event log. State is never mutated directly — append an event, the projector applies it.

Event kinds: user_turn, assistant_turn, time_skip, event_triggered, edge_update, scene_transition, entity_state_change, activity_change.

This buys: free rewind, trivial replay-debugging, schema migrations against the same log, branching ("what if BotA had said yes").

Determinism on replay: LLM calls are nondeterministic. Store the outcome in the event payload — on replay, use the stored outcome. Never re-call the LLM during replay.

Snapshots every N events / M minutes so we don't replay everything on load. Log is source of truth.

Prompt construction

A speaker's prompt is assembled from their edges and their witnessed memories — never the global state. BotA and BotB are effectively two separate agents who happen to share a scene.

Order (for speaker BotA, with you and BotB present):

1. BotA identity + current state
2. BotA → You edge
3. BotA → BotB edge
4. Group node (only if all three present)
5. World state (time, weather, location)
6. Active scene description
7. Activity snapshot for all present entities
8. Active threads
9. Recent dialogue window
10. Retrieved memories (top-K, witness-filtered, BotA-owned)
11. Currently active events + their props

After every utterance, run a state-update pass on every present entity, not just the speaker. Silent witnesses still update edges.

Memory retrieval

• Always-loaded: pinned, current scene, active threads, recent N scenes (no retrieval).
• Retrieved: top-K vector search over the speaker's memory store, filtered by witness flag, with recency + significance boosts.
• Keep K small. Bloated retrieval poisons the prompt.
• Phase 1: SQLite FTS5 is enough. Vector search comes at Phase 4.

Implementation phases

1. Core loop: schema, entities + edges, single container, event log + projector, single-bot conversation, one LLM backend, streaming UI, manual rollback.
2. Multi-entity: second bot, group node, scene configs, witness filtering, per-POV memories, activity/containers, scene transitions with compression.
3. Events & skips: event queue with triggers, time skips, active threads, significance classifier.
4. Polish: vector retrieval, branching, surgical delete + regenerate, snapshots, backups, impact-preview UI for rewinds.

Don't jump phases. Phase 1 must work end-to-end before Phase 2 lands.

Conventions for working in this repo

• Don't bypass the event log. Any state change goes through an event. If you're tempted to UPDATE a row directly, you're doing it wrong.
• Don't collapse directed edges. botA → botB and botB → botA are independent. Asymmetry is the point.
• Don't promote event props to memory. Only the four promotion categories above survive an event closing.
• Per-POV, not omniscient. When writing scene summaries, write one per witness, from their angle.
• Witness filter every memory read. A bot must never see a memory their bit isn't set on.
• Activity block is always in the prompt. It's the spatial anchor that prevents "leaning on the kitchen counter while in a car" failures.
• Streaming on the inference path; non-blocking bookkeeping (significance classification, embeddings, snapshots) runs while the LLM streams.
• No Docker, no extra services. SQLite + a process. Push back on suggestions to add infrastructure.

Open decisions (deferred — don't pre-decide)

• Token budget strategy (during Phase 1, with real prompts)
• Embedding model (Phase 4)
• sqlite-vss vs sqlite-vec (Phase 4)
• UI framework (local web app / Tauri / Electron / native — TBD)
• Inference hosting (start with a cloud API, re-evaluate later)
• Character template format (during Phase 1)
• Multi-session / multi-character casts: out of scope for v1. Leave cheap schema hooks only.

Phase 1 status

Phase 1 shipped end-to-end across 35 tasks (T0–T35). The single-bot core loop is functional: event log + projector, schema + migrations, settings/bot authoring, kickoff confirm, streaming turns, drawer rendering, regenerate/rewind, scene close + per-POV summaries, significance classifier, snapshots/backups, first-run navigation, and friendly 404/500 pages. 168 tests passing.

Deferred to Phase 2: second bot, group node, scene configurations, witness filtering across multi-entity scenes, activity/containers, scene-transition compression. Phase 3: event queue + triggers, time skips, active threads. Phase 4: vector retrieval, branching, surgical delete + regenerate, impact-preview UI.

