Files
interactive-story/backend/tools/m11_long_run.py
T
JesseMarkowitzandClaude Opus 5 144406cd48 M11: what the server will actually read
The release-validation milestone, and the thing it had to settle first was
whether any of the earlier evidence meant what it said. M8 measured a deployment
enforcing a 4,096-token input window while the application budgeted 16,384.
Every request returned 200. What Ollama does with the excess is drop the oldest
tokens, and the oldest tokens here are the system block — the narrator's rules
and the campaign canon. A hundred-turn certification against that server would
have looked perfect and proved nothing, which is why this milestone could not
begin with a hundred turns.

So the application asks now. Ollama's window is a property of how a model was
loaded rather than of the request — sending num_ctx is accepted, ignored, and
worse, reloads the model at the server's own default — so the only honest move
is to find out and then tell the truth about it. /api/ps reports what a resident
model is being served with, /api/show what an unloaded one will load with, both
on the same host inference already uses, through the same endpoint policy and
the same TLS trust store. A verified window is a ceiling on the budget; an
unverified one leaves the budget alone and is recorded as unverified in the
turn's own provenance, so an old turn can be asked afterwards whether it was
built against a checked window. There is no third behaviour, and in particular
no hard-coded 4,096: a number the server did not say would be right on one
machine and wrong on the next.

The proof that this is doing something is a campaign whose canon sits at the
front of the prompt, 120 turns of history, and a 4,096-token window. The canon
is still there afterwards and the oldest history is gone. The same campaign
built the old way produces a prompt more than twice the window — the defect,
reproduced, so the fix is measured against it rather than asserted.

Two defects the validation found on its own, and they are the same defect twice:
something was true and nobody was told. A manual state correction of four
changes with one bad reference applied three, returned 201, and said nothing —
while recording the refusal on the audit row nobody reads. It came to light
because the identity diagnostic's own fixture was refused that way and the whole
run proceeded on a campaign with no scene, which would have read as a model
failure. And the narration-length setting moved no number: brief, medium and
long each became one English sentence, while the numeric hint the model actually
reads was derived from the global reply cap and said the same thing for all
three. Both now say what they did.

The other two post-M8 findings are closed as well. The tab said AI D&D, which no
document had ever claimed it did not; it says Interactive Story now, with the
open campaign first, and the name is the owner's decision rather than a
find-and-replace to something narrower than the engine. After an Undo the reader
could not tell where they had landed; the control row now ends with
"Moment 11 · later story ahead", from the server's own answer, in the word the
transcript already uses, with none of head, branch or depth anywhere near it.

The identity diagnostic exists and the root cause does not. That campaign was
destroyed, so no cause can be established — what M11 owes the finding is
something that can classify the next occurrence, and a diagnostic that makes only
the judgements a program can honestly make: duplicate keys, shared names,
protagonist drift, state and context disagreeing. Whether prose misattributed a
line is left to a person reading it beside its prompt, because a regex cannot
read dialogue and one that pretended to would produce exactly the confident wrong
answer this finding is about. Its detectors are proved to fire against a planted
second Alice.

Two entities may still share a display name. That was checked first, as the
finding asked, and left permitted: a mother and a daughter, or a stranger giving
a false name, are ordinary fiction, and refusing them to guard against a model
mistake would refuse the wrong thing. What was missing was that it happened
silently. It is reported now.

Evidence, not inference: a hundred accepted turns against a real narrator with
genuine process restarts; a real browser against the built SPA; a container with
no network at all; a campaign moved into a data directory that never existed.
Each was discarded and re-run whenever the product changed under it, and the runs
that were thrown away are listed in the report with the reason, along with ten
defects in the harnesses themselves — because a harness that has only ever
agreed with itself is not evidence, and two of M8's five harness defects were
masking real ones.

