merge: T75 phase 2.5 docs sweep + phase 2.6 backlog

T44 carried a defensive degrade-to-1:1 block in post_turn for the
case where chat.guest_bot_id pointed at a deleted bot. T47 then
fixed the root cause by adding a bot_reset cascade that clears
guest_bot_id from any chat that referenced the deleted bot, so the
post_turn defensive block was rendered dead.

Remove the orphan-clear branch and replace it with a comment
documenting that get_bot now returns a real row when guest_bot_id
is non-None. The cascade behavior is pinned by
test_reset_clears_guest_reference_in_other_chats in tests/test_reset.py.

CLAUDE.md

@@ -174,13 +174,7 @@ Deferred to Phase 2: second bot, group node, scene configurations, witness filte
 
 ### Phase 1.5 cleanup backlog
 
-Small follow-ups identified during Phase 1 reviews. Pick up at any time; none are blocking.
-
-- **`open_db` refactor.** `chat/web/bots.py:get_conn()` duplicates the context-manager body to add `check_same_thread=False`. Extend `open_db(path, *, check_same_thread=True)` and have `get_conn` call it directly — eliminates the duplicated PRAGMA setup and ensures any future PRAGMA tweak only happens in one place.
-- **Regenerate broadcasts `turn_html` over SSE.** Currently a refresh is needed (see T29 limitation above). Mirror the broadcast logic from `chat/web/turns.py:post_turn` after the new `assistant_turn` lands.
-- **`bot_reset` purges orphaned "you" activity rows** (see limitation above). Either delete `activity` rows by chat-membership or accept the noise indefinitely; the projection-layer fix is one extra `DELETE FROM activity WHERE entity_id='you' AND container_id IN (SELECT id FROM containers WHERE chat_id IN (...))` clause inside `_apply_bot_reset`.
-- **Drawer edits for the deferred v1 fields**: edge_trust slider, edge_summary textarea, memory pov_summary textarea, knowledge_facts add/remove. The `manual_edit` projector already supports `edge_trust` / `edge_summary` / `memory_pov_summary` target_kinds — only the routes are missing. Knowledge_facts needs a new dispatch branch.
-- **NICE trim order in prompt assembly** drops previous-scene first instead of last (T18 review). Greedy-cuts heuristic vs spec listing order; revisit if v1 play surfaces a real regression.
+All items shipped — see Phase 2.5 status below.
 
 ## Phase 2 status
 
@@ -194,15 +188,28 @@ Phase 2 shipped end-to-end across **13 tasks** (T36–T48 wave). The multi-entit
 
 ### Phase 2.5 / 3 backlog
 
-Carry-overs from Phase 2 reviews and implementer notes. None are blocking; pick up at any time.
+All items shipped — see Phase 2.5 status below.
 
-- **Interjection regenerate**: regenerate currently only acts on the addressee turn. Phase 2.5 should extend regenerate to cover the interjection turn too.
-- **Classifier-based addressee detection**: substring match is brittle (e.g., names that are common English words, or names appearing inside a quoted aside). A small classifier call could disambiguate.
-- **LLM-merged group meta-summary**: current `group_node.summary` is a naive concat of host + guest per-POV summaries. Phase 2.5 should polish with an LLM-merged group view.
-- **First-meeting gate**: the drawer's "have they met?" textarea fires every time. Phase 2.5 should check whether the host→guest edge already exists and offer a "they already know each other" toggle to skip re-seeding.
-- **Witness flag editing**: drawer doesn't allow editing memory witness flags (read-only). Phase 2.5+ may expose this.
-- **Significance for interjection memories**: the interjection's `memory_written` event doesn't enqueue a `SignificanceJob` (per the T44 implementer note). Phase 2.5 should wire this in so interjection memories are scored alongside primary turns.
-- **Stale guest reference defensive degrade in `post_turn`**: T44 added a degrade-to-1:1 when `chat.guest_bot_id` points at a deleted bot. T47 fixes the root cause (resets clear the reference); the degrade can probably be removed but is harmless.
-- **Scene close on cancel**: scene close runs even when the primary turn is cancelled. Behavior may be intentional but could be argued either way; revisit if it surfaces a real UX regression.
-- **Dual `ACTIVITIES:` block**: T43's prompt assembly adds a second `ACTIVITIES:` block for guest activity. Cleaner would be a single block with three bullets and per-bullet trim.
-- **Witness role hardcoded in prompt assembly**: `chat/services/prompt.py:436` hardcodes `witness_role="host"` regardless of which bot is speaking. Phase 2.5 should derive the role from chat membership (e.g. `"host" if speaker_bot_id == chat.host_bot_id else "guest"`) so guest-as-speaker prompts retrieve the right memory slice. Test contract pinned in `tests/test_witness_filter_multi.py`.
+## 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
+
+New follow-ups discovered during Phase 2.5 execution. None are blocking; pick up at any time.
+
+- **Frontend handler for `turn_html_replace` SSE event (from T73.1 review)**: regenerate's backend broadcast lands, but no live tab swaps the regenerated turn until a JS handler is wired. The existing `turn_html` event uses HTMX `sse-swap` to append; `turn_html_replace` ships JSON with `supersedes_id` for replacement semantics. Phase 2.6 should wire the JS to swap the prior turn's DOM node in place.
+- **Cancel/stop hook for in-flight regenerate streams (from T73 review)**: `post_turn` registers stream tasks in `_in_flight_tasks` so the user can stop them. Regenerate doesn't. A user clicking "Stop" mid-regenerate has no cancel hook today.
+- **DRY: regenerate vs post_turn (from T73 review)**: recent-dialogue assembly and prior-edges block are duplicated between `chat/services/regenerate.py` and `chat/web/turns.py`. Extract to shared helpers analogous to `_gather_state_update_inputs`.
+- **Sibling-discovery query optimization (from T73 review)**: `regenerate.py`'s sibling-assistant-turn lookup scans all non-superseded `assistant_turn` rows globally. Adding a `chat_id` predicate via JSON extraction (or a denormalized column) bounds the cost to per-chat scale.
+- **`_witness_role_for` defensive coding (from T71 review)**: helper returns `"guest"` when `host_bot_id is None`, which is wrong for Phase-1 chats. Defensive: `return "host" if host_bot_id is None or speaker_bot_id == host_bot_id else "guest"`. Not exercised by current tests; harden as a precaution.
+- **Confidence type tightening (from T74 review)**: `chat/services/addressee.py::AddresseeDecision.confidence` could be typed as `Literal["high","medium","low"]` for stricter validation. Currently `str` with a comment.
+- **Scene-close-on-cancel UX revisit**: T74.3 pinned the existing behavior (close fires even on cancel). If real play-testing surfaces a regression, revisit.

chat/services/addressee.py

@@ -0,0 +1,108 @@
+"""Addressee classifier service (T74.1).
+
+Phase 2 (T44) detected the addressee — host vs. guest — with a simple
+case-insensitive whole-word substring match against the bots' names.
+That worked for the obvious case ("BotB, what do you think?") but lost
+the long tail: pronouns, paraphrases, indirect address, narrative
+focus on a particular party. T74.1 swaps the substring helper for a
+classifier call that reads the prose holistically.
+
+The substring helper in :mod:`chat.web.turns` is kept as a fast-path
+for the no-guest case (only one bot present means there is nothing to
+classify) and as a non-breaking fallback for the regenerate path. The
+multi-entity branch in :func:`chat.web.turns.post_turn` calls
+:func:`detect_addressee` from this module.
+
+Failure mode: classifier flake or low-confidence response degrades to
+the host (the default speaker per Phase 2's host-keeps-the-floor
+bias). The decision carries ``confidence`` and ``reason`` so callers
+that want to log degraded decisions can distinguish a real "host" call
+from a fallback.
+"""
+
+from __future__ import annotations
+
+from pydantic import BaseModel
+
+from chat.llm.classify import classify
+from chat.llm.client import LLMClient
+
+
+class AddresseeDecision(BaseModel):
+    """Which present bot the user is addressing.
+
+    ``addressee_id`` is the chosen bot's id. ``confidence`` is one of
+    ``"high"`` / ``"medium"`` / ``"low"`` — callers may treat ``"low"``
+    as a soft fallback to the host. ``reason`` is a short free-form
+    string. The classifier-failure fallback uses ``reason="fallback"``
+    so it's distinguishable from a real low-confidence call.
+    """
+
+    addressee_id: str
+    confidence: str = "medium"  # "high" | "medium" | "low"
+    reason: str = ""
+
+
+_SYSTEM = (
+    "Given a user's turn prose and the names of present bots, decide "
+    "which bot the user is addressing. If the user is speaking to no "
+    "specific bot (descriptive narration, action without dialogue), "
+    "default to the host. Output strict JSON matching the schema. "
+    "The addressee_id MUST be one of the ids supplied in the user "
+    "message — do not invent ids."
+)
+
+
+async def detect_addressee(
+    client: LLMClient,
+    *,
+    classifier_model: str,
+    user_prose: str,
+    host_id: str,
+    host_name: str,
+    guest_id: str | None,
+    guest_name: str | None,
+    timeout_s: float = 30.0,
+) -> AddresseeDecision:
+    """Classify which present bot the user is addressing.
+
+    Defaults to host on classifier failure or when the classifier picks
+    an id that isn't one of the supplied ids. The caller is expected to
+    only invoke this in the multi-entity case (a guest is present);
+    when no guest is present the substring fast-path in
+    :mod:`chat.web.turns` is used instead and this function is not
+    called.
+    """
+    fallback = AddresseeDecision(
+        addressee_id=host_id, confidence="low", reason="fallback"
+    )
+    user = (
+        f"Host: {host_name} (id={host_id})\n"
+        + (
+            f"Guest: {guest_name} (id={guest_id})\n"
+            if guest_id is not None
+            else ""
+        )
+        + f"\nUser prose:\n{user_prose}"
+    )
+    decision = await classify(
+        client,
+        model=classifier_model,
+        system=_SYSTEM,
+        user=user,
+        schema=AddresseeDecision,
+        default=fallback,
+        timeout_s=timeout_s,
+    )
+    # Defensive: if the classifier returned an id outside the supplied
+    # set, treat it as a fallback to the host. This catches pathological
+    # outputs that pass schema validation but pick a phantom id.
+    valid_ids = {host_id}
+    if guest_id is not None:
+        valid_ids.add(guest_id)
+    if decision.addressee_id not in valid_ids:
+        return fallback
+    return decision
+
+
+__all__ = ["AddresseeDecision", "detect_addressee"]

chat/services/regenerate.py

@@ -26,6 +26,7 @@ Phase 1 simplifications (per the plan's "bound it" guidance):
   so affinity/trust/knowledge reflect the new output.
 - The route does not broadcast a fresh ``turn_html`` SSE event; T34
   polishes UI swaps. The user refreshes the page to see the new turn.
+  *(T73.1 closed this gap — see Phase 2.5 changes below.)*
 
 Phase 2 changes (T44):
 
@@ -42,6 +43,27 @@ Phase 2 changes (T44):
   is not invoked here. If the prior turn fired an interjection it
   remains attached to the original assistant_turn (which is superseded
   alongside the regenerated turn) — Phase 2.5 will revisit.
+
+Phase 2.5 changes:
+
+- T73.1: After the new ``assistant_turn`` lands we publish a
+  ``turn_html_replace`` SSE event carrying the rendered HTML for the
+  regenerated turn plus the original assistant_turn's event_id as
+  ``supersedes_id`` so connected tabs can swap the prior DOM node
+  in-place. We use a NEW event name (rather than re-using ``turn_html``)
+  because the existing HTMX ``sse-swap="turn_html"`` consumer expects a
+  raw-HTML body and an *append* semantic; ``turn_html_replace`` is a
+  JSON payload (sse.py auto-serialises when extra keys accompany
+  ``data``) so the front-end JS can read ``supersedes_id`` and replace
+  the right node.
+- T73.2: Interjection regeneration. When the original assistant_turn
+  group included an interjection beat we redo BOTH the primary and the
+  interjection — re-running ``detect_interjection`` against the new
+  primary text. If the classifier returns False this time we supersede
+  the original interjection without appending a replacement.
+- T73.3: The defensive degrade-to-1:1 for stale ``guest_bot_id``
+  references was removed — Phase 2 T47 fixed the root cause (resets
+  clear the reference) so the guard is dead code.
 """
 
 from __future__ import annotations
@@ -51,6 +73,7 @@ from sqlite3 import Connection
 
 from chat.config import Settings
 from chat.eventlog.log import append_and_apply, append_event
+from chat.services.interjection import detect_interjection
 from chat.services.memory_write import record_turn_memory_for_present
 from chat.services.multi_state_update import compute_state_updates_for_present
 from chat.services.prompt import assemble_narrative_prompt
@@ -58,6 +81,7 @@ from chat.state.edges import get_edge
 from chat.state.entities import get_bot, get_you
 from chat.state.world import active_scene, get_chat
 from chat.web.pubsub import publish
+from chat.web.render import render_turn_html
 
 
 async def regenerate_assistant_turn(
@@ -90,13 +114,13 @@ async def regenerate_assistant_turn(
 
     # Phase 2: surface the guest (if any) so the prompt assembler and
     # downstream multi-entity passes see the same shape post_turn does.
+    # Phase 2 T47 made bot_reset cascade-clear ``chat.guest_bot_id`` when
+    # the referenced bot is purged (verified by tests/test_reset.py), so
+    # we trust the column here: it's either a valid bot id or NULL.
     guest_bot_id = chat.get("guest_bot_id")
-    guest_bot: dict | None = None
-    if guest_bot_id is not None:
-        guest_bot = get_bot(conn, guest_bot_id)
-        if guest_bot is None:
-            # Stale guest reference — degrade to single-bot regenerate.
-            guest_bot_id = None
+    guest_bot: dict | None = (
+        get_bot(conn, guest_bot_id) if guest_bot_id is not None else None
+    )
 
