v1.1: harden context window and narrator protocol boundary

WP-A1 and WP-A2, implemented in sequence, plus the corrective work the owner
asked for at review. Reported in
planning/reports/v1.1/V1.1-WP-A1-A2-REPORT.md (corrective addendum §R).
Planning package v4.2.

WP-A1: context-window safety reserve
- The prompt leaves max(256, ceil(5% of the effective window)) tokens free
  beside the reply. That is 256 at 4,096 and 820 at 16,384. The value is fixed,
  not a setting, and not calibrated per model.
- M6's 64-token margin is gone. Separators and the chat hint are priced
  exactly; tokenizer drift is the reserve's job.
- Protected context that cannot fit raises ContextOverflow before the model
  is called.
- Streams set stream_options.include_usage. Measured on Ollama 0.33, a stream
  sent no usage without it.
- Each sent turn records fits, exceeded, truncation_suspected or unknown.
  The status is returned on the done event, logged when bad, and shown in the
  context inspector. The turn is always kept.
- Accounting is per-attempt data (attempts.ATTEMPT_KEYS).
- Corrective: a cold model is loaded before its turn is built. When the
  window is unverified but the server answered, contextwindow.ensure_window
  makes one bounded POST /api/generate naming only the model. It sends no
  prompt, generates nothing and writes nothing. It then probes again, and the
  turn is built to that answer. If the load fails, or the window is still
  unknown, the turn falls back to the old behaviour.
- Real host, 4,096 window:
  - v1 cold turn: sent 13,875, the server read 2,050.
  - Same turn after the correction: the window was verified, 3,082 sent,
    3,097 read, fits, 499 tokens left beside the reply.
  - Verified turns elsewhere left 275-2,297 tokens against v1's 23-42.

WP-A2: protocol echo and genre-neutral state prompting
- The vocabulary is shown as the JSON object the model sends, not as
  name(field, ...). This costs 121 tokens.
- The example uses character-1, item-1 and location-1.
- The extractor removes shapes anchored to application-owned text:
  - a vocabulary call line;
  - an echoed length hint;
  - the renderer's scene line left last;
  - an empty fence opener.
- Corrective R5: the echoed continue hint is recognised by its own sentence
  ("Output only story text"). A Hard-limit-opened bracket is removed only
  directly above an echo already cut from the same reply.
- Replay of all 518 real v1 replies: 9 changed, 0 flagged, and no story prose
  removed. That is unchanged by R5.
- Replay of 64 v1.1 replies: 3 changed, 0 flagged. The depth-16 instruction
  tail is removed.
- Identity diagnostic after the correction:
  - 0 identity signals;
  - 0 prompt example identifiers proposed;
  - 0/10 stored turns with protocol or instruction shapes.
- 50-turn run: 51 accepted, 0 of 54 stored turns carry protocol.
- SPECS, render.py and validate.py are identical to v1.0.0.

Compatibility: a real v1.0.0 database reads identically on v1.0.0 and v1.1,
field for field, with schema and user_version 94 unchanged. Undo, redo, Save
Point restore, export and import all work on it. There is no schema,
migration or bundle-format change.

Verification: the backend suite passes 1,534 with 17 skipped and 0 failed.
The frontend passes 165/165, and lint and the build are clean. The offline
container and the browser regression were re-run on this tree (see §R.3).

One test was re-calibrated, not weakened: test_history_block_trim's prefix
test had assumed which turn holds the floor at a 2,048 budget.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VvegagkhuCZoFPdv4M1egY

backend/app/routers/adventures/turns.py

@@ -6,6 +6,7 @@ lock guards one set only while one module owns it. And a test that replaces
 `OpenAICompatibleProvider` or `generate_turn` patches this module, which every
 caller reads through.
 """
+import logging
 import threading
 
 from fastapi import Depends, HTTPException, Request
@@ -28,6 +29,8 @@ from .deps import CurrentUser, current_adventure, router
 from .nodes import _move_to_after, next_depth
 from .paging import annotate_takes
 
+log = logging.getLogger(__name__)
+
 
 def world_delta_of(snapshot: dict | None) -> dict | None:
     """Returns the bulk-read slice of a context snapshot, for `Action.world_delta`.
@@ -212,8 +215,18 @@ async def _generate_turn(
     # network calls — and cached per endpoint and model, so it costs one short
     # request per session rather than one per turn. An unverified window does
     # not block the turn; it is recorded as unverified in the snapshot below.
-    window = await contextwindow.probe(settings.endpoint_url, settings.model,
-                                          declared=settings.context_window_override)
+    #
+    # v1.1 WP-A1 corrective: a model that is not resident cannot report its window,
+    # and a turn built to the configured budget against it was silently cut in the
+    # A1 evidence (13,875 tokens sent, 2,050 read). So an unverified window gets
+    # one bounded attempt to load the model, and one more probe, before the
+    # prompt is assembled. No story text is generated by it and nothing is
+    # written. A window still unverified afterwards changes nothing below.
+    window, preflight = await contextwindow.ensure_window(
+        settings.endpoint_url, settings.model,
+        declared=settings.context_window_override,
+        warm_timeout=float(settings.model_timeout_seconds or 300),
+    )
     try:
         system_text, story_text, snapshot = build_context(
             adventure,
@@ -232,6 +245,9 @@ async def _generate_turn(
         yield turn_error(str(exc))
         return
 
+    if isinstance(snapshot.get("window"), dict):
+        snapshot["window"]["preflight"] = preflight
+
     parts = PromptParts(system=system_text, story=story_text)
 
     provider = OpenAICompatibleProvider(
@@ -318,6 +334,24 @@ async def _generate_turn(
     # prompt came from cache rather than being billed in full. This is recorded
     # per attempt, next to the prompt it priced.
     snapshot["usage"] = provider.last_usage
+    # v1.1 WP-A1: what the server says it read, against what was sent. Recorded
+    # and shown, never acted on: the narration has already streamed to the
+    # reader, and discarding an accepted turn over an accounting discrepancy
+    # would lose story to hide a problem. A server that cut the prompt answers
+    # 200 either way, so this record is the only place the cut is visible.
+    tokens = snapshot.get("tokens") or {}
+    accounting = contextwindow.classify_usage(
+        provider.last_usage,
+        estimate=tokens.get("estimate") or tokens.get("total") or 0,
+        budget=tokens.get("budget") or settings.context_token_budget,
+        max_output_tokens=settings.max_output_tokens,
+        window_verified=bool((snapshot.get("window") or {}).get("verified")),
+    )
+    snapshot["accounting"] = accounting
+    if accounting["status"] in (contextwindow.EXCEEDED,
+                                contextwindow.TRUNCATION_SUSPECTED):
+        log.warning("turn accounting for adventure %s: %s — %s",
+                    adventure.id, accounting["status"], accounting["detail"])
 
     reasoning = "".join(reasoning_chunks).strip() or None
     ai_action = models.Action(
@@ -381,7 +415,8 @@ async def _generate_turn(
     db.commit()
     db.refresh(ai_action)
     yield _SAVED
-    yield sse({"type": "done", "action": action_json(ai_action, db)})
+    yield sse({"type": "done", "action": action_json(ai_action, db),
+               "accounting": accounting})
     # Phase 6: schedule summarization and embedding without waiting for them.
     # The task opens its own database session.
     memorybank.schedule_post_turn(adventure)