No dependency was added, removed or upgraded. No acceptance test was retired,
relaxed or reclassified. M11 is implemented and verified; it is not accepted, and
there is no release tag.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qyn3oRd4D6pi72nKBG725B
2026-09-07 14:01:20 -04:00

637 lines
27 KiB
Python

"""M11 M01-M04: a real 100-turn campaign, against a real narrator, over HTTP.
python -m tools.m11_long_run --turns 100 --out <dir>
Run from `backend/`. Reads `AIDND_TEST_ENDPOINT`, `AIDND_TEST_MODEL` and
optionally `AIDND_TEST_EMBED_MODEL`.
## Why this is a script that spawns servers rather than a test
M01's pass condition is not "100 requests succeeded". It is **100+ accepted turns
with no continuity, state, history, authority, lineage or recovery corruption**,
across genuine application restarts, with a fact planted at the beginning
recoverable at the end through memory rather than through the transcript.
Three of those words decide the shape of this harness:
*Accepted* — a turn counts when the application committed it, so every turn is
checked for a committed action and a state document, not for an HTTP 200.
*Restarts* — M02 says a new test client, a reconnected browser and a reopened
session do **not** count. So the storyteller runs as a real `uvicorn` process,
started with the command `DEVELOPMENT.md` documents, and is killed and restarted
at planned points. Everything that survives crosses as bytes on disk.
*Recoverable* — the planted clue has to be pushed out of the recent-history
window and then retrieved, so the run measures the window at intervals and the
recall check at the end asks the application what it would actually send.
## What it records
A JSON line per turn (`timeline.jsonl`) with the context measurements M03 wants,
a `measurements.json` of the sampled checkpoints, the recall evidence for M04,
and the final bundle. Everything is written as it happens, so a run that dies at
turn 80 still leaves 80 turns of evidence rather than nothing.
"""
from __future__ import annotations
import argparse
import json
import os
import socket
import subprocess
import sys
import time
import urllib.error
import urllib.request
from datetime import datetime
from pathlib import Path
HERE = Path(__file__).resolve().parent
BACKEND = HERE.parent
ENDPOINT = os.environ.get("AIDND_TEST_ENDPOINT", "")
MODEL = os.environ.get("AIDND_TEST_MODEL", "")
EMBED_MODEL = os.environ.get("AIDND_TEST_EMBED_MODEL", "")
#: The planted clue. Distinctive enough that its presence anywhere is
#: unambiguous, and phrased as something a story would actually establish.
CLUE = "the silver key opens the crypt beneath the Old Abbey"
CLUE_SENTINEL = "SILVER-KEY-CRYPT-OLD-ABBEY"
CANON = [
"The dead do not return. No rite, relic or bargain has ever returned anyone.",
"The abbey crypt has been sealed since the founding.",
"Aldric is the protagonist and the one the reader plays.",
]
CANON_MD = """# Westhaven
## The Old Abbey
The abbey above Westhaven has stood since the founding. Its crypt is sealed.
## What cannot happen here
The dead do not return. No rite, relic or bargain in Westhaven has ever
returned anyone from death, and none ever will.
"""
REFERENCE_MD = """# Roads and weather of the Fen
The fen road floods between the autumn rains and the first hard frost. Traders
take the ridge track instead, which adds a day.
"""
INSPIRATION_MD = """# Tone notes
Rain on slate. Lamplight through smoke. People who say less than they mean.
"""
#: The beats the campaign plays through, cycled. Written so the story keeps
#: moving and keeps giving the state extractor something to do, rather than a
#: hundred repetitions of one sentence.
BEATS = [
"I ask Mara what she has heard about the abbey.",
"I walk down to the waterfront and watch the boats.",
"I ask the ferryman about the fen road.",
"I look through my pack for anything useful.",
"I go back to the tavern and sit by the fire.",
"I ask Mara whether Edrin has been seen.",
"I take the ridge track north out of town.",
"I stop at the shrine on the ridge and look back at Westhaven.",
"I talk to the trader waiting out the rain.",
"I check the sky and decide whether to press on.",
]
def free_port() -> int:
with socket.socket() as s:
s.bind(("127.0.0.1", 0))