     # 1. Locate the original assistant_turn event.
     row = conn.execute(
@@ -108,6 +132,33 @@ async def regenerate_assistant_turn(
         raise ValueError("assistant_turn event not found")
     original_assistant_payload = json.loads(row[0])
     original_user_turn_id = original_assistant_payload.get("user_turn_id")
+
+    # 1a. Look up any sibling interjection beat in the same turn group
+    # (T73.2). The original group is (primary + optional interjection),
+    # both pinned to the same ``user_turn_id``. The interjection has a
+    # populated ``interjection_of`` field in its payload — its speaker is
+    # the silent witness (the bot that wasn't the primary addressee).
+    # Filter on ``superseded_by IS NULL`` so prior regenerates of this
+    # group don't reappear as siblings.
+    original_interjection_event_id: int | None = None
+    original_interjection_payload: dict | None = None
+    if original_user_turn_id is not None:
+        sibling_cur = conn.execute(
+            "SELECT id, payload_json FROM event_log "
+            "WHERE kind = 'assistant_turn' "
+            "  AND id != ? "
+            "  AND superseded_by IS NULL",
+            (original_assistant_event_id,),
+        )
+        for sib_id, sib_payload_json in sibling_cur.fetchall():
+            sib_payload = json.loads(sib_payload_json)
+            if sib_payload.get("user_turn_id") != original_user_turn_id:
+                continue
+            if not sib_payload.get("interjection_of"):
+                continue
+            original_interjection_event_id = sib_id
+            original_interjection_payload = sib_payload
+            break
     # Phase 2 v2 regenerates only the addressee turn — preserve whichever
     # bot the original turn was attributed to, falling back to the host
     # for legacy rows that pre-date multi-entity support.
@@ -238,6 +289,27 @@ async def regenerate_assistant_turn(
         (new_assistant_event_id, original_assistant_event_id),
     )
 
+    # 7a. Broadcast a turn_html_replace SSE event so connected tabs can
+    # swap the prior assistant_turn DOM node in-place (T73.1, Phase 1.5
+    # backlog #2). Uses a separate event name from post_turn's
+    # ``turn_html`` (which is append-only) because regenerate is a
+    # *replace* operation — see module docstring for the rationale.
+    speaker_name_for_render = (
+        speaker_bot.get("name", "bot") if speaker_bot is not None else "bot"
+    )
+    new_turn_html = render_turn_html(
+        speaker_name_for_render, new_text, role="bot"
+    )
+    await publish(
+        chat_id,
+        {
+            "event": "turn_html_replace",
+            "data": new_turn_html,
+            "turn_id": new_assistant_event_id,
+            "supersedes_id": original_assistant_event_id,
+        },
+    )
+
     # 8. Re-run downstream classifier passes (memory write + state update
     # for every directed pair across present entities). Significance is
     # intentionally skipped on regenerate (the prior score remains
@@ -317,6 +389,234 @@ async def regenerate_assistant_turn(
             },
         )
 
+    # 9. Interjection regenerate branch (T73.2). When the original
+    # assistant_turn group included a follow-on interjection beat we need
+    # to revisit that beat against the regenerated primary. Three outcomes:
+    #
+    #   - No original interjection: nothing to do; we already short-circuit
+    #     above by leaving ``original_interjection_event_id`` as None.
+    #   - Original interjection + classifier returns True: stream a fresh
+    #     interjection from the silent witness, append it (with
+    #     ``interjection_of`` linking to the new primary speaker), and
+    #     supersede the original interjection's row. Also re-run memory
+    #     + state-update so the second beat moves edges + writes memories.
+    #   - Original interjection + classifier returns False: supersede the
+    #     original interjection without appending a replacement. The
+    #     regenerated group becomes "primary only" because the new primary
+    #     no longer warrants a follow-on. No memory / state work needed
+    #     for the absent beat.
+    #
+    # ``superseded_by`` on the original interjection's row points at the
+    # *new primary* in the no-replacement case (rather than NULL or a
+    # nonexistent id) so the row is consistently hidden by the standard
+    # ``superseded_by IS NULL`` timeline filter and the back-pointer
+    # leads somewhere meaningful for an "originally said …" affordance.
+    if original_interjection_event_id is not None and guest_bot is not None:
+        # Identify the silent witness from the original interjection's
+        # speaker_id (which is the bot that interjected last time). When
+        # we regenerate we keep the *same pair of present entities*, so
+        # the silent witness is whichever bot isn't the new primary
+        # speaker — derive it from present rather than reusing the prior
+        # speaker_id verbatim, in case the regenerated primary swapped
+        # who held the floor.
+        if speaker_bot_id == host_bot_id:
+            silent_witness = guest_bot
+        else:
+            silent_witness = host_bot
+        silent_witness_id = silent_witness.get("id")
+
+        edge_w_to_addr = get_edge(conn, silent_witness_id, speaker_bot_id) or {
+            "affinity": 50,
+            "trust": 50,
+            "summary": "",
+        }
+        edge_w_to_you = get_edge(conn, silent_witness_id, "you") or {
+            "affinity": 50,
+            "trust": 50,
+            "summary": "",
+        }
+
+        decision = await detect_interjection(
+            client,
+            classifier_model=settings.classifier_model,
+            addressee_name=speaker_bot.get("name", "bot"),
+            addressee_just_said=new_text,
+            silent_witness_name=silent_witness.get("name", "bot"),
+            silent_witness_persona=silent_witness.get("persona") or "",
+            silent_witness_edge_to_addressee=edge_w_to_addr,
+            silent_witness_edge_to_you=edge_w_to_you,
+            you_just_said=prose_for_prompt or "",
+            timeout_s=settings.classifier_timeout_s,
+        )
+
+        if decision.should_interject:
+            # Re-read recent so the just-appended primary is in the prompt.
+            interject_cur = conn.execute(
+                "SELECT id, kind, payload_json FROM event_log "
+                "WHERE kind IN ('user_turn', 'user_turn_edit', 'assistant_turn') "
+                "  AND superseded_by IS NULL AND hidden = 0 "
+                "ORDER BY id DESC LIMIT 20",
+            )
+            interject_rows = list(reversed(interject_cur.fetchall()))
+            interject_recent: list[dict] = []
+            for _eid, kind, payload_json in interject_rows:
+                p = json.loads(payload_json)
+                if p.get("chat_id") != chat_id:
+                    continue
+                if kind in ("user_turn", "user_turn_edit"):
+                    interject_recent.append(
+                        {"speaker": you_name, "text": p.get("prose", "")}
+                    )
+                else:
+                    spk = p.get("speaker_id", "bot")
+                    if spk == host_bot_id:
+                        spk_name = host_bot.get("name", "bot")
+                    elif spk == guest_bot.get("id"):
+                        spk_name = guest_bot.get("name", "bot")
+                    else:
+                        spk_name = "bot"
+                    interject_recent.append(
+                        {"speaker": spk_name, "text": p.get("text", "")}
+                    )
+            if interject_recent and interject_recent[-1].get("speaker") == you_name:
+                interject_recent = interject_recent[:-1]
+
+            interject_messages = assemble_narrative_prompt(
+                conn,
+                chat_id=chat_id,
+                speaker_bot_id=silent_witness_id,
+                addressee=speaker_bot_id,
+                user_turn_prose=prose_for_prompt or None,
+                recent_dialogue=interject_recent,
+                budget_soft=settings.narrative_budget_soft,
+                budget_hard=settings.narrative_budget_hard,
+                guest_id=guest_bot_id,
+            )
+
+            interject_accumulated: list[str] = []
+            async for chunk in client.stream(
+                interject_messages,
+                model=settings.narrative_model,
+                max_tokens=settings.narrative_max_tokens,
+                temperature=settings.narrative_temperature,
+            ):
+                interject_accumulated.append(chunk)
+                await publish(
+                    chat_id,
+                    {
+                        "event": "token",
+                        "text": chunk,
+                        "speaker_id": silent_witness_id,
+                    },
+                )
+            interject_text = "".join(interject_accumulated)
+
+            new_interjection_event_id = append_event(
+                conn,
+                kind="assistant_turn",
+                payload={
+                    "chat_id": chat_id,
+                    "speaker_id": silent_witness_id,
+                    "text": interject_text,
+                    "truncated": False,
+                    "user_turn_id": (
+                        new_user_event_id
+                        if new_user_event_id is not None
+                        else original_user_turn_id
+                    ),
+                    "regenerated_from": original_interjection_event_id,
+                    "interjection_of": speaker_bot_id,
+                },
+            )
+
+            # Supersede the original interjection by the new one.
+            conn.execute(
+                "UPDATE event_log SET superseded_by = ? WHERE id = ?",
+                (new_interjection_event_id, original_interjection_event_id),
+            )
+
+            # Broadcast a replace event so connected tabs swap the prior
+            # interjection node in-place (mirrors T73.1's primary swap).
+            interject_html = render_turn_html(
+                silent_witness.get("name", "bot"), interject_text, role="bot"
+            )
+            await publish(
+                chat_id,
+                {
+                    "event": "turn_html_replace",
+                    "data": interject_html,
+                    "turn_id": new_interjection_event_id,
+                    "supersedes_id": original_interjection_event_id,
+                },
+            )
+
+            # Memory write for the new interjection beat (one event per
+            # present witness).
+            record_turn_memory_for_present(
+                conn,
+                chat_id=chat_id,
+                host_bot_id=host_bot_id,
+                guest_bot_id=guest_bot_id,
+                narrative_text=interject_text,
+                scene_id=scene["id"] if scene else None,
+                chat_clock_at=chat.get("time"),
+            )
+
+            # Re-run the multi-pair state-update with the post-interjection
+            # dialogue tail so deltas land on the post-primary baseline.
+            recent_post_interject = recent_for_update + [
+                {
+                    "speaker": silent_witness.get("name", "bot"),
+                    "text": interject_text,
+                }
+            ]
+            prior_edges_post: dict[tuple[str, str], dict] = {}
+            for src in present_ids:
+                for tgt in present_ids:
+                    if src == tgt:
+                        continue
+                    edge = get_edge(conn, src, tgt) or {
+                        "affinity": 50,
+                        "trust": 50,
+                        "summary": "",
+                    }
+                    prior_edges_post[(src, tgt)] = edge
+
+            state_updates_post = await compute_state_updates_for_present(
+                client,
+                classifier_model=settings.classifier_model,
+                present_ids=present_ids,
+                present_names=present_names,
+                personas=personas,
+                prior_edges=prior_edges_post,
+                recent_dialogue=recent_post_interject,
+                timeout_s=settings.classifier_timeout_s,
+            )
+            for src_id, tgt_id, update in state_updates_post:
+                append_and_apply(
+                    conn,
+                    kind="edge_update",
+                    payload={
+                        "source_id": src_id,
+                        "target_id": tgt_id,
+                        "chat_id": chat_id,
+                        "affinity_delta": update.affinity_delta,
+                        "trust_delta": update.trust_delta,
+                        "knowledge_facts": update.knowledge_facts,
+                        "last_interaction_at": last_at,
+                        "last_interaction_chat_id": chat_id,
+                    },
+                )
+        else:
+            # Classifier said "no follow-on this time" — supersede the
+            # original interjection without a replacement. Point the
+            # back-pointer at the new primary so the row is consistently
+            # hidden by the standard timeline filter.
+            conn.execute(
+                "UPDATE event_log SET superseded_by = ? WHERE id = ?",
+                (new_assistant_event_id, original_interjection_event_id),
+            )
+
     return new_text
 
 

chat/state/manual_edit.py

@@ -6,7 +6,7 @@ be reversed by emitting an inverse ``manual_edit`` later. This module
 applies the new value to the appropriate target table; the snapshot of
 ``prior_value`` is taken by the route handler before this fires.
 
-Phase 1 covers four target kinds:
+Phase 1 covers five target kinds:
 - ``edge_affinity`` and ``edge_trust`` — slider edits on a specific edge,
   clamped to 0..100.
 - ``memory_significance`` — dropdown edit, clamped to 0..3.
@@ -17,8 +17,18 @@ Phase 1 covers four target kinds:
   field. Driven by T27 from the classifier's ``relationship_summary``
   output combined with the prior summary.
 
-Other §6.4 editable fields (activity verb / attention / posture,
-knowledge_facts list manipulation) are deferred to Phase 1.5.
+T72.1 (Phase 2.5) adds one list-shaped edit:
+- ``edge_knowledge_fact`` — add/remove a single fact on an edge's
+  ``knowledge_json`` list. Payload carries an ``action`` of ``"add"`` or
+  ``"remove"`` and a ``fact`` string; remove matches the first occurrence
+  by string equality so the route handler doesn't have to track fact
+  indices across re-renders.
+
+T72.3 adds a per-flag witness toggle:
+- ``memory_witness`` — flip one of ``witness_you`` / ``witness_host`` /
+  ``witness_guest`` on a memory row. Payload's ``new_value`` is a dict
+  ``{"flag": "you"|"host"|"guest", "value": 0|1}`` and ``prior_value``
+  mirrors the same shape so an inverse edit can restore the flag.
 