Known v1 limitations (read before extending)

• Drawer edits scope: only affinity, significance, and pin can be hand-edited from the drawer. Other v1 fields (knowledge, summary text, traits) are deferred to Phase 1.5.
• Cold-load snapshot path is wired and unit-tested but rarely exercised in dev — long-running sessions are the only realistic trigger.
• WAL sidecar files (-wal, -shm) are not captured in nightly backups; the nightly snapshot is a fresh .backup() so this is fine for restore but worth knowing if you copy the db file by hand.
• HTMX SSE event names may need a version check if you bump the htmx CDN URL in base.html — the swap targets are name-coupled.
• "You" activity rows can linger after bot_reset (the reset purges the bot's chats and the bot's own activity row but not the "you" row that was associated with those chats). Cosmetic, fixed in Phase 1.5.
• Projector replay is non-idempotent for plain INSERT events. After appending, call apply_event(conn, event) for the new row only — calling project(conn) re-runs every handler from scratch and will trip uniqueness or duplicate inserts.
• 8-pin auto-cap eviction is FIFO over the auto-pinned set only. Manual pins survive the eviction; this is by design (manual intent > auto-pin signal).
• Regenerate (T29) does not broadcast turn_html over SSE — the page must refresh to show the regenerated turn. Acceptable for v1 single-tab usage; Phase 1.5 should wire the SSE event.
• First-run middleware fires only on bare / and /chats. Sub-paths like /chats/<id> and /chats/<id>/drawer pass through (correct: HTMX partials should not page-redirect, and a deep-link to a missing chat should 404, not redirect mid-setup).

Phase 1.5 cleanup backlog

All items shipped — see Phase 2.5 status below.

Phase 2 status

Phase 2 shipped end-to-end across 13 tasks (T36–T48 wave). The multi-entity surface is functional: chats can host a guest bot, the prompt assembly is guest-aware, post-turn fans out across all directed pairs, and scene close writes a per-POV summary per present witness plus a group_node summary.

• Multi-entity scene support: chats can now have a guest bot (you + host + guest). The 3-entity cap holds. New event kinds: guest_added, guest_removed, group_node_initialized, group_node_updated. New table: group_node (members, summary, dynamic, threads).
• Drawer guest UX: add/remove guest from the drawer side panel. The "have they met?" prose seed is parsed by the relationship_seed classifier into inter-bot directed edges (host↔guest).
• Multi-entity turn flow: post_turn assembles narrative with the guest-aware prompt; writes memories for all present bot witnesses; runs state updates for all directed pairs (6 with 3 entities); detects interjections via classifier (default false; the addressee gets the floor first).
• Per-POV scene close summaries: each present witness with a memory store gets their own per-POV summary on close; group_node summary updated alongside.
• Bot reset cascade: resetting a bot now also clears chats.guest_bot_id references in other chats (root-cause fix for stale-guest references after T47).

Phase 2.5 / 3 backlog

All items shipped — see Phase 2.5 status below.

Phase 2.5 status

Phase 2.5 cleanup shipped end-to-end across 8 tasks (T68–T75). Two CLAUDE.md backlogs (Phase 1.5 cleanup, Phase 2.5/3) are now empty; deferred follow-ups discovered during execution are tracked in a new "Phase 2.6 / 3 backlog" section below.

• open_db with check_same_thread parameter (T68): refactored chat/db/connection.py so chat/web/bots.py:get_conn no longer duplicates the PRAGMA setup. Default behavior preserved.
• bot_reset cross-chat cleanup (T69): now purges orphaned "you" activity rows. Note: this also fixed a latent FK constraint crash that was lurking in the projector — activity.container_id is FK-referenced and the prior code would have crashed on any reset of a bot whose chat had a non-NULL container_id "you" activity row. The bug was masked because no prior test seeded such a row.
• LLM-merged group meta-summary (T70): replaces Phase 2 T45's naive concat with a classifier merge call. Falls back to the naive concat on classifier failure.
• prompt.py polish (T71): witness role parametric (host vs guest derived from chat membership); single ACTIVITIES: block with bullet-level trim; NICE trim order kept with documented rationale (greedy cheapest-impact-first beats spec-listing order in practice).
• Drawer polish (T72): deferred v1 edits (edge_trust slider, edge_summary textarea, memory pov_summary textarea, knowledge_facts add/remove) + first-meeting gate (Add-guest form disables prose textarea when host→guest edge already exists; "re-seed anyway" toggle re-enables) + witness flag inline-edit (per-memory checkboxes for [you, host, guest] flags). Two new manual_edit projector branches: edge_knowledge_fact and memory_witness.
• Regenerate polish (T73): regenerate now broadcasts turn_html_replace over SSE (NEW event distinct from turn_html to avoid breaking the existing append-semantic consumer); regenerate covers interjection turns (re-detects + re-streams or supersedes); defensive stale-guest degrade removed.
• Turn-flow polish + addressee service (T74): classifier-based addressee detection (substring helper kept as no-guest fast path); SignificanceJob enqueued for interjection memories; scene-close-on-cancel pinned with comment + regression test (close detection is genuinely user-prose-only); defensive stale-guest degrade removed.