return s.getsockname()[1]
class Storyteller:
"""The real application, started the way `DEVELOPMENT.md` says to start it."""
def __init__(self, db_path: Path, log: Path):
self.db_path = db_path
self.port = free_port()
self.log_path = log
self.proc = None
self.starts = 0
def start(self) -> None:
self.starts += 1
handle = open(self.log_path, "ab")
self.proc = subprocess.Popen(
[str(BACKEND / ".venv/bin/uvicorn"), "app.main:app",
"--host", "127.0.0.1", "--port", str(self.port)],
cwd=str(BACKEND), stdout=handle, stderr=subprocess.STDOUT,
env={**os.environ, "AIDND_DB_PATH": str(self.db_path),
"AIDND_DATABASE_URL": "", "DATABASE_URL": ""},
)
deadline = time.monotonic() + 90
while time.monotonic() < deadline:
if self.proc.poll() is not None:
raise SystemExit(f"server exited early; see {self.log_path}")
try:
self.call("GET", "/settings")
return
except (urllib.error.URLError, ConnectionError, OSError):
time.sleep(0.1)
raise SystemExit(f"server never became ready; see {self.log_path}")
def stop(self) -> None:
if self.proc and self.proc.poll() is None:
self.proc.terminate()
try:
self.proc.wait(timeout=20)
except subprocess.TimeoutExpired:
self.proc.kill()
self.proc.wait(timeout=20)
def listening(self) -> bool:
try:
self.call("GET", "/settings")
return True
except Exception:
return False
def restart(self) -> None:
"""A genuine OS process boundary, proved gone before it is replaced."""
self.stop()
assert not self.listening(), "the old process is still answering"
self.port = free_port()
self.start()
def call(self, method: str, path: str, payload=None, timeout=600):
data = json.dumps(payload).encode() if payload is not None else None
request = urllib.request.Request(
f"http://127.0.0.1:{self.port}/api{path}", data=data, method=method,
headers={"Content-Type": "application/json"} if data else {},
)
with urllib.request.urlopen(request, timeout=timeout) as response:
body = response.read().decode()
return json.loads(body) if body else None
def stream(self, path: str, payload, timeout=900) -> list[dict]:
"""A turn. The reply is SSE, and a failed turn is an event, not a status.
`app/sse.py`: "A failed turn is still an HTTP 200 response, because the
error is reported inside the stream the client is already reading." A
harness that read the status code would call every failure a success —
which is precisely the class of harness defect M8's review warned about.
"""
request = urllib.request.Request(
f"http://127.0.0.1:{self.port}/api{path}",
data=json.dumps(payload).encode(), method="POST",
headers={"Content-Type": "application/json"},
)
events: list[dict] = []
with urllib.request.urlopen(request, timeout=timeout) as response:
for raw in response:
line = raw.decode(errors="replace").strip()
if line.startswith("data:"):
try:
events.append(json.loads(line[5:].strip()))
except json.JSONDecodeError:
pass
return events
class Run:
"""One long campaign, and everything measured about it."""
def __init__(self, server: Storyteller, out: Path):
self.server = server
self.out = out
self.timeline = (out / "timeline.jsonl").open("a")
self.adv = 0
self.accepted = 0
self.events: list[dict] = []
# ------------------------------------------------------------ recording
def note(self, kind: str, **fields) -> None:
entry = {"at": datetime.now().isoformat(timespec="seconds"),
"kind": kind, "accepted_turns": self.accepted, **fields}
self.events.append(entry)
self.timeline.write(json.dumps(entry, default=str) + "\n")
self.timeline.flush()
# ------------------------------------------------------------- campaign
def setup(self) -> None:
settings = self.server.call("PUT", "/settings", {
"endpoint_url": ENDPOINT, "model": MODEL,
"embedding_model": EMBED_MODEL, "context_token_budget": 16384,
"max_output_tokens": 500, "model_timeout_seconds": 600,
"memory_top_k": 4,
})
self.note("settings", model=settings["model"],
budget=settings["context_token_budget"])
created = self.server.call("POST", "/adventures", {
"title": "Continuity Test (M11 long run)",
"opening": (
"Rain over Westhaven. Aldric sits in the Crooked Lantern with a "
"silver key in his pocket and no-one to give it to."
),
"canon_rules": CANON,
"persona_name": "Aldric",
"narration_length": "brief",
})
self.adv = created["id"]
self.note("campaign", id=self.adv)