 Pin toggles intentionally use the existing ``memory_pin_changed`` event
 (registered in :mod:`chat.state.memory`) rather than ``manual_edit`` so
@@ -27,11 +37,14 @@ the projection writes both ``pinned`` and ``auto_pinned`` atomically.
 
 from __future__ import annotations
 
+import json
 from sqlite3 import Connection
 
 from chat.eventlog.log import Event
 from chat.eventlog.projector import on
 
+_VALID_WITNESS_FLAGS = {"you", "host", "guest"}
+
 
 def _clamp(value: int, lo: int, hi: int) -> int:
     return max(lo, min(hi, value))
@@ -87,5 +100,43 @@ def _apply_manual_edit(conn: Connection, e: Event) -> None:
                 target_id["target_id"],
             ),
         )
+    elif kind == "edge_knowledge_fact":
+        # T72.1: add or remove a single fact on an edge's knowledge list.
+        # ``target_id`` is the {"source_id", "target_id"} edge pair;
+        # ``new_value`` carries ``{"action": "add"|"remove", "fact": str}``.
+        # Remove matches by string equality (first occurrence) so callers
+        # don't have to thread a fact_index through re-rendered drawers.
+        action = new_value["action"]
+        fact = str(new_value["fact"])
+        row = conn.execute(
+            "SELECT knowledge_json FROM edges "
+            "WHERE source_id = ? AND target_id = ?",
+            (target_id["source_id"], target_id["target_id"]),
+        ).fetchone()
+        if row is not None:
+            knowledge = json.loads(row[0])
+            if action == "add":
+                knowledge.append(fact)
+            elif action == "remove" and fact in knowledge:
+                knowledge.remove(fact)
+            conn.execute(
+                "UPDATE edges SET knowledge_json = ? "
+                "WHERE source_id = ? AND target_id = ?",
+                (
+                    json.dumps(knowledge),
+                    target_id["source_id"],
+                    target_id["target_id"],
+                ),
+            )
+    elif kind == "memory_witness":
+        # T72.3: toggle one of the three witness flags on a memory row.
+        # ``new_value`` is the dict ``{"flag", "value"}``; ``prior_value``
+        # mirrors the same shape so an inverse edit restores the flag.
+        flag = new_value["flag"]
+        if flag in _VALID_WITNESS_FLAGS:
+            conn.execute(
+                f"UPDATE memories SET witness_{flag} = ? WHERE id = ?",
+                (1 if int(new_value["value"]) else 0, int(target_id)),
+            )
     # Unknown target_kind: silently no-op for v1. Future kinds (activity
-    # fields, knowledge_facts list manipulation) extend the dispatch above.
+    # fields, etc.) extend the dispatch above.

chat/templates/_drawer.html

@@ -100,24 +100,71 @@
   <section class="drawer-section">
     <h3>Add guest</h3>
     {% if available_guests %}
-      <form class="inline-edit"
+      {% set first_guest_id = available_guests[0].id %}
+      {% set first_existing = existing_guest_edges.get(first_guest_id, False) %}
+      <form class="inline-edit add-guest-form"
             hx-post="/chats/{{ chat.id }}/drawer/guest/add"
             hx-target="#drawer" hx-swap="innerHTML">
         <label>
           Bot:
-          <select name="guest_bot_id" required>
+          <select name="guest_bot_id" required class="add-guest-select">
             {% for b in available_guests %}
-              <option value="{{ b.id }}">{{ b.name }}</option>
+              <option value="{{ b.id }}"
+                      data-existing-edge="{{ 'true' if existing_guest_edges.get(b.id) else 'false' }}">
+                {{ b.name }}{% if existing_guest_edges.get(b.id) %} (already met){% endif %}
+              </option>
             {% endfor %}
           </select>
         </label>
+        <p class="muted add-guest-existing-note"
+           {% if not first_existing %}hidden{% endif %}>
+          they already know each other (edge exists from a prior chat)
+        </p>
+        <label class="add-guest-reseed-label"
+               {% if not first_existing %}hidden{% endif %}>
+          <input type="checkbox" name="reseed" value="1" class="add-guest-reseed">
+          re-seed anyway
+        </label>
         <label>
           Have they met before? Describe how (leave blank if not):
           <textarea name="relationship_prose" rows="3"
+                    class="add-guest-prose"
+                    {% if first_existing %}disabled{% endif %}
                     placeholder="e.g. Old college friends who studied physics together."></textarea>
         </label>
         <button type="submit">Add guest</button>
       </form>
+      <script>
+        (function () {
+          var form = document.currentScript.previousElementSibling;
+          while (form && !form.classList.contains('add-guest-form')) {
+            form = form.previousElementSibling;
+          }
+          if (!form) return;
+          var sel = form.querySelector('.add-guest-select');
+          var prose = form.querySelector('.add-guest-prose');
+          var reseed = form.querySelector('.add-guest-reseed');
+          var note = form.querySelector('.add-guest-existing-note');
+          var reseedLabel = form.querySelector('.add-guest-reseed-label');
+          function refresh() {
+            var opt = sel.options[sel.selectedIndex];
+            var existing = opt && opt.getAttribute('data-existing-edge') === 'true';
+            if (existing) {
+              note.removeAttribute('hidden');
+              reseedLabel.removeAttribute('hidden');
+              prose.disabled = !reseed.checked;
+            } else {
+              note.setAttribute('hidden', '');
+              reseedLabel.setAttribute('hidden', '');
+              reseed.checked = false;
+              prose.disabled = false;
+            }
+          }
+          sel.addEventListener('change', refresh);
+          reseed.addEventListener('change', refresh);
+          refresh();
+        })();
+      </script>
     {% else %}
       <p class="muted">No other bots authored yet.</p>
     {% endif %}
@@ -156,19 +203,95 @@
           </label>
           <button type="submit">Save</button>
         </form>
-        {% if edge_b2y.summary %}<p class="muted">{{ edge_b2y.summary }}</p>{% endif %}
-        {% if edge_b2y.knowledge %}
-          <details><summary>Knowledge ({{ edge_b2y.knowledge|length }})</summary>
-            <ul>{% for fact in edge_b2y.knowledge %}<li>{{ fact }}</li>{% endfor %}</ul>
-          </details>
-        {% endif %}
+        <form class="inline-edit"
+              hx-post="/chats/{{ chat.id }}/drawer/edge/trust"
+              hx-target="#drawer" hx-swap="innerHTML">
+          <input type="hidden" name="source_id" value="{{ host_bot.id }}">
+          <input type="hidden" name="target_id" value="you">
+          <label>
+            Trust:
+            <input type="range" name="new_value" min="0" max="100"
+                   value="{{ edge_b2y.trust }}"
+                   oninput="this.nextElementSibling.value = this.value">
+            <output>{{ edge_b2y.trust }}</output>
+          </label>
+          <button type="submit">Save</button>
+        </form>
+        <form class="inline-edit"
+              hx-post="/chats/{{ chat.id }}/drawer/edge/summary"
+              hx-target="#drawer" hx-swap="innerHTML">
+          <input type="hidden" name="source_id" value="{{ host_bot.id }}">
+          <input type="hidden" name="target_id" value="you">
+          <label>
+            Summary:
+            <textarea name="new_summary" rows="3" maxlength="2000">{{ edge_b2y.summary or "" }}</textarea>
+          </label>
+          <button type="submit">Save summary</button>
+        </form>
+        <details>
+          <summary>Knowledge ({{ (edge_b2y.knowledge or [])|length }})</summary>
+          {% if edge_b2y.knowledge %}
+            <ul>
+              {% for fact in edge_b2y.knowledge %}
+                <li>
+                  {{ fact }}
+                  <form class="inline-edit"
+                        hx-post="/chats/{{ chat.id }}/drawer/edge/knowledge-facts"
+                        hx-target="#drawer" hx-swap="innerHTML">
+                    <input type="hidden" name="source_id" value="{{ host_bot.id }}">
+                    <input type="hidden" name="target_id" value="you">
+                    <input type="hidden" name="action" value="remove">
+                    <input type="hidden" name="fact" value="{{ fact }}">
+                    <button type="submit">Remove</button>
+                  </form>
+                </li>
+              {% endfor %}
+            </ul>
+          {% endif %}
+          <form class="inline-edit"
+                hx-post="/chats/{{ chat.id }}/drawer/edge/knowledge-facts"
+                hx-target="#drawer" hx-swap="innerHTML">
+            <input type="hidden" name="source_id" value="{{ host_bot.id }}">
+            <input type="hidden" name="target_id" value="you">
+            <input type="hidden" name="action" value="add">
+            <label>
+              Add fact:
+              <input type="text" name="fact" maxlength="500" required>
+            </label>
+            <button type="submit">Add</button>
+          </form>
+        </details>
       </div>
     {% endif %}
     {% if edge_y2b %}
       <div class="edge-row">
         <strong>you &rarr; {{ host_bot.name }}</strong>
         <p>Affinity: {{ edge_y2b.affinity }}/100 &middot; Trust: {{ edge_y2b.trust }}/100</p>
-        {% if edge_y2b.summary %}<p class="muted">{{ edge_y2b.summary }}</p>{% endif %}
+        <form class="inline-edit"
+              hx-post="/chats/{{ chat.id }}/drawer/edge/trust"
+              hx-target="#drawer" hx-swap="innerHTML">
+          <input type="hidden" name="source_id" value="you">
+          <input type="hidden" name="target_id" value="{{ host_bot.id }}">
+          <label>
+            Trust:
+            <input type="range" name="new_value" min="0" max="100"
+                   value="{{ edge_y2b.trust }}"
+                   oninput="this.nextElementSibling.value = this.value">
+            <output>{{ edge_y2b.trust }}</output>
+          </label>
+          <button type="submit">Save</button>
+        </form>
+        <form class="inline-edit"
+              hx-post="/chats/{{ chat.id }}/drawer/edge/summary"
+              hx-target="#drawer" hx-swap="innerHTML">
+          <input type="hidden" name="source_id" value="you">
+          <input type="hidden" name="target_id" value="{{ host_bot.id }}">
+          <label>
+            Summary:
+            <textarea name="new_summary" rows="3" maxlength="2000">{{ edge_y2b.summary or "" }}</textarea>
+          </label>
+          <button type="submit">Save summary</button>
+        </form>
       </div>
     {% endif %}
     {% if not edge_b2y and not edge_y2b %}
@@ -224,6 +347,33 @@
               <input type="hidden" name="pinned" value="{{ 0 if m.pinned else 1 }}">
               <button type="submit">{{ 'Unpin' if m.pinned else 'Pin' }}</button>
             </form>
+            <div class="witness-row">
+              {% for flag in ['you', 'host', 'guest'] %}
+                {% set witnessed = m['witness_' ~ flag] %}
+                <form class="inline-edit"
+                      hx-post="/chats/{{ chat.id }}/drawer/memory/witness"
+                      hx-target="#drawer" hx-swap="innerHTML">
+                  <input type="hidden" name="memory_id" value="{{ m.id }}">
+                  <input type="hidden" name="flag" value="{{ flag }}">
+                  <input type="hidden" name="new_value" value="{{ 0 if witnessed else 1 }}">
+                  <label>
+                    <input type="checkbox" {% if witnessed %}checked{% endif %}
+                           onchange="this.form.requestSubmit()">
+                    {{ flag }}
+                  </label>
+                </form>
+              {% endfor %}
+            </div>
+            <details>
+              <summary>Edit POV summary</summary>
+              <form class="inline-edit"
+                    hx-post="/chats/{{ chat.id }}/drawer/memory/pov-summary"
+                    hx-target="#drawer" hx-swap="innerHTML">
+                <input type="hidden" name="memory_id" value="{{ m.id }}">
+                <textarea name="new_summary" rows="3" maxlength="2000">{{ m.pov_summary }}</textarea>
+                <button type="submit">Save</button>
+              </form>
+            </details>
           </li>
         {% endfor %}
       </ul>

chat/web/drawer.py

@@ -1,4 +1,4 @@
-"""Chat drawer — read view (T24) and inline edits (T25).
+"""Chat drawer — read view (T24) and inline edits (T25, T72).
 
 The GET endpoint renders an HTML partial showing the current scene +
 container, per-entity activity, host <-> you edges, pinned memories with
@@ -13,14 +13,16 @@ returning the refreshed drawer partial so HTMX can swap it in:
 * pin toggle on a memory (emits ``memory_pin_changed`` with
   ``auto_pinned=0`` so a manual pin is not subject to auto-eviction).
 
+T72 (Phase 2.5) extends the inline-edit set to cover the remaining
+§6.4 editable fields whose state-layer support already lands in the
+``manual_edit`` projector: edge trust slider, edge summary textarea,
+memory POV summary textarea, and per-edge knowledge-fact add/remove. It
+also exposes a witness-flag toggle (``you/host/guest``) per memory row
+and a "first-meeting gate" on the Add-guest form so an existing edge
+isn't quietly overwritten by a re-seed.
+
 Each ``manual_edit`` payload snapshots the prior value alongside the new
 one so a later inverse edit can restore state (§6.4 final paragraph).
-
-Other §6.4 editable fields (activity verb/attention/posture, edge_trust,
-edge summary, knowledge_facts list, memory pov_summary) are deferred to
-a Phase 1.5 follow-up — the dispatch in :mod:`chat.state.manual_edit`
-already accepts more ``target_kind`` values, so adding their routes is a
-mechanical extension.
 """
 
 from __future__ import annotations
@@ -55,6 +57,13 @@ PIN_CAP = 8
 # Recent-memories list is bounded to keep the drawer cheap to render.
 RECENT_LIMIT = 10
 
+# T72.1 caps on free-form textarea edits. Edge summaries and per-POV
+# memory summaries are drawer-driven prose — bound them so a stray paste
+# can't blow up the projected row size or the SSE drawer refresh payload.
+EDGE_SUMMARY_MAX = 2000
+MEMORY_POV_SUMMARY_MAX = 2000
+KNOWLEDGE_FACT_MAX = 500
+
 
 @router.get("/chats/{chat_id}/drawer", response_class=HTMLResponse)
 async def drawer(chat_id: str, request: Request, conn=Depends(get_conn)):
@@ -104,14 +113,25 @@ async def drawer(chat_id: str, request: Request, conn=Depends(get_conn)):
         available_guests = [
             b for b in list_bots(conn) if b["id"] != chat["host_bot_id"]
         ]
+    # T72.2 first-meeting gate: pre-compute whether a host->candidate edge
+    # already exists. Template renders the prose textarea disabled and the
+    # POST handler skips ``seed_inter_bot_edges`` (preserving the existing
+    # edge content) unless the user explicitly toggles "re-seed anyway".
+    existing_guest_edges = {
+        b["id"]: get_edge(conn, chat["host_bot_id"], b["id"]) is not None
+        for b in available_guests
+    }
     group_node = get_group_node(conn, chat_id)
 