Phase 2.6 / 3 backlog

All items shipped — see Phase 3.5 status below.

Phase 3 status

Phase 3 shipped end-to-end across 19 tasks (T49–T67). Events with full lifecycle, time skips, active threads, significance refinements, and meanwhile scenes are functional. Schema baseline is now version 11 (migrations 0009 events, 0010 threads, 0011 meanwhile_scenes). Test count grew from ~247 (Phase 2) to ~315 (+68 new tests across the wave).

• Wave 1 — schema + lifecycle handlers (parallel):
  • T49 events table + lifecycle handlers (event_planned, event_started, event_completed, event_cancelled, event_expired).
  • T50 time_skip event handlers (elision and jump variants).
  • T51 threads table + handlers (thread_opened, thread_updated, thread_closed).
• Wave 2 — detection / narration services (parallel):
  • T52 event-lifecycle detection service (planned→active→completed transitions inferred from narration).
  • T53 skip narration service (elision + jump prose).
  • T54 synthesized-memories service for jump skips (LLM-summarized intervening time).
  • T55 thread-detection service (open/update/close inferred from recent dialogue).
• Wave 3 — promotion + ranking (parallel):
  • T56 event-completion promotion service (objects → inventory, knowledge → edge knowledge, relationship deltas → edge summary; everything else stays in the closed event).
  • T57 significance-aware retrieval ranking — SQL-side SIGNIFICANCE_RANK_BIAS plus the existing Python composite re-rank.
  • T58 scene compression keeps key quotes when significance ≥ 2; thread emission piggybacks on scene close.
• Wave 4 — drawer UX (single):
  • T59 drawer additions: events panel, threads panel, skip controls.
• Wave 5a — prompt + turn flow integration (parallel):
  • T60 prompt assembly includes active events + open threads in the speaker's prompt.
  • T61 turn flow invokes event-detection + completion promotion alongside existing post-turn fan-out.
• Wave 5b — natural-language skip surface (single):
  • T62 classifier-driven skip command at the user-input layer; shared skip controllers extracted into chat/web/skip.py.
• Wave 6a — meanwhile schema (single):
  • T63 meanwhile-scene schema + state (scene config 4: host+guest, no "you").
• Wave 6b — meanwhile turn flow (parallel):
  • T64 meanwhile turn flow (host+guest, no "you" in the prompt or witness writes).
  • T65 meanwhile summary digest surfaces to the next "you"-present scene.
• Wave 7 — integration + docs (parallel):
  • T66 cross-feature integration tests covering events × skips × threads × meanwhile.
  • T67 documentation (this section).

Phase 3.5 / 4 backlog

All items shipped — see Phase 3.5 status below.

Phase 3.5 status

Phase 3.5 cleanup shipped end-to-end across 12 tasks (T76–T87). Two CLAUDE.md backlogs (Phase 2.6/3, Phase 3.5/4) are now empty; deferred follow-ups discovered during execution are tracked in a new "Phase 3.6 / 4 backlog" section below. Test count grew from 315 (Phase 3) to 343 (+28 new tests).

• Wave 1 — trivial polish (parallel):
  • T76 narrate_skip timeout_s plumbed through to client.generate.
  • T77 AddresseeDecision.confidence typed as Literal["high","medium","low"].
  • T78 search_memories docstring notes SQL-side significance bias (SIGNIFICANCE_RANK_BIAS).
  • T79 _witness_role_for defensive host_bot_id is None handling (returns "host" for Phase-1 chats).
• Wave 2 — scene_summarize polish (single):
  • T80 five T58 follow-ups: re-close suffix bloat guard, transcript scoping by scene, swallowed-exception logging in detect_threads, chat-clock closed_at, and three new tests covering T58 gaps (200-char truncation, thread_updated/thread_closed candidate paths, try/except fallback).
• Wave 3 — typed exception (single):
  • T81 ChatNotFoundError replaces string-prefix sniff in skip routes; mapped to 404 (vs 400 for other ValueError cases).
• Wave 4 — turn-flow wiring (single):
  • T82 consume_pending_meanwhile_digests wired into post_turn (closes T66 gap; meanwhile digests no longer pile up); natural-language skip dispatch now runs scene close detection first.
• Wave 5 — regenerate polish (single):
  • T83 five sub-fixes — cancel/stop hook (regenerate registers stream task in _in_flight_tasks); DRY extraction of read_recent_dialogue and gather_prior_edges into chat/services/turn_common.py; chat-scoped sibling-assistant-turn lookup; lifecycle-rollback warning log on regenerate; ordering-symmetry comment between post_turn and regenerate event-detection paths.
• Wave 6 — final polish (parallel):
  • T84 unified record_turn_memory API with you_present kwarg; record_meanwhile_memory becomes a thin wrapper.
  • T85 JSON-build audit (no findings) + meanwhile cancel route-level test.
  • T86 frontend turn_html_replace SSE handler + turn_id stamping on rendered HTML so the in-place swap actually works.