for name, body, kind in (("canon.md", CANON_MD, "canon"),
("reference.md", REFERENCE_MD, "reference"),
("inspiration.md", INSPIRATION_MD, "inspiration")):
self.upload(name, body, kind)
# The cast and the opening scene, as accepted state rather than prose.
self.correct([
{"type": "create_entity", "entity": "aldric",
"entity_type": "character", "name": "Aldric"},
{"type": "create_entity", "entity": "mara",
"entity_type": "character", "name": "Mara"},
{"type": "create_entity", "entity": "edrin",
"entity_type": "character", "name": "Edrin"},
{"type": "create_entity", "entity": "tavern",
"entity_type": "location", "name": "The Crooked Lantern"},
{"type": "create_entity", "entity": "abbey",
"entity_type": "location", "name": "The Old Abbey"},
{"type": "create_entity", "entity": "silver_key",
"entity_type": "item", "name": "the silver key"},
{"type": "set_possession", "item": "silver_key", "owner": "aldric"},
{"type": "set_scene",
"summary": "Aldric and Mara in the Crooked Lantern, rain outside.",
"location": "tavern", "present": ["aldric", "mara"]},
], note="the opening cast")
def upload(self, name: str, body: str, classification: str) -> None:
"""Multipart by hand: the harness speaks HTTP, not the test client."""
boundary = "----m11longrun"
parts = (
f"--{boundary}\r\nContent-Disposition: form-data; name=\"classification\""
f"\r\n\r\n{classification}\r\n"
f"--{boundary}\r\nContent-Disposition: form-data; name=\"file\"; "
f"filename=\"{name}\"\r\nContent-Type: text/markdown\r\n\r\n{body}\r\n"
f"--{boundary}--\r\n"
).encode()
request = urllib.request.Request(
f"http://127.0.0.1:{self.server.port}/api/adventures/{self.adv}/knowledge",
data=parts, method="POST",
headers={"Content-Type": f"multipart/form-data; boundary={boundary}"},
)
with urllib.request.urlopen(request, timeout=120) as response:
body_out = json.loads(response.read().decode())
self.note("knowledge", file=name, classification=classification,
id=body_out["id"])
def correct(self, events, note="") -> None:
self.server.call("POST", f"/adventures/{self.adv}/state/corrections",
{"events": events, "note": note})
self.note("state_correction", events=len(events), note=note)
# ----------------------------------------------------------------- play
def turn(self, text: str, *, kind="do") -> dict:
started = time.monotonic()
before = self.count_actions()
try:
events = self.server.stream(f"/adventures/{self.adv}/actions",
{"type": kind, "text": text})
except urllib.error.HTTPError as exc:
self.note("turn_failed", text=text, status=exc.code,
detail=exc.read().decode()[:300])
return {"accepted": False}
errors = [e for e in events if e.get("type") == "error"]
if errors:
self.note("turn_error", text=text, detail=errors[0].get("detail", "")[:300])
return {"accepted": False, "error": errors[0].get("detail", "")}
after = self.count_actions()
if after <= before:
self.note("turn_not_accepted", text=text)
return {"accepted": False}
self.accepted += 1
seconds = time.monotonic() - started
sample = self.measure()
self.note("turn", text=text, seconds=round(seconds, 1), **sample)
return {"accepted": True, "seconds": seconds, **sample}
def count_actions(self) -> int:
return self.server.call("GET", f"/adventures/{self.adv}/actions?limit=1")["total"]
def measure(self) -> dict:
"""M03's numbers, read from the prompt the app would send right now."""
report = self.server.call("GET", f"/adventures/{self.adv}/context")
tokens = report["tokens"]
sections = {s["label"]: s["tokens"] for s in report["sections"]}
window = report.get("window") or {}
return {
"total_actions": report["history"]["total"],
"history_included": report["history"]["included"],
"prompt_tokens": tokens["total"],
"budget": tokens["budget"],
"configured_budget": tokens.get("configured_budget"),
"output_reserve": tokens["output_reserve"],
"protected": tokens["protected"],
"available_for_history": tokens["available_for_history"],
"summary_tokens": sections.get("story_summary", 0),
"memory_tokens": sections.get("memories", 0),
"knowledge_tokens": sum(
v for k, v in sections.items() if k.startswith("knowledge")),
"state_tokens": sections.get("narrative_state", 0),
"canon_tokens": sections.get("campaign_canon", 0),
"window_verified": window.get("verified"),
"window_tokens": window.get("tokens"),