     # Recent memories from host's POV (witness_host = 1), most recent first.
     # Raw query keeps this read self-contained — no projector helper exposes
-    # "latest N for an owner" yet and the drawer is the only consumer.
+    # "latest N for an owner" yet and the drawer is the only consumer. The
+    # three witness flags ride along so T72.3's per-row checkboxes can
+    # render the current state without a second query per memory.
     recent_rows = conn.execute(
         """
-        SELECT id, pov_summary, significance, pinned, created_at
+        SELECT id, pov_summary, significance, pinned, created_at,
+               witness_you, witness_host, witness_guest
         FROM memories
         WHERE owner_id = ? AND witness_host = 1
         ORDER BY id DESC
@@ -126,6 +146,9 @@ async def drawer(chat_id: str, request: Request, conn=Depends(get_conn)):
             "significance": r[2],
             "pinned": r[3],
             "created_at": r[4],
+            "witness_you": r[5],
+            "witness_host": r[6],
+            "witness_guest": r[7],
         }
         for r in recent_rows
     ]
@@ -152,6 +175,7 @@ async def drawer(chat_id: str, request: Request, conn=Depends(get_conn)):
             "edge_y2g": edge_y2g,
             "edge_g2y": edge_g2y,
             "available_guests": available_guests,
+            "existing_guest_edges": existing_guest_edges,
             "group_node": group_node,
             "recent_memories": recent_memories,
             "pinned": pinned,
@@ -342,6 +366,281 @@ async def toggle_memory_pin(
     return await drawer(chat_id, request, conn)
 
 
+# --- T72.1 deferred v1 drawer edits --------------------------------------
+#
+# These four endpoints round out the §6.4 editable surface — the
+# ``manual_edit`` projector already dispatches ``edge_trust``,
+# ``edge_summary``, and ``memory_pov_summary`` (T25); ``edge_knowledge_fact``
+# is a new dispatch branch added alongside this commit. Each route follows
+# the T25 pattern: snapshot the prior value, append + apply ``manual_edit``,
+# then re-render the drawer partial.
+
+
+@router.post(
+    "/chats/{chat_id}/drawer/edge/trust",
+    response_class=HTMLResponse,
+)
+async def edit_edge_trust(
+    chat_id: str,
+    request: Request,
+    source_id: str = Form(...),
+    target_id: str = Form(...),
+    new_value: int = Form(...),
+    conn=Depends(get_conn),
+):
+    chat = get_chat(conn, chat_id)
+    if chat is None:
+        raise HTTPException(status_code=404, detail=f"chat not found: {chat_id}")
+
+    if not 0 <= int(new_value) <= 100:
+        raise HTTPException(
+            status_code=400,
+            detail=f"trust must be in [0, 100], got {new_value}",
+        )
+
+    edge = get_edge(conn, source_id, target_id)
+    if edge is None:
+        raise HTTPException(
+            status_code=404,
+            detail=f"edge not found: {source_id}->{target_id}",
+        )
+
+    prior = int(edge["trust"])
+    append_and_apply(
+        conn,
+        kind="manual_edit",
+        payload={
+            "target_kind": "edge_trust",
+            "target_id": {"source_id": source_id, "target_id": target_id},
+            "prior_value": prior,
+            "new_value": int(new_value),
+        },
+    )
+    return await drawer(chat_id, request, conn)
+
+
+@router.post(
+    "/chats/{chat_id}/drawer/edge/summary",
+    response_class=HTMLResponse,
+)
+async def edit_edge_summary(
+    chat_id: str,
+    request: Request,
+    source_id: str = Form(...),
+    target_id: str = Form(...),
+    new_summary: str = Form(...),
+    conn=Depends(get_conn),
+):
+    chat = get_chat(conn, chat_id)
+    if chat is None:
+        raise HTTPException(status_code=404, detail=f"chat not found: {chat_id}")
+
+    if len(new_summary) > EDGE_SUMMARY_MAX:
+        raise HTTPException(
+            status_code=400,
+            detail=(
+                f"edge summary exceeds {EDGE_SUMMARY_MAX} chars "
+                f"(got {len(new_summary)})"
+            ),
+        )
+
+    edge = get_edge(conn, source_id, target_id)
+    if edge is None:
+        raise HTTPException(
+            status_code=404,
+            detail=f"edge not found: {source_id}->{target_id}",
+        )
+
+    prior = edge.get("summary") or ""
+    append_and_apply(
+        conn,
+        kind="manual_edit",
+        payload={
+            "target_kind": "edge_summary",
+            "target_id": {"source_id": source_id, "target_id": target_id},
+            "prior_value": prior,
+            "new_value": new_summary,
+        },
+    )
+    return await drawer(chat_id, request, conn)
+
+
+@router.post(
+    "/chats/{chat_id}/drawer/memory/pov-summary",
+    response_class=HTMLResponse,
+)
+async def edit_memory_pov_summary(
+    chat_id: str,
+    request: Request,
+    memory_id: int = Form(...),
+    new_summary: str = Form(...),
+    conn=Depends(get_conn),
+):
+    chat = get_chat(conn, chat_id)
+    if chat is None:
+        raise HTTPException(status_code=404, detail=f"chat not found: {chat_id}")
+
+    if len(new_summary) > MEMORY_POV_SUMMARY_MAX:
+        raise HTTPException(
+            status_code=400,
+            detail=(
+                f"memory pov_summary exceeds {MEMORY_POV_SUMMARY_MAX} chars "
+                f"(got {len(new_summary)})"
+            ),
+        )
+
+    # 404 when the memory either doesn't exist or belongs to a different
+    # chat — the drawer never surfaces cross-chat memories so editing one
+    # would be a path-traversal-style mistake.
+    row = conn.execute(
+        "SELECT pov_summary FROM memories WHERE id = ? AND chat_id = ?",
+        (int(memory_id), chat_id),
+    ).fetchone()
+    if row is None:
+        raise HTTPException(
+            status_code=404,
+            detail=f"memory not found in chat: {memory_id}",
+        )
+
+    prior = row[0] or ""
+    append_and_apply(
+        conn,
+        kind="manual_edit",
+        payload={
+            "target_kind": "memory_pov_summary",
+            "target_id": int(memory_id),
+            "prior_value": prior,
+            "new_value": new_summary,
+        },
+    )
+    return await drawer(chat_id, request, conn)
+
+
+@router.post(
+    "/chats/{chat_id}/drawer/edge/knowledge-facts",
+    response_class=HTMLResponse,
+)
+async def edit_edge_knowledge_facts(
+    chat_id: str,
+    request: Request,
+    source_id: str = Form(...),
+    target_id: str = Form(...),
+    action: str = Form(...),
+    fact: str = Form(...),
+    conn=Depends(get_conn),
+):
+    """Add or remove a single knowledge_fact on an edge.
+
+    Remove semantics are by string match (first occurrence) — the drawer
+    re-renders after every edit so threading a stable index through is
+    fragile when concurrent ``edge_update`` events can append more facts
+    between renders. The projector is a no-op when the fact isn't found,
+    keeping the route idempotent for stale form submissions.
+    """
+    chat = get_chat(conn, chat_id)
+    if chat is None:
+        raise HTTPException(status_code=404, detail=f"chat not found: {chat_id}")
+
+    if action not in ("add", "remove"):
+        raise HTTPException(
+            status_code=400,
+            detail=f"action must be 'add' or 'remove', got {action!r}",
+        )
+
+    if len(fact) > KNOWLEDGE_FACT_MAX:
+        raise HTTPException(
+            status_code=400,
+            detail=(
+                f"fact exceeds {KNOWLEDGE_FACT_MAX} chars (got {len(fact)})"
+            ),
+        )
+    if not fact.strip():
+        raise HTTPException(status_code=400, detail="fact must not be empty")
+
+    edge = get_edge(conn, source_id, target_id)
+    if edge is None:
+        raise HTTPException(
+            status_code=404,
+            detail=f"edge not found: {source_id}->{target_id}",
+        )
+
+    prior = list(edge.get("knowledge") or [])
+    append_and_apply(
+        conn,
+        kind="manual_edit",
+        payload={
+            "target_kind": "edge_knowledge_fact",
+            "target_id": {"source_id": source_id, "target_id": target_id},
+            "prior_value": prior,
+            "new_value": {"action": action, "fact": fact},
+        },
+    )
+    return await drawer(chat_id, request, conn)
+
+
+# --- T72.3 witness flag inline-edit --------------------------------------
+#
+# Witness flags decide which entities can recall a memory (§7 retrieval).
+# Editing them is rare but high-impact — flipping ``witness_guest`` from 0
+# to 1 makes the memory available to the guest's prompt context. The route
+# follows the T25 / T72.1 pattern: snapshot prior, append + apply
+# ``manual_edit`` with a ``{flag, value}`` payload, refresh the partial.
+
+
+_VALID_WITNESS_FLAGS = ("you", "host", "guest")
+
+
+@router.post(
+    "/chats/{chat_id}/drawer/memory/witness",
+    response_class=HTMLResponse,
+)
+async def edit_memory_witness(
+    chat_id: str,
+    request: Request,
+    memory_id: int = Form(...),
+    flag: str = Form(...),
+    new_value: int = Form(...),
+    conn=Depends(get_conn),
+):
+    chat = get_chat(conn, chat_id)
+    if chat is None:
+        raise HTTPException(status_code=404, detail=f"chat not found: {chat_id}")
+
+    if flag not in _VALID_WITNESS_FLAGS:
+        raise HTTPException(
+            status_code=400,
+            detail=(
+                f"flag must be one of {list(_VALID_WITNESS_FLAGS)}, "
+                f"got {flag!r}"
+            ),
+        )
+
+    row = conn.execute(
+        f"SELECT witness_{flag} FROM memories "
+        "WHERE id = ? AND chat_id = ?",
+        (int(memory_id), chat_id),
+    ).fetchone()
+    if row is None:
+        raise HTTPException(
+            status_code=404,
+            detail=f"memory not found in chat: {memory_id}",
+        )
+
+    prior_int = int(row[0])
+    new_int = 1 if int(new_value) else 0
+    append_and_apply(
+        conn,
+        kind="manual_edit",
+        payload={
+            "target_kind": "memory_witness",
+            "target_id": int(memory_id),
+            "prior_value": {"flag": flag, "value": prior_int},
+            "new_value": {"flag": flag, "value": new_int},
+        },
+    )
+    return await drawer(chat_id, request, conn)
+
+
 # --- T42 guest add/remove -------------------------------------------------
 #
 # Adding a guest fans out into up to four events: a ``guest_added`` to flip
@@ -381,6 +680,7 @@ async def add_guest(
     request: Request,
     guest_bot_id: str = Form(...),
     relationship_prose: str = Form(""),
+    reseed: str = Form(""),
     conn=Depends(get_conn),
     client=Depends(get_llm_client),
 ):
@@ -412,17 +712,32 @@ async def add_guest(
             detail=f"host bot not found: {chat['host_bot_id']}",
         )
 
-    settings = request.app.state.settings
-    seed = await seed_inter_bot_edges(
-        client,
-        classifier_model=settings.classifier_model,
-        bot_a_id=chat["host_bot_id"],
-        bot_a_name=host_bot["name"],
-        bot_b_id=guest_bot_id,
-        bot_b_name=guest_bot["name"],
-        relationship_prose=relationship_prose,
-        timeout_s=settings.classifier_timeout_s,
+    # T72.2 first-meeting gate: when an edge already exists from a prior
+    # chat, the textarea is rendered disabled. Submission without the
+    # explicit "re-seed anyway" toggle skips ``seed_inter_bot_edges``
+    # entirely so the existing edge content (affinity, trust, knowledge,
+    # summaries) survives. ``guest_added`` and ``group_node_initialized``
+    # still fire so the chat picks up the new participant.
+    existing_edge = (
+        get_edge(conn, chat["host_bot_id"], guest_bot_id) is not None
     )
+    reseed_requested = reseed.lower() in ("1", "true", "on", "yes")
+    skip_seed = existing_edge and not reseed_requested
+
+    settings = request.app.state.settings
+    if skip_seed:
+        seed = None
+    else:
+        seed = await seed_inter_bot_edges(
+            client,
+            classifier_model=settings.classifier_model,
+            bot_a_id=chat["host_bot_id"],
+            bot_a_name=host_bot["name"],
+            bot_b_id=guest_bot_id,
+            bot_b_name=guest_bot["name"],
+            relationship_prose=relationship_prose,
+            timeout_s=settings.classifier_timeout_s,
+        )
 
     append_and_apply(
         conn,
@@ -437,7 +752,7 @@ async def add_guest(
     # per-direction summary is set via the per-pov scene-close path
     # (T27), not direct edge_update. We therefore drop seed.*_summary
     # here; the deltas + knowledge_facts are what materializes.
-    if not _seed_is_default(seed):
+    if seed is not None and not _seed_is_default(seed):
         append_and_apply(
             conn,
             kind="edge_update",

chat/web/turns.py

@@ -55,6 +55,7 @@ from fastapi import APIRouter, Depends, Form, HTTPException, Request
 from fastapi.responses import HTMLResponse, RedirectResponse, Response
 
 from chat.eventlog.log import append_and_apply, append_event
+from chat.services.addressee import detect_addressee
 from chat.services.background import SignificanceJob
 from chat.services.interjection import detect_interjection
 from chat.services.memory_write import record_turn_memory_for_present
@@ -235,11 +236,12 @@ async def post_turn(
     guest_bot = None
     guest_bot_id = chat.get("guest_bot_id")
     if guest_bot_id is not None:
+        # T47's bot_reset cascade clears guest_bot_id from any chat that
+        # referenced the deleted bot, so by the time we read it here it's
+        # either None or a live bot id. The previous defensive
+        # degrade-to-1:1 block (T44) was rendered dead by T47 and removed
+        # in T74.4 — get_bot now returns a real row.
         guest_bot = get_bot(conn, guest_bot_id)
-        # If the chat references a deleted guest we degrade to single-bot
-        # rather than 404 — the chat is still usable as a 1:1.
-        if guest_bot is None:
-            guest_bot_id = None
 
     settings = request.app.state.settings
 
@@ -262,8 +264,25 @@ async def post_turn(
 
     # 3. Determine the addressee. Done before assistant_turn_started so the
     # placeholder reflects the bot the user is actually talking to (host
-    # in 1:1, host-or-guest in multi-entity).
-    addressee_id = _detect_addressee_id(prose, host_bot, guest_bot)
+    # in 1:1, host-or-guest in multi-entity). T74.1 routes the multi-entity
+    # case through the addressee classifier; the no-guest case still uses
+    # the substring fast-path because there is nothing to classify when
+    # only one bot is present (and a classifier round-trip there would
+    # just be throughput overhead).
+    if guest_bot is None:
+        addressee_id = _detect_addressee_id(prose, host_bot, guest_bot)
+    else:
+        decision = await detect_addressee(
+            client,
+            classifier_model=settings.classifier_model,
+            user_prose=prose,
+            host_id=host_bot["id"],
+            host_name=host_bot["name"],
+            guest_id=guest_bot["id"],
+            guest_name=guest_bot["name"],
+            timeout_s=settings.classifier_timeout_s,
+        )
+        addressee_id = decision.addressee_id
     addressee_bot = (
         guest_bot if (guest_bot is not None and addressee_id == guest_bot["id"])
         else host_bot
@@ -598,7 +617,7 @@ async def post_turn(
 