Phase 3.6 / 4 backlog

New follow-ups discovered during Phase 3.5 reviews and execution. None are blocking; pick up at any time.

From T80 review

• read_recent_dialogue chat-id pushdown: helper filters chat_id post-fetch in Python. Could push the json_extract(payload_json, '$.chat_id') = ? predicate into SQL (matching T83.3's pattern) for tighter LIMIT semantics. Currently a chat-with-many-other-chats can have its 50-row LIMIT consumed by foreign rows.
• Lifecycle warning wording in regenerate: T83.4's warning log lists ALL lifecycle event ids that exist after the original assistant_turn id, not just ones produced by the superseded turn. For the typical "regenerate the most recent" flow these are identical, but if a user regenerates an OLDER turn, the warning will list intervening-turn lifecycle events that legitimately stand. Tighten warning wording to "lifecycle transitions at-or-after turn X" (operator-friendly); a code-level fix would require a schema change to add explicit back-reference from lifecycle events to their producing turn.

From T84 review

• record_turn_memory legacy single-bot function still exists alongside the unified record_turn_memory_for_present. Could be consolidated in a follow-up.

From T86 fix-up

• Test fixtures + tests/test_phase3_integration.py that seed turns directly via append_event+project may need updating once any new test asserts the rendered HTML carries the new turn ids end-to-end. Existing tests pass because they don't read the stamped attribute, but they're brittle if the contract evolves.

Deferred items (carry-over)

• Scene-close-on-cancel UX revisit (Phase 2.5 carry-over): T74.3 pinned the existing behavior; revisit if real play-testing surfaces a regression.
• Cross-feature canned-queue brittleness: meanwhile-scene close test required a canned response for T65's digest call after T64+T65 merge. Future close-path additions will keep extending the queue. Consider a structured fixture builder rather than positional canned arrays. NOT addressed in Phase 3.5.
• Lifecycle-transition rollback in regenerate: T83.4 added a warning log; actual rollback (with proper schema linkage from lifecycle event back to producing turn) is Phase 4 work.

Phase 4 status

Phase 4 polish shipped end-to-end across 15 tasks (T88–T102). Vector retrieval is functional via pure-Python cosine over a JSON-blob embeddings table (sqlite-vec deferred — host Python lacks loadable extensions). Branching is data-model + drawer UI. Surgical delete with cascade preview, hide-from-view soft delete, significance review panel, snapshot UX, and cross-chat search all surface from the drawer or top-bar. Test count grew from 343 (Phase 3.5) to ~413 (+70 new tests).