                 # Memory write for the interjection beat — a second pair
                 # of memory_written events (host + guest POVs).
-                record_turn_memory_for_present(
+                interject_memory_results = record_turn_memory_for_present(
                     conn,
                     chat_id=chat_id,
                     host_bot_id=host_bot["id"],
@@ -608,6 +627,33 @@ async def post_turn(
                     chat_clock_at=chat.get("time"),
                 )
 
+                # T74.2: enqueue a significance pass for the interjection
+                # memory. Mirrors the primary-turn enqueue pattern above —
+                # we score on the host's memory id since the prose is
+                # identical across both POVs (per-POV rewrite happens at
+                # scene close in T45). Without this enqueue the
+                # interjection beat lands in memory but never gets scored,
+                # so it can never auto-pin even when it carries a pivotal
+                # moment.
+                interject_host_event = interject_memory_results.get(
+                    host_bot["id"]
+                )
+                interject_host_memory_id = (
+                    interject_host_event[1] if interject_host_event else None
+                )
+                if (
+                    worker is not None
+                    and interject_host_memory_id is not None
+                ):
+                    worker.enqueue(
+                        SignificanceJob(
+                            memory_id=interject_host_memory_id,
+                            narrative_text=interjection_text,
+                            prior_dialogue=recent_post_interject,
+                            host_bot_id=host_bot["id"],
+                        )
+                    )
+
     # 9. Scene-close detection (Plan §7.2, T26). Runs AFTER assistant_turn
     # and the optional interjection so the bots' responses are part of
     # the closing scene's final beat — closing before narrative would
@@ -623,6 +669,15 @@ async def post_turn(
     # close in the same chat) — we have nothing to close. T13 (kickoff)
     # is the only scene-opener path in v1; Phase 2-3 will handle
     # automatic re-opening with the next container.
+    #
+    # T74.3: this branch deliberately runs even when ``cancelled`` is
+    # True. Close detection consumes only the user's prose (which is
+    # fully appended to the event_log BEFORE streaming starts) and the
+    # current container name; it does NOT consume the bot's output.
+    # A user who types "we're done here, fade out" and then hits Stop
+    # mid-stream still meant to close the scene — the cancelled bot
+    # beat doesn't invalidate that intent. Pinned by
+    # test_cancelled_turn_still_closes_scene_when_user_prose_signals_close.
     if scene is not None and prose.strip():
         container = None
         if scene.get("container_id") is not None:

tests/test_addressee.py

@@ -0,0 +1,99 @@
+"""Addressee classifier service tests (T74.1).
+
+Covers :func:`chat.services.addressee.detect_addressee`:
+
+- Classifier picks the guest -> ``addressee_id == guest_id``.
+- Classifier picks the host -> ``addressee_id == host_id``.
+- Classifier flakes (3 bad-JSON responses, exhausting the built-in
+  retry budget in :func:`chat.llm.classify.classify`) -> fallback to
+  the host with ``reason="fallback"``.
+"""
+
+from __future__ import annotations
+
+import json
+
+import pytest
+
+from chat.llm.mock import MockLLMClient
+from chat.services.addressee import AddresseeDecision, detect_addressee
+
+
+@pytest.mark.asyncio
+async def test_classifier_picks_guest():
+    """Classifier returns the guest id verbatim — caller propagates it."""
+    canned = [
+        json.dumps(
+            {
+                "addressee_id": "bot_b",
+                "confidence": "high",
+                "reason": "user named BotB",
+            }
+        )
+    ]
+    client = MockLLMClient(canned=canned)
+
+    result = await detect_addressee(
+        client,
+        classifier_model="test-model",
+        user_prose="BotB, what do you think?",
+        host_id="bot_a",
+        host_name="BotA",
+        guest_id="bot_b",
+        guest_name="BotB",
+    )
+
+    assert isinstance(result, AddresseeDecision)
+    assert result.addressee_id == "bot_b"
+    assert result.confidence == "high"
+
+
+@pytest.mark.asyncio
+async def test_classifier_picks_host():
+    """Classifier returns the host id — caller propagates it."""
+    canned = [
+        json.dumps(
+            {
+                "addressee_id": "bot_a",
+                "confidence": "medium",
+                "reason": "narration aimed at host",
+            }
+        )
+    ]
+    client = MockLLMClient(canned=canned)
+
+    result = await detect_addressee(
+        client,
+        classifier_model="test-model",
+        user_prose="I lean back and stretch.",
+        host_id="bot_a",
+        host_name="BotA",
+        guest_id="bot_b",
+        guest_name="BotB",
+    )
+
+    assert result.addressee_id == "bot_a"
+    assert result.confidence == "medium"
+
+
+@pytest.mark.asyncio
+async def test_classifier_failure_falls_back_to_host():
+    """Three bad-JSON responses exhaust the retry budget and the
+    classifier-failure fallback returns ``host_id`` with
+    ``reason="fallback"``."""
+    canned = ["not json", "still not json", "garbage"]
+    client = MockLLMClient(canned=canned)
+
+    result = await detect_addressee(
+        client,
+        classifier_model="test-model",
+        user_prose="anything",
+        host_id="bot_a",
+        host_name="BotA",
+        guest_id="bot_b",
+        guest_name="BotB",
+    )
+
+    assert result.addressee_id == "bot_a"
+    assert result.reason == "fallback"
+    assert result.confidence == "low"

tests/test_drawer_edits_extended.py

@@ -0,0 +1,403 @@
+"""T72: deferred v1 drawer edits + witness flag inline-edit.
+
+T25 shipped affinity / significance / pin. T72.1 fills in the rest of the
+§6.4 editable surface whose ``manual_edit`` projector dispatch was already
+in place (or, for ``edge_knowledge_fact``, added alongside the route):
+
+* ``POST /chats/{chat_id}/drawer/edge/trust`` — slider 0..100.
+* ``POST /chats/{chat_id}/drawer/edge/summary`` — textarea, capped 2000.
+* ``POST /chats/{chat_id}/drawer/memory/pov-summary`` — textarea, capped.
+* ``POST /chats/{chat_id}/drawer/edge/knowledge-facts`` — add/remove fact.
+
+T72.3 layers a witness-flag toggle on top:
+
+* ``POST /chats/{chat_id}/drawer/memory/witness`` — ``manual_edit`` with
+  ``target_kind`` = ``memory_witness`` and a ``{flag, value}`` payload.
+
+Each test asserts (a) the ``manual_edit`` event lands in the log,
+(b) the projected table reflects the new value, and (c) the response is
+the refreshed drawer partial.
+"""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+import pytest
+from fastapi.testclient import TestClient
+
+from chat.app import app
+from chat.db.connection import open_db
+from chat.eventlog.log import append_event
+from chat.eventlog.projector import project
+
+
+@pytest.fixture
+def client(tmp_path, monkeypatch):
+    cfg = tmp_path / "config.toml"
+    cfg.write_text('featherless_api_key = "test"\n')
+    monkeypatch.setenv("CHAT_CONFIG_PATH", str(cfg))
+    db = tmp_path / "test.db"
+    monkeypatch.setenv("CHAT_DB_PATH", str(db))
+    with TestClient(app) as c:
+        if hasattr(app.state, "background_worker"):
+            app.state.background_worker.enabled = False
+        yield c
+
+
+def _seed(db: Path) -> None:
+    """Seed a chat with one host bot, one host->you edge with a fact and
+    summary already set, and one memory authored by ``bot_a`` witnessed by
+    all three roles. Tests reach into projected state to verify edits.
+    """
+    with open_db(db) as conn:
+        append_event(
+            conn,
+            kind="bot_authored",
+            payload={
+                "id": "bot_a",
+                "name": "BotA",
+                "persona": "...",
+                "voice_samples": [],
+                "traits": [],
+                "backstory": "",
+                "initial_relationship_to_you": "",
+                "kickoff_prose": "",
+            },
+        )
+        append_event(
+            conn,
+            kind="chat_created",
+            payload={
+                "id": "chat_bot_a",
+                "host_bot_id": "bot_a",
+                "initial_time": "2026-04-26T20:00:00+00:00",
+                "narrative_anchor": "Day 1",
+                "weather": "",
+            },
+        )
+        # Materialise edge bot_a -> you with a knowledge_fact already on it
+        # so the remove path has something to consume.
+        append_event(
+            conn,
+            kind="edge_update",
+            payload={
+                "source_id": "bot_a",
+                "target_id": "you",
+                "chat_id": "chat_bot_a",
+                "affinity_delta": 0,
+                "knowledge_facts": ["studied physics together"],
+            },
+        )
+        append_event(
+            conn,
+            kind="memory_written",
+            payload={
+                "owner_id": "bot_a",
+                "chat_id": "chat_bot_a",
+                "pov_summary": "Original summary text.",
+                "witness_you": 1,
+                "witness_host": 1,
+                "witness_guest": 0,
+                "significance": 1,
+            },
+        )
+        project(conn)
+
+
+# --- T72.1 tests ----------------------------------------------------------
+
+
+def test_edit_edge_trust_emits_manual_edit_and_updates(client, tmp_path):
+    _seed(tmp_path / "test.db")
+    response = client.post(
+        "/chats/chat_bot_a/drawer/edge/trust",
+        data={"source_id": "bot_a", "target_id": "you", "new_value": "73"},
+    )
+    assert response.status_code == 200
+    # Refresh shows the new trust value somewhere in the partial.
+    assert "73" in response.text
+
+    with open_db(tmp_path / "test.db") as conn:
+        rows = conn.execute(
+            "SELECT payload_json FROM event_log WHERE kind = 'manual_edit'"
+        ).fetchall()
+        assert len(rows) == 1
+        payload = json.loads(rows[0][0])
+        assert payload["target_kind"] == "edge_trust"
+        assert payload["prior_value"] == 50
+        assert payload["new_value"] == 73
+        assert payload["target_id"] == {
+            "source_id": "bot_a",
+            "target_id": "you",
+        }
+
+        from chat.state.edges import get_edge
+
+        edge = get_edge(conn, "bot_a", "you")
+        assert edge["trust"] == 73
+
+
+def test_edit_edge_trust_400_on_out_of_range(client, tmp_path):
+    _seed(tmp_path / "test.db")
+    response = client.post(
+        "/chats/chat_bot_a/drawer/edge/trust",
+        data={"source_id": "bot_a", "target_id": "you", "new_value": "150"},
+    )
+    assert response.status_code == 400
+
+
+def test_edit_edge_summary_emits_manual_edit_and_updates(client, tmp_path):
+    _seed(tmp_path / "test.db")
+    response = client.post(
+        "/chats/chat_bot_a/drawer/edge/summary",
+        data={
+            "source_id": "bot_a",
+            "target_id": "you",
+            "new_summary": "BotA respects you and shares lab notes.",
+        },
+    )
+    assert response.status_code == 200
+
+    with open_db(tmp_path / "test.db") as conn:
+        rows = conn.execute(
+            "SELECT payload_json FROM event_log WHERE kind = 'manual_edit'"
+        ).fetchall()
+        assert len(rows) == 1
+        payload = json.loads(rows[0][0])
+        assert payload["target_kind"] == "edge_summary"
+        assert payload["new_value"].startswith("BotA respects")
+        assert payload["target_id"] == {
+            "source_id": "bot_a",
+            "target_id": "you",
+        }
+
+        summary = conn.execute(
+            "SELECT summary FROM edges "
+            "WHERE source_id = ? AND target_id = ?",
+            ("bot_a", "you"),
+        ).fetchone()[0]
+        assert "respects" in summary
+
+    # And the refreshed partial echoes the new summary back.
+    assert "respects" in response.text
+
+
+def test_edit_edge_summary_400_on_overflow(client, tmp_path):
+    _seed(tmp_path / "test.db")
+    response = client.post(
+        "/chats/chat_bot_a/drawer/edge/summary",
+        data={
+            "source_id": "bot_a",
+            "target_id": "you",
+            "new_summary": "x" * 2001,
+        },
+    )
+    assert response.status_code == 400
+
+
+def test_edit_memory_pov_summary_emits_manual_edit_and_updates(
+    client, tmp_path
+):
+    _seed(tmp_path / "test.db")
+    with open_db(tmp_path / "test.db") as conn:
+        memory_id = conn.execute("SELECT id FROM memories LIMIT 1").fetchone()[0]
+
+    response = client.post(
+        "/chats/chat_bot_a/drawer/memory/pov-summary",
+        data={
+            "memory_id": str(memory_id),
+            "new_summary": "Cleaner per-POV restatement of the moment.",
+        },
+    )
+    assert response.status_code == 200
+
+    with open_db(tmp_path / "test.db") as conn:
+        rows = conn.execute(
+            "SELECT payload_json FROM event_log WHERE kind = 'manual_edit'"
+        ).fetchall()
+        assert len(rows) == 1
+        payload = json.loads(rows[0][0])
+        assert payload["target_kind"] == "memory_pov_summary"
+        assert payload["prior_value"] == "Original summary text."
+        assert payload["new_value"].startswith("Cleaner per-POV")
+        assert payload["target_id"] == memory_id
+
+        pov = conn.execute(
+            "SELECT pov_summary FROM memories WHERE id = ?", (memory_id,)
+        ).fetchone()[0]
+        assert pov.startswith("Cleaner per-POV")
+
+    assert "Cleaner per-POV" in response.text
+
+
+def test_edit_memory_pov_summary_404_when_wrong_chat(client, tmp_path):
+    _seed(tmp_path / "test.db")
+    with open_db(tmp_path / "test.db") as conn:
+        memory_id = conn.execute("SELECT id FROM memories LIMIT 1").fetchone()[0]
+        # Re-home the memory to a different chat to confirm the route's
+        # cross-chat guard fires.
+        conn.execute(
+            "UPDATE memories SET chat_id = 'other_chat' WHERE id = ?",
+            (memory_id,),
+        )
+        conn.commit()
+
+    response = client.post(
+        "/chats/chat_bot_a/drawer/memory/pov-summary",
+        data={"memory_id": str(memory_id), "new_summary": "..."},
+    )
+    assert response.status_code == 404
+
+
+def test_edit_edge_knowledge_facts_add_emits_event_and_appends(client, tmp_path):
+    _seed(tmp_path / "test.db")
+    response = client.post(
+        "/chats/chat_bot_a/drawer/edge/knowledge-facts",
+        data={
+            "source_id": "bot_a",
+            "target_id": "you",
+            "action": "add",
+            "fact": "lent you a textbook",
+        },
+    )
+    assert response.status_code == 200
+
+    with open_db(tmp_path / "test.db") as conn:
+        rows = conn.execute(
+            "SELECT payload_json FROM event_log WHERE kind = 'manual_edit'"
+        ).fetchall()
+        assert len(rows) == 1
+        payload = json.loads(rows[0][0])
+        assert payload["target_kind"] == "edge_knowledge_fact"
+        assert payload["new_value"] == {
+            "action": "add",
+            "fact": "lent you a textbook",
+        }
+        # Prior value snapshots the entire knowledge list before the edit.
+        assert payload["prior_value"] == ["studied physics together"]
+
+        from chat.state.edges import get_edge
+
+        edge = get_edge(conn, "bot_a", "you")
+        assert "lent you a textbook" in edge["knowledge"]
+        assert "studied physics together" in edge["knowledge"]
+
+    assert "lent you a textbook" in response.text
+
+
+def test_edit_edge_knowledge_facts_remove_drops_matching_fact(client, tmp_path):
+    _seed(tmp_path / "test.db")
+    response = client.post(
+        "/chats/chat_bot_a/drawer/edge/knowledge-facts",
+        data={
+            "source_id": "bot_a",
+            "target_id": "you",
+            "action": "remove",
+            "fact": "studied physics together",
+        },
+    )
+    assert response.status_code == 200
+
+    with open_db(tmp_path / "test.db") as conn:
+        from chat.state.edges import get_edge
+
+        edge = get_edge(conn, "bot_a", "you")
+        assert "studied physics together" not in edge["knowledge"]
+
+        rows = conn.execute(
+            "SELECT payload_json FROM event_log WHERE kind = 'manual_edit'"
+        ).fetchall()
+        payload = json.loads(rows[0][0])
+        assert payload["target_kind"] == "edge_knowledge_fact"
+        assert payload["new_value"]["action"] == "remove"
+
+
+def test_edit_edge_knowledge_facts_400_on_bad_action(client, tmp_path):
+    _seed(tmp_path / "test.db")
+    response = client.post(
+        "/chats/chat_bot_a/drawer/edge/knowledge-facts",
+        data={
+            "source_id": "bot_a",
+            "target_id": "you",
+            "action": "delete",
+            "fact": "x",
+        },
+    )
+    assert response.status_code == 400
+
+
+# --- T72.3 tests (witness flag inline-edit) -------------------------------
+
+
+def test_witness_flag_toggle_updates_memory_row(client, tmp_path):
+    """Memory seeded with witness [you=1, host=1, guest=0]; toggling
+    ``guest`` to 1 lands as ``witness_guest = 1`` after projection.
+    """
+    _seed(tmp_path / "test.db")
+    with open_db(tmp_path / "test.db") as conn:
+        memory_id = conn.execute("SELECT id FROM memories LIMIT 1").fetchone()[0]
+
+    response = client.post(
+        "/chats/chat_bot_a/drawer/memory/witness",
+        data={
+            "memory_id": str(memory_id),
+            "flag": "guest",
+            "new_value": "1",
+        },
+    )
+    assert response.status_code == 200
+
+    with open_db(tmp_path / "test.db") as conn:
+        row = conn.execute(
+            "SELECT witness_you, witness_host, witness_guest "
+            "FROM memories WHERE id = ?",
+            (memory_id,),
+        ).fetchone()
+        assert row == (1, 1, 1)
+
+
+def test_witness_flag_toggle_emits_manual_edit_event(client, tmp_path):
+    _seed(tmp_path / "test.db")
+    with open_db(tmp_path / "test.db") as conn:
+        memory_id = conn.execute("SELECT id FROM memories LIMIT 1").fetchone()[0]
+
+    response = client.post(
+        "/chats/chat_bot_a/drawer/memory/witness",
+        data={
+            "memory_id": str(memory_id),
+            "flag": "guest",
+            "new_value": "1",
+        },
+    )
+    assert response.status_code == 200
+
+    with open_db(tmp_path / "test.db") as conn:
+        rows = conn.execute(
+            "SELECT payload_json FROM event_log WHERE kind = 'manual_edit'"
+        ).fetchall()
+        assert len(rows) == 1
+        payload = json.loads(rows[0][0])
+        assert payload["target_kind"] == "memory_witness"
+        assert payload["target_id"] == memory_id
+        assert payload["prior_value"] == {"flag": "guest", "value": 0}
+        assert payload["new_value"] == {"flag": "guest", "value": 1}
+
+
+def test_witness_flag_toggle_400_on_bad_flag(client, tmp_path):
+    _seed(tmp_path / "test.db")
+    with open_db(tmp_path / "test.db") as conn:
+        memory_id = conn.execute("SELECT id FROM memories LIMIT 1").fetchone()[0]
+
+    response = client.post(
+        "/chats/chat_bot_a/drawer/memory/witness",
+        data={
+            "memory_id": str(memory_id),
+            "flag": "narrator",
+            "new_value": "1",
+        },
+    )
+    assert response.status_code == 400
+
+

tests/test_drawer_guest.py

@@ -265,6 +265,162 @@ def test_drawer_remove_guest_clears_and_closes_scene(client, tmp_path):
         assert kinds.index("scene_closed") < kinds.index("guest_removed")
 
 
+# --- T72.2 first-meeting gate ----------------------------------------------
+
+
+def _seed_host_to_guest_edge(db: Path) -> None:
+    """Materialise a bot_a -> bot_b edge so the gate's check fires."""
+    from chat.eventlog.log import append_and_apply
+
+    with open_db(db) as conn:
+        append_and_apply(
+            conn,
+            kind="edge_update",
+            payload={
+                "source_id": "bot_a",
+                "target_id": "bot_b",
+                "chat_id": "chat_bot_a",
+                "affinity_delta": 0,
+                "knowledge_facts": ["already met before"],
+            },
+        )
+
+
+def test_add_guest_form_disables_prose_when_edge_exists(client, tmp_path):
+    """When host->candidate edge already exists, the GET partial renders
+    the textarea disabled and surfaces the "already know each other"
+    message so the user knows submitting will skip the seed.
+    """
+    _seed_chat(tmp_path / "test.db")
+    _seed_host_to_guest_edge(tmp_path / "test.db")
+
+    response = client.get("/chats/chat_bot_a/drawer")
+    assert response.status_code == 200
+    body = response.text
+    # Note + disabled state both present. The textarea sits next to the
+    # ``add-guest-prose`` class so we can match it specifically.
+    assert "already know each other" in body
+    assert 'class="add-guest-prose"' in body
+    # The textarea for the first (auto-selected) candidate should be
+    # disabled in the initial markup since an edge exists.
+    assert "disabled" in body.split('class="add-guest-prose"', 1)[1].split(">", 1)[0]
+    # And the option carries the ``data-existing-edge="true"`` attribute
+    # the inline JS uses to flip state on subsequent select changes.
+    assert 'data-existing-edge="true"' in body
+
+
+def test_add_guest_with_existing_edge_skips_seed_call(client, tmp_path):
+    """Submitting the Add-guest form WITHOUT toggling re-seed must skip
+    ``seed_inter_bot_edges`` entirely. We assert this via an empty mock
+    queue: if the seed function had been called it would have consumed
+    a canned response (or raised because none was available).
+    """
+    _seed_chat(tmp_path / "test.db")
+    _seed_host_to_guest_edge(tmp_path / "test.db")
+
+    # Empty queue: any classifier call would raise inside MockLLMClient.
+    canned_queue: list[str] = []
+    _override_llm(canned_queue)
+    try:
+        response = client.post(
+            "/chats/chat_bot_a/drawer/guest/add",
+            data={
+                "guest_bot_id": "bot_b",
+                "relationship_prose": "ignored prose",
+                # NO reseed flag — gate should suppress the seed call.
+            },
+        )
+        assert response.status_code == 200
+    finally:
+        app.dependency_overrides.clear()
+
+    with open_db(tmp_path / "test.db") as conn:
+        from chat.state.edges import get_edge
+        from chat.state.world import get_chat
+
+        chat = get_chat(conn, "chat_bot_a")
+        assert chat["guest_bot_id"] == "bot_b"
+
+        # The pre-seeded knowledge fact survives — proof the seed didn't run
+        # and overwrite the existing edge.
+        edge = get_edge(conn, "bot_a", "bot_b")
+        assert "already met before" in edge["knowledge"]
+
+        # Exactly one guest_added; no new edge_update events between
+        # bot_a and bot_b (the pre-seed edge_update from the test setup
+        # is the only edge_update on this pair).
+        added = conn.execute(
+            "SELECT COUNT(*) FROM event_log WHERE kind = 'guest_added'"
+        ).fetchone()[0]
+        assert added == 1
+
+        edge_updates = conn.execute(
+            "SELECT payload_json FROM event_log WHERE kind = 'edge_update'"
+        ).fetchall()
+        # Only the pre-seed edge_update from _seed_host_to_guest_edge.
+        ab_updates = [
+            json.loads(p[0])
+            for p in edge_updates
+            if {
+                json.loads(p[0]).get("source_id"),
+                json.loads(p[0]).get("target_id"),
+            }
+            == {"bot_a", "bot_b"}
+        ]
+        assert len(ab_updates) == 1
+        assert ab_updates[0]["knowledge_facts"] == ["already met before"]
+
+
+def test_add_guest_with_existing_edge_and_reseed_runs_seed(client, tmp_path):
+    """Toggling ``re-seed anyway`` flips the gate off — the existing
+    flow runs (seed produces deltas, two ``edge_update`` events fire).
+    """
+    _seed_chat(tmp_path / "test.db")
+    _seed_host_to_guest_edge(tmp_path / "test.db")
+
+    canned = json.dumps(
+        {
+            "a_to_b_summary": "reconnected",
+            "a_to_b_knowledge_facts": ["new fact"],
+            "a_to_b_affinity_delta": 2,
+            "a_to_b_trust_delta": 1,
+            "b_to_a_summary": "reconnected",
+            "b_to_a_knowledge_facts": [],
+            "b_to_a_affinity_delta": 1,
+            "b_to_a_trust_delta": 0,
+        }
+    )
+    _override_llm([canned])
+    try:
+        response = client.post(
+            "/chats/chat_bot_a/drawer/guest/add",
+            data={
+                "guest_bot_id": "bot_b",
+                "relationship_prose": "fresh prose",
+                "reseed": "1",
+            },
+        )
+        assert response.status_code == 200
+    finally:
+        app.dependency_overrides.clear()
+
+    with open_db(tmp_path / "test.db") as conn:
+        edge_updates = conn.execute(
+            "SELECT payload_json FROM event_log WHERE kind = 'edge_update'"
+        ).fetchall()
+        # Pre-seed (1) + two from the re-seed = 3 edge_updates total.
+        ab_updates = [
+            json.loads(p[0])
+            for p in edge_updates
+            if {
+                json.loads(p[0]).get("source_id"),
+                json.loads(p[0]).get("target_id"),
+            }
+            == {"bot_a", "bot_b"}
+        ]
+        assert len(ab_updates) == 3
+
+
 def test_drawer_with_guest_renders_guest_and_group_sections(client, tmp_path):
     _seed_chat(tmp_path / "test.db")
     from chat.eventlog.log import append_and_apply