• Wave 1 — schema + Phase 3.6 carry-overs (parallel):
  • T88 embeddings table + projector handlers (pure-Python cosine, JSON-blob storage; sqlite-vec deferred).
  • T89 branches table + handlers (main bootstrapped; is_active flag; partial unique index).
  • T90 Phase 3.6 carry-overs trio — read_recent_dialogue chat-id SQL pushdown, lifecycle warning wording tightening, legacy record_turn_memory removed.
• Wave 2 — services (parallel):
  • T91 embedding generation service (Phase 4 ships a deterministic SHA-256-derived pseudo-embedding; real model swap is Phase 4.5+).
  • T92 vector search service via pure-Python cosine.
  • T93 cross-chat search service (FTS5 across all owners, no witness filter — admin-style).
• Wave 3 — services (parallel):
  • T94 branching service (branch_from_event, switch_active_branch, list_branches_with_metadata).
  • T95 delete-impact computation service (cascade preview, no DB mutation).
• Wave 4 — combined retrieval (single):
  • T96 combined FTS + vector retrieval ranking via reciprocal-rank fusion (RRF, RRF_CONST=60); existing significance/recency boost applied as final pass.
• Wave 5 — memory write hook + backfill (single):
  • T97 EmbeddingWorker drains queue and emits embedding_indexed events; memory_write enqueues per memory_written; backfill_embeddings script for existing memories; ALL 4 production call sites wired (turns, regenerate, meanwhile, drawer).
• Wave 6 — drawer Phase 4 bundle (single, 5 sub-features):
  • T98.1 branching UI (Branches panel + 3 routes).
  • T98.2 significance review panel (distribution bar chart + per-memory edit).
  • T98.3 hide-from-view toggle + turn_hidden manual_edit branch.
  • T98.4 surgical delete with cascade preview (reuses existing rewind path; pre-rewind snapshot preserved).
  • T98.5 remaining v1 edits — narrative_anchor + weather drawer affordances + 2 new manual_edit branches.
• Wave 7 — UX surfaces (parallel):
  • T99 snapshot UX (manual trigger, list, restore with hard-confirm, preview).
  • T100 cross-chat search UX (top-bar form + results page).
• Wave 8 — polish (parallel):
  • T101 cross-feature integration tests (5 multi-feature scenarios).
  • T102 documentation (this section).

Phase 4.5 / 5 backlog

New follow-ups discovered during Phase 4 reviews and execution. None are blocking; pick up at any time.

From T88 review

• embeddings FK lacks ON DELETE CASCADE: deindex events are the only deletion path; if memories ever get deleted directly (raw SQL), embedding rows orphan. Defensible since projector model uses explicit deindex events, but worth a comment or ON DELETE CASCADE addition.

From T89 review

• list_branches(chat_id=...) filter leaks global branches (chat_id IS NULL) into every chat scope. Intentional? Document.
• Branch-switch to nonexistent silently leaves zero active branches — log a warning when this would happen.

From T91 review

• Real embedding model swap: Phase 4 ships pseudo-embedding (deterministic SHA-256 hash). Phase 4.5+ should swap to a real model (Featherless bge-small-en-v1.5 if available; or local sentence-transformers/all-MiniLM-L6-v2). The 384-dim is hardcoded in 0012_embeddings.sql; if dim changes, migrate first.
• timeout_s unused on pseudo path — fine, but log when non-default model falls through to fallback so misconfigured callers don't silently degrade.

From T96 review

• Duplicate MAX(id) lookup between _composite_rerank and the fused-path tail — DRY follow-up.
• fts_rank=None for vector-only rows — document downstream contract.

From T98 review

• event_id <= 0 guard in delete_turn — currently silently rewinds everything if event_id is 0. Add if event_id <= 0: 400.
• html.escape() on compute_delete_impact output rendered into the modal — defense in depth (currently model-controlled strings, but if event payload fields ever appear in descriptions, autoescape needed).
• Extract delete-impact modal HTML to a Jinja partial — testability + autoescape inheritance.

From T99 review

• Hoist datetime/timezone imports to module level in chat/web/snapshots.py.
• kind defaulting in restore/preview — reject missing kind rather than silent 404.
• created_at from file mtime vs filename-encoded timestamp — small drift if files copied; document.

From T100 review

• Hardcoded k=50 — extract to module constant.
• N+1 lookups (get_bot/get_chat/get_scene per row) — fine at k=50, revisit if k grows.
• FTS highlighting via snippet() — Phase 4 skipped this; UX nice-to-have.
• Result links chat-level only — memories table has no event_id column; deep-linking to specific turn requires schema addition.

Deferred items

• sqlite-vec swap when host Python supports enable_load_extension.
• Real embedding model with proper semantic similarity.
• Branching read-side filter: T89 ships data-model + UI but event readers don't yet consult is_active. Each branch is metadata-only labeled ranges. Consult-on-read is Phase 4.5+ work.
• Bulk significance re-rate in drawer (T98.2 deferred — only per-memory edit shipped).
• Vector index optimization (HNSW) — only relevant if memory counts grow past pure-Python feasibility.
• scene-close-on-cancel UX revisit (Phase 2.5 carry-over).
• Cross-feature canned-queue brittleness fixture builder (Phase 3 carry-over).
• Full lifecycle-rollback in regenerate — Phase 3.5 T83.4 shipped a warning log; proper rollback needs schema-level back-references (triggered_by_assistant_turn_id payload field).