tests/test_regenerate.py

@@ -271,3 +271,394 @@ def test_regenerate_404_when_assistant_turn_missing(client, tmp_path):
         assert response.status_code == 404
     finally:
         app.dependency_overrides.clear()
+
+
+def _seed_with_interjection_group(db_path):
+    """Seed a multi-entity scene with a (primary + interjection) group.
+
+    Returns ``(user_turn_id, primary_at_id, interjection_at_id)``.
+
+    The primary speaker is the host (bot_a); the silent witness who
+    interjected is the guest (bot_b). Mirrors the convention in
+    chat/web/turns.py — both assistant_turns share the same
+    ``user_turn_id`` and the interjection's payload carries
+    ``interjection_of=<primary speaker_id>``.
+    """
+    with open_db(db_path) as conn:
+        for bot_id, name, persona in (
+            ("bot_a", "BotA", "thoughtful"),
+            ("bot_b", "BotB", "loud"),
+        ):
+            append_event(
+                conn,
+                kind="bot_authored",
+                payload={
+                    "id": bot_id,
+                    "name": name,
+                    "persona": persona,
+                    "voice_samples": [],
+                    "traits": [],
+                    "backstory": "",
+                    "initial_relationship_to_you": "",
+                    "kickoff_prose": "",
+                },
+            )
+        append_event(
+            conn,
+            kind="chat_created",
+            payload={
+                "id": "chat_multi",
+                "host_bot_id": "bot_a",
+                "guest_bot_id": "bot_b",
+                "initial_time": "2026-04-26T20:00:00+00:00",
+                "narrative_anchor": "Day 1",
+                "weather": "",
+            },
+        )
+        for src, tgt in (
+            ("bot_a", "you"),
+            ("you", "bot_a"),
+            ("bot_b", "you"),
+            ("you", "bot_b"),
+            ("bot_a", "bot_b"),
+            ("bot_b", "bot_a"),
+        ):
+            append_event(
+                conn,
+                kind="edge_update",
+                payload={
+                    "source_id": src,
+                    "target_id": tgt,
+                    "chat_id": "chat_multi",
+                },
+            )
+        for entity_id in ("you", "bot_a", "bot_b"):
+            append_event(
+                conn,
+                kind="activity_change",
+                payload={
+                    "entity_id": entity_id,
+                    "posture": "sitting",
+                    "action": {"verb": "talking"},
+                    "attention": "",
+                    "holding": [],
+                    "status": {},
+                },
+            )
+        ut_id = append_event(
+            conn,
+            kind="user_turn",
+            payload={
+                "chat_id": "chat_multi",
+                "prose": "hello",
+                "segments": [],
+            },
+        )
+        primary_id = append_event(
+            conn,
+            kind="assistant_turn",
+            payload={
+                "chat_id": "chat_multi",
+                "speaker_id": "bot_a",
+                "text": "Original primary.",
+                "truncated": False,
+                "user_turn_id": ut_id,
+            },
+        )
+        interjection_id = append_event(
+            conn,
+            kind="assistant_turn",
+            payload={
+                "chat_id": "chat_multi",
+                "speaker_id": "bot_b",
+                "text": "Original interjection!",
+                "truncated": False,
+                "user_turn_id": ut_id,
+                "interjection_of": "bot_a",
+            },
+        )
+        project(conn)
+        return ut_id, primary_id, interjection_id
+
+
+def test_regenerate_broadcasts_turn_html_over_sse(
+    tmp_path, monkeypatch
+):
+    """T73.1: regenerate publishes a ``turn_html_replace`` SSE event so
+    connected tabs swap the prior turn's DOM node in place.
+
+    The event carries:
+    - ``data``: rendered HTML for the new turn
+    - ``turn_id``: event_id of the new assistant_turn
+    - ``supersedes_id``: event_id of the original assistant_turn
+    """
+    import asyncio
+
+    from chat.config import Settings
+    from chat.db.migrate import apply_migrations
+    from chat.services import regenerate as regenerate_module
+    from chat.services.regenerate import regenerate_assistant_turn
+
+    db_path = tmp_path / "test.db"
+    cfg = tmp_path / "config.toml"
+    cfg.write_text('featherless_api_key = "test"\n')
+    monkeypatch.setenv("CHAT_CONFIG_PATH", str(cfg))
+    monkeypatch.setenv("CHAT_DB_PATH", str(db_path))
+    apply_migrations(db_path)
+
+    ut_id, at_id = _seed_with_one_turn(db_path)
+
+    published: list[tuple[str, dict]] = []
+
+    async def _capture(chat_id, event):
+        published.append((chat_id, event))
+
+    # Patch the imported reference inside the regenerate module so the
+    # service's call site goes through our spy.
+    monkeypatch.setattr(regenerate_module, "publish", _capture)
+
+    narrative_canned = "Refreshed reply."
+    state_canned = json.dumps(
+        {"affinity_delta": 0, "trust_delta": 0, "knowledge_facts": []}
+    )
+    canned = [narrative_canned, state_canned, state_canned]
+    mock_client = MockLLMClient(canned=list(canned))
+
+    settings = Settings(featherless_api_key="test")
+
+    with open_db(db_path) as conn:
+        new_text = asyncio.run(
+            regenerate_assistant_turn(
+                conn,
+                mock_client,
+                settings=settings,
+                chat_id="chat_bot_a",
+                original_assistant_event_id=at_id,
+            )
+        )
+        assert new_text == narrative_canned
+
+        # Find the new assistant_turn event_id for cross-checking.
+        cur = conn.execute(
+            "SELECT id FROM event_log "
+            "WHERE kind = 'assistant_turn' AND id != ? "
+            "AND superseded_by IS NULL",
+            (at_id,),
+        ).fetchone()
+        new_at_id = cur[0]
+
+    # Filter out per-token publishes; we want the replace broadcast.
+    replace_calls = [
+        ev for (_cid, ev) in published if ev.get("event") == "turn_html_replace"
+    ]
+    assert len(replace_calls) == 1
+    payload = replace_calls[0]
+    assert payload["supersedes_id"] == at_id
+    assert payload["turn_id"] == new_at_id
+    # The HTML carries the new narrative text and the speaker name.
+    assert "Refreshed reply." in payload["data"]
+    assert "BotA" in payload["data"]
+    # Sanity: every publish targeted this chat.
+    for cid, _ev in published:
+        assert cid == "chat_bot_a"
+
+
+def test_regenerate_with_interjection_redoes_both_turns(tmp_path, monkeypatch):
+    """T73.2: when the original turn group included an interjection, both
+    the primary and the interjection are regenerated.
+
+    Setup: 3-entity scene (host BotA + guest BotB + you) with a prior
+    (primary by BotA + interjection by BotB) group. Mock the
+    interjection classifier to return ``should_interject=True`` so the
+    follow-on regenerates too.
+
+    Assert: 2 new assistant_turns exist for the same user_turn_id, the
+    second carrying ``interjection_of`` pointing at the new primary's
+    speaker_id. Both originals are superseded.
+    """
+    import asyncio
+
+    from chat.config import Settings
+    from chat.db.migrate import apply_migrations
+    from chat.services import regenerate as regenerate_module
+    from chat.services.interjection import InterjectionDecision
+    from chat.services.regenerate import regenerate_assistant_turn
+
+    db_path = tmp_path / "test.db"
+    cfg = tmp_path / "config.toml"
+    cfg.write_text('featherless_api_key = "test"\n')
+    monkeypatch.setenv("CHAT_CONFIG_PATH", str(cfg))
+    monkeypatch.setenv("CHAT_DB_PATH", str(db_path))
+    apply_migrations(db_path)
+
+    ut_id, primary_id, interjection_id = _seed_with_interjection_group(db_path)
+
+    # Stub detect_interjection so the classifier "fires" with new prose.
+    async def _stub_should_interject(*_args, **_kwargs):
+        return InterjectionDecision(should_interject=True, reason="fired")
+
+    monkeypatch.setattr(
+        regenerate_module, "detect_interjection", _stub_should_interject
+    )
+
+    # Canned queue:
+    #   1. New primary narrative stream.
+    #   2-7. Six state-update classifier calls (one per directed pair
+    #        across host/you/guest = 6 pairs) for the primary pass.
+    #   8.   New interjection narrative stream.
+    #   9-14. Six state-update classifier calls for the post-interjection
+    #         pass.
+    state_canned = json.dumps(
+        {"affinity_delta": 0, "trust_delta": 0, "knowledge_facts": []}
+    )
+    canned: list[str] = []
+    canned.append("New primary text.")
+    canned.extend([state_canned] * 6)
+    canned.append("New interjection text!")
+    canned.extend([state_canned] * 6)
+    mock_client = MockLLMClient(canned=list(canned))
+
+    settings = Settings(featherless_api_key="test")
+
+    with open_db(db_path) as conn:
+        new_text = asyncio.run(
+            regenerate_assistant_turn(
+                conn,
+                mock_client,
+                settings=settings,
+                chat_id="chat_multi",
+                original_assistant_event_id=primary_id,
+            )
+        )
+        assert new_text == "New primary text."
+
+        # Both originals are superseded.
+        primary_super = conn.execute(
+            "SELECT superseded_by FROM event_log WHERE id = ?", (primary_id,)
+        ).fetchone()[0]
+        interjection_super = conn.execute(
+            "SELECT superseded_by FROM event_log WHERE id = ?",
+            (interjection_id,),
+        ).fetchone()[0]
+        assert primary_super is not None
+        assert interjection_super is not None
+
+        # Two NEW assistant_turn events exist (the regenerated primary
+        # and the regenerated interjection), both pinned to the same
+        # user_turn_id as the originals.
+        cur = conn.execute(
+            "SELECT id, payload_json FROM event_log "
+            "WHERE kind = 'assistant_turn' AND id NOT IN (?, ?) "
+            "ORDER BY id",
+            (primary_id, interjection_id),
+        ).fetchall()
+        assert len(cur) == 2
+        new_primary_id, new_primary_payload_json = cur[0]
+        new_interjection_id, new_interjection_payload_json = cur[1]
+        new_primary_payload = json.loads(new_primary_payload_json)
+        new_interjection_payload = json.loads(new_interjection_payload_json)
+
+        assert new_primary_payload["text"] == "New primary text."
+        assert new_primary_payload["speaker_id"] == "bot_a"
+        assert new_primary_payload["user_turn_id"] == ut_id
+        assert new_primary_payload["regenerated_from"] == primary_id
+        assert "interjection_of" not in new_primary_payload
+
+        assert new_interjection_payload["text"] == "New interjection text!"
+        assert new_interjection_payload["speaker_id"] == "bot_b"
+        assert new_interjection_payload["user_turn_id"] == ut_id
+        assert new_interjection_payload["regenerated_from"] == interjection_id
+        # interjection_of links to the new primary's speaker (matches
+        # the existing convention in chat/web/turns.py).
+        assert new_interjection_payload["interjection_of"] == "bot_a"
+
+        # The originals' supersede pointers reach the new ones.
+        assert primary_super == new_primary_id
+        assert interjection_super == new_interjection_id
+
+
+def test_regenerate_drops_interjection_when_classifier_returns_false(
+    tmp_path, monkeypatch
+):
+    """T73.2: when the original group included an interjection but the
+    classifier returns False this time, the new group is primary-only.
+
+    The original interjection is still superseded (we don't leave it
+    visible in the timeline alongside a regenerated primary it no longer
+    follows from), but no replacement assistant_turn is appended.
+    """
+    import asyncio
+
+    from chat.config import Settings
+    from chat.db.migrate import apply_migrations
+    from chat.services import regenerate as regenerate_module
+    from chat.services.interjection import InterjectionDecision
+    from chat.services.regenerate import regenerate_assistant_turn
+
+    db_path = tmp_path / "test.db"
+    cfg = tmp_path / "config.toml"
+    cfg.write_text('featherless_api_key = "test"\n')
+    monkeypatch.setenv("CHAT_CONFIG_PATH", str(cfg))
+    monkeypatch.setenv("CHAT_DB_PATH", str(db_path))
+    apply_migrations(db_path)
+
+    ut_id, primary_id, interjection_id = _seed_with_interjection_group(db_path)
+
+    async def _stub_no_interject(*_args, **_kwargs):
+        return InterjectionDecision(
+            should_interject=False, reason="quiet"
+        )
+
+    monkeypatch.setattr(
+        regenerate_module, "detect_interjection", _stub_no_interject
+    )
+
+    # Canned queue: primary narrative + 6 state-update calls. No
+    # interjection stream because the classifier short-circuits.
+    state_canned = json.dumps(
+        {"affinity_delta": 0, "trust_delta": 0, "knowledge_facts": []}
+    )
+    canned: list[str] = ["New primary text."] + [state_canned] * 6
+    mock_client = MockLLMClient(canned=list(canned))
+
+    settings = Settings(featherless_api_key="test")
+
+    with open_db(db_path) as conn:
+        new_text = asyncio.run(
+            regenerate_assistant_turn(
+                conn,
+                mock_client,
+                settings=settings,
+                chat_id="chat_multi",
+                original_assistant_event_id=primary_id,
+            )
+        )
+        assert new_text == "New primary text."
+
+        # Original primary superseded by the new primary.
+        primary_super = conn.execute(
+            "SELECT superseded_by FROM event_log WHERE id = ?", (primary_id,)
+        ).fetchone()[0]
+        # Original interjection ALSO superseded — we don't leave a
+        # dangling beat attached to a regenerated primary that no longer
+        # warrants a follow-on. Back-pointer goes to the new primary.
+        interjection_super = conn.execute(
+            "SELECT superseded_by FROM event_log WHERE id = ?",
+            (interjection_id,),
+        ).fetchone()[0]
+        assert primary_super is not None
+        assert interjection_super is not None
+        assert interjection_super == primary_super  # both point at new primary
+
+        # Exactly ONE new assistant_turn — the primary; no replacement
+        # interjection.
+        cur = conn.execute(
+            "SELECT payload_json FROM event_log "
+            "WHERE kind = 'assistant_turn' AND id NOT IN (?, ?) "
+            "AND superseded_by IS NULL",
+            (primary_id, interjection_id),
+        ).fetchall()
+        assert len(cur) == 1
+        new_primary_payload = json.loads(cur[0][0])
+        assert new_primary_payload["text"] == "New primary text."
+        assert "interjection_of" not in new_primary_payload

tests/test_turn_flow.py

@@ -405,14 +405,15 @@ def test_multi_bot_turn_no_interjection(app_state_setup, tmp_path):
     1 user_turn + 1 assistant_turn + 6 *post-turn* edge_updates + 2
     memory_written events. Single turn_html broadcast.
 
-    Canned queue (8 calls):
+    Canned queue (11 calls):
       1. parse_turn
-      2. narrative stream (primary, addressee = host because the prose
+      2. detect_addressee (T74.1) -> host
+      3. narrative stream (primary, addressee = host because the prose
          doesn't name the guest)
-      3-8. 6 state-update calls (one per directed pair across {you,
+      4-9. 6 state-update calls (one per directed pair across {you,
          bot_a, bot_b})
-      9. detect_interjection -> should_interject=False
-      10. detect_scene_close -> should_close=False
+      10. detect_interjection -> should_interject=False
+      11. detect_scene_close -> should_close=False
     """
     _seed_chat_with_guest(tmp_path / "test.db")
     canned_parse = json.dumps(
@@ -420,6 +421,9 @@ def test_multi_bot_turn_no_interjection(app_state_setup, tmp_path):
     )
     canned = [
         canned_parse,
+        json.dumps(
+            {"addressee_id": "bot_a", "confidence": "medium", "reason": "host"}
+        ),
         "Greetings.",
         _zero_state(), _zero_state(), _zero_state(),
         _zero_state(), _zero_state(), _zero_state(),
@@ -474,14 +478,15 @@ def test_multi_bot_turn_with_interjection(app_state_setup, tmp_path):
     1 user_turn + 2 assistant_turns + (6 + 6) post-turn edge_updates +
     4 memory_written events.
 
-    Canned queue (16 calls):
+    Canned queue (17 calls):
       1. parse_turn
-      2. narrative stream (primary)
-      3-8. 6 state-update calls (post-primary)
-      9. detect_interjection -> should_interject=True
-      10. narrative stream (interjection)
-      11-16. 6 state-update calls (post-interjection)
-      17. detect_scene_close -> should_close=False
+      2. detect_addressee (T74.1) -> host
+      3. narrative stream (primary)
+      4-9. 6 state-update calls (post-primary)
+      10. detect_interjection -> should_interject=True
+      11. narrative stream (interjection)
+      12-17. 6 state-update calls (post-interjection)
+      18. detect_scene_close -> should_close=False
     """
     _seed_chat_with_guest(tmp_path / "test.db")
     canned_parse = json.dumps(
@@ -489,6 +494,9 @@ def test_multi_bot_turn_with_interjection(app_state_setup, tmp_path):
     )
     canned = [
         canned_parse,
+        json.dumps(
+            {"addressee_id": "bot_a", "confidence": "medium", "reason": "host"}
+        ),
         "Primary beat.",
         _zero_state(), _zero_state(), _zero_state(),
         _zero_state(), _zero_state(), _zero_state(),
@@ -555,14 +563,15 @@ def test_multi_bot_turn_scene_close_writes_per_pov_summaries(
     rewrites fire for both bots (memory.pov_summary changes for each).
     Interjection short-circuits at False so the queue stays compact.
 
-    Canned queue (12 calls):
+    Canned queue (13 calls):
       1. parse_turn
-      2. narrative stream (primary)
-      3-8. 6 state-update calls
-      9. detect_interjection -> False (no follow-on stream)
-      10. detect_scene_close -> True
-      11. apply_scene_close_summary host POV
-      12. apply_scene_close_summary guest POV
+      2. detect_addressee (T74.1) -> host
+      3. narrative stream (primary)
+      4-9. 6 state-update calls
+      10. detect_interjection -> False (no follow-on stream)
+      11. detect_scene_close -> True
+      12. apply_scene_close_summary host POV
+      13. apply_scene_close_summary guest POV
     """
     _seed_chat_with_guest(tmp_path / "test.db")
     canned_parse = json.dumps(
@@ -588,6 +597,9 @@ def test_multi_bot_turn_scene_close_writes_per_pov_summaries(
     )
     canned = [
         canned_parse,
+        json.dumps(
+            {"addressee_id": "bot_a", "confidence": "medium", "reason": "host"}
+        ),
         "Goodnight.",
         _zero_state(), _zero_state(), _zero_state(),
         _zero_state(), _zero_state(), _zero_state(),
@@ -639,12 +651,20 @@ def test_multi_bot_turn_scene_close_writes_per_pov_summaries(
 
 
 def test_addressee_detection_routes_to_named_bot(app_state_setup, tmp_path):
-    """Prose that names the guest by name routes the primary turn to the
-    guest. Interjection (when fired) makes the host the silent witness
-    and the second assistant_turn carries the host as speaker.
+    """T74.1: the multi-entity addressee call goes through the classifier;
+    when the classifier returns the guest, the primary turn routes there.
+    Interjection (when fired) makes the host the silent witness and the
+    second assistant_turn carries the host as speaker.
 
-    Canned queue: same shape as the with-interjection test (16 calls)
-    plus the trailing scene_close decision.
+    Canned queue (with classifier-led addressee = guest):
+      1. parse_turn
+      2. detect_addressee -> bot_b (the guest)
+      3. narrative stream (primary, addressee = guest)
+      4-9. 6 state-update calls
+      10. detect_interjection -> True
+      11. interjection narrative stream
+      12-17. 6 state-update calls (post-interjection)
+      18. detect_scene_close -> False
     """
     _seed_chat_with_guest(tmp_path / "test.db")
     canned_parse = json.dumps(
@@ -652,6 +672,13 @@ def test_addressee_detection_routes_to_named_bot(app_state_setup, tmp_path):
     )
     canned = [
         canned_parse,
+        json.dumps(
+            {
+                "addressee_id": "bot_b",
+                "confidence": "high",
+                "reason": "user named BotB",
+            }
+        ),
         "BotB pondering.",
         _zero_state(), _zero_state(), _zero_state(),
         _zero_state(), _zero_state(), _zero_state(),
@@ -680,9 +707,192 @@ def test_addressee_detection_routes_to_named_bot(app_state_setup, tmp_path):
         primary_payload = json.loads(rows[0][0])
         interjection_payload = json.loads(rows[1][0])
 
-    # Primary speaker is the guest because the prose names BotB and not
-    # BotA (case-insensitive whole-word match).
+    # Primary speaker is the guest because the addressee classifier
+    # picked bot_b for the prose ("BotB, what do you think?").
     assert primary_payload["speaker_id"] == "bot_b"
     # Interjection follow-on goes to the silent witness — the host.
     assert interjection_payload["speaker_id"] == "bot_a"
     assert interjection_payload["interjection_of"] == "bot_b"
+
+
+def test_cancelled_turn_still_closes_scene_when_user_prose_signals_close(
+    app_state_setup, tmp_path
+):
+    """T74.3 regression: a cancelled primary stream still triggers scene
+    close when the user prose carries a hard close signal.
+
+    Rationale (also documented in turns.py near the close-detection
+    branch): close detection only consumes the user's prose, which is
+    fully appended to the event_log BEFORE streaming starts. The
+    cancelled bot beat doesn't invalidate the user's intent to close.
+
+    Implementation: install a MockLLMClient whose ``stream`` raises
+    CancelledError on the first iteration. The classifier calls (parse,
+    addressee, scene_close, per-POV summaries) are still served from
+    the canned queue. The post_turn route ultimately re-raises
+    CancelledError after recording the partial — TestClient surfaces
+    that as an exception, so we drive the request inside ``with
+    pytest.raises``. Despite the exception, the scene_closed event
+    must land in the event_log.
+    """
+    from typing import AsyncIterator, Sequence
+
+    _seed_chat_with_guest(tmp_path / "test.db")
+    canned_parse = json.dumps(
+        {"segments": [{"kind": "narration", "text": "we are done here, fade out"}]}
+    )
+    pov_payload = json.dumps(
+        {
+            "summary": "BotA noticed the day winding down.",
+            "knowledge_facts": [],
+            "relationship_summary": "warmer",
+        }
+    )
+    pov_payload_guest = json.dumps(
+        {
+            "summary": "BotB watched the scene close.",
+            "knowledge_facts": [],
+            "relationship_summary": "warmer",
+        }
+    )
+    # Canned queue: parse + addressee + 6 state-updates +
+    # scene_close=True + 2 per-POV summaries. NO interjection slot
+    # because the cancel path short-circuits the interjection branch.
+    canned = [
+        canned_parse,
+        json.dumps(
+            {"addressee_id": "bot_a", "confidence": "medium", "reason": "host"}
+        ),
+        # NOTE: no narrative slot — the stream is hijacked below to
+        # raise CancelledError on first iteration; it never pulls a
+        # canned response.
+        _zero_state(), _zero_state(), _zero_state(),
+        _zero_state(), _zero_state(), _zero_state(),
+        json.dumps({"should_close": True, "reason": "fade out signaled"}),
+        pov_payload,
+        pov_payload_guest,
+    ]
+
+    class _CancelOnStreamMock:
+        """Mock LLM client that serves ``generate`` from a canned queue
+        and raises CancelledError on the FIRST iteration of ``stream``.
+
+        Mirrors :class:`chat.llm.mock.MockLLMClient` for ``generate`` but
+        diverges on ``stream`` to simulate a mid-stream cancel.
+        """
+
+        def __init__(self, canned: list[str]) -> None:
+            self._canned = list(canned)
+
+        async def generate(
+            self, messages: Sequence, *, model: str, **params
+        ) -> str:
+            return self._canned.pop(0)
+
+        async def stream(
+            self, messages: Sequence, *, model: str, **params
+        ) -> AsyncIterator[str]:
+            # Yield a CancelledError on first iteration to simulate the
+            # /turns/cancel route firing mid-stream.
+            raise asyncio.CancelledError
+            yield  # pragma: no cover — keeps this an async generator.
+
+    from chat.web.kickoff import get_llm_client
+
+    mock = _CancelOnStreamMock(canned=list(canned))
+    app.dependency_overrides[get_llm_client] = lambda: mock
+
+    try:
+        # FastAPI/Starlette handles the re-raised CancelledError as an
+        # internal failure — TestClient surfaces it as a 500 response.
+        # We don't assert on the status here; the regression is whether
+        # the scene_closed event still landed in the event_log.
+        try:
+            app_state_setup.post(
+                "/chats/chat_bot_a/turns",
+                data={"prose": "we are done here, fade out"},
+            )
+        except BaseException:
+            # Some Starlette/asyncio versions propagate the
+            # CancelledError out of the test client; that's fine — the
+            # partial-record + scene-close still ran before the raise.
+            pass
+    finally:
+        app.dependency_overrides.clear()
+
+    with open_db(tmp_path / "test.db") as conn:
+        scene_close_count = conn.execute(
+            "SELECT COUNT(*) FROM event_log WHERE kind = 'scene_closed'"
+        ).fetchone()[0]
+        assistant_payload = conn.execute(
+            "SELECT payload_json FROM event_log "
+            "WHERE kind = 'assistant_turn' ORDER BY id"
+        ).fetchall()
+
+    # Scene close lands despite the cancel.
+    assert scene_close_count == 1
+    # The cancelled assistant_turn was still recorded (truncated=True).
+    assert len(assistant_payload) == 1
+    assert json.loads(assistant_payload[0][0])["truncated"] is True
+
+
+def test_interjection_enqueues_significance_job(app_state_setup, tmp_path):
+    """T74.2: when an interjection fires, the interjection memory is
+    enqueued for significance scoring just like the primary memory.
+
+    Capture enqueued ``SignificanceJob``s by replacing the background
+    worker's ``enqueue`` method with a list-append. Without T74.2, the
+    interjection memory would never be scored — only the primary's
+    enqueue would land. We therefore expect TWO jobs after a turn that
+    has both a primary and an interjection beat: one for the primary
+    memory, one for the interjection memory.
+    """
+    _seed_chat_with_guest(tmp_path / "test.db")
+    canned_parse = json.dumps(
+        {"segments": [{"kind": "dialogue", "text": "tell me"}]}
+    )
+    canned = [
+        canned_parse,
+        json.dumps(
+            {"addressee_id": "bot_a", "confidence": "medium", "reason": "host"}
+        ),
+        "Primary beat.",
+        _zero_state(), _zero_state(), _zero_state(),
+        _zero_state(), _zero_state(), _zero_state(),
+        json.dumps({"should_interject": True, "reason": "jealous"}),
+        "Interjection beat!",
+        _zero_state(), _zero_state(), _zero_state(),
+        _zero_state(), _zero_state(), _zero_state(),
+        json.dumps({"should_close": False, "reason": "no signal"}),
+    ]
+    _override_llm(canned)
+
+    captured_jobs: list = []
+    worker = app.state.background_worker
+    # Re-enable enqueue capture even though the worker's loop is disabled
+    # — we want to count enqueues without the loop running classifier work.
+    worker.enabled = True
+    original_enqueue = worker.enqueue
+    worker.enqueue = captured_jobs.append  # type: ignore[assignment]
+
+    try:
+        response = app_state_setup.post(
+            "/chats/chat_bot_a/turns", data={"prose": "tell me"}
+        )
+        assert response.status_code == 204
+    finally:
+        worker.enqueue = original_enqueue  # type: ignore[assignment]
+        worker.enabled = False
+        app.dependency_overrides.clear()
+
+    # Expect 2 enqueues: 1 for the primary memory + 1 for the
+    # interjection memory.
+    assert len(captured_jobs) == 2
+
+    # Both jobs should reference distinct memory ids — the primary's
+    # host-POV memory and the interjection's host-POV memory.
+    memory_ids = [job.memory_id for job in captured_jobs]
+    assert len(set(memory_ids)) == 2
+    # The two narrative texts should be the two streamed beats.
+    narrative_texts = sorted(job.narrative_text for job in captured_jobs)
+    assert narrative_texts == ["Interjection beat!", "Primary beat."]