# Whether the *planted clue* is still visible anywhere in the
# assembled prompt. Named for what it measures: an earlier version
# called this `canon_present`, which it never was — the campaign
# canon's presence is `canon_tokens`, which is non-zero on every
# turn. This one going to zero is M04's precondition: the clue has
# left the recent-history window and can only come back through
# memory, summary or state.
"clue_in_prompt": CLUE_SENTINEL in json.dumps(report["sections"]),
}
def state(self) -> dict:
return self.server.call("GET", f"/adventures/{self.adv}/state")
def head(self) -> tuple:
page = self.server.call("GET", f"/adventures/{self.adv}/actions?limit=1")
return page["total"], page["can_undo"], page["can_redo"]
def main() -> int:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--turns", type=int, default=100)
parser.add_argument("--out", required=True)
args = parser.parse_args()
if not (ENDPOINT and MODEL):
print("set AIDND_TEST_ENDPOINT and AIDND_TEST_MODEL")
return 2
out = Path(args.out)
out.mkdir(parents=True, exist_ok=True)
db_path = out / "campaign.db"
server = Storyteller(db_path, out / "server.log")
server.start()
run = Run(server, out)
started = datetime.now()
try:
run.setup()
# ---- The planted clue, at the very beginning. ----
run.turn(f"I tell Mara quietly that {CLUE} — {CLUE_SENTINEL}.")
run.correct([
{"type": "add_fact", "subject": "aldric", "predicate": "knows",
"object": "abbey", "detail": f"{CLUE} ({CLUE_SENTINEL})",
"fact_id": "silver-key-opens-crypt"},
], note="the planted clue, as accepted state")
run.note("clue_planted", sentinel=CLUE_SENTINEL)
# ---- The long middle. ----
plan = _schedule(args.turns)
beat = 0
while run.accepted < args.turns:
step = plan.get(run.accepted + 1)
if step:
try:
_do_step(run, server, step)
except Exception as exc: # noqa: BLE001
# A step that fails is a finding, not a reason to lose the
# other ninety turns. It is recorded loudly and the campaign
# goes on, because an abandoned run proves nothing at all.
run.note("step_failed", step=step,
error=f"{type(exc).__name__}: {exc}"[:300])
run.turn(BEATS[beat % len(BEATS)])
beat += 1
# ---- M04: the recall check, with controls. ----
run.note("recall_begin")
recall = _recall(run)
(out / "recall.json").write_text(json.dumps(recall, indent=2))
# ---- Export the whole thing, for the recovery evidence. ----
bundle = server.call("GET", f"/adventures/{run.adv}/export")
(out / "bundle.json").write_text(json.dumps(bundle))
run.note("exported", bytes=len((out / "bundle.json").read_bytes()))
summary = {
"accepted_turns": run.accepted,
"restarts": server.starts - 1,
"elapsed_seconds": round((datetime.now() - started).total_seconds()),
"recall": recall,
"final_state": run.state()["document"],
"final_measurement": run.measure(),
"db_bytes": db_path.stat().st_size,
}
(out / "summary.json").write_text(json.dumps(summary, indent=2, default=str))
print(json.dumps({k: v for k, v in summary.items()
if k not in ("final_state",)}, indent=2, default=str)[:2000])
return 0
finally:
server.stop()
run.timeline.close()
def _schedule(turns: int) -> dict:
"""Where each required history operation happens. Spread, not clustered."""
unit = max(1, turns // 13)
return {
unit * 1: "save_point_1",
unit * 2: "restart",
unit * 3: "undo_redo",
unit * 4: "retry",
unit * 5: "save_point_2",
unit * 6: "restart_with_retained_history",
unit * 7: "retry",
unit * 8: "undo_diverge",
unit * 9: "take_selection",
unit * 10: "failed_call",
unit * 11: "restore_save_point",
unit * 12: "restart",
}
def _do_step(run: Run, server: Storyteller, step: str) -> None:
adv = run.adv
if step == "save_point_1":
point = server.call("POST", f"/adventures/{adv}/checkpoints",
{"name": "Before the ridge", "note": "planted clue is behind us"})
run.note("save_point", id=point["id"], name=point["name"])
elif step == "save_point_2":
point = server.call("POST", f"/adventures/{adv}/checkpoints",
{"name": "On the ridge", "note": ""})
run.note("save_point", id=point["id"], name=point["name"])
elif step in ("restart", "restart_with_retained_history"):
if step == "restart_with_retained_history":
server.call("POST", f"/adventures/{adv}/undo")
run.note("undo", why="leave retained history across the restart")
before = _snapshot(run)
server.restart()
after = _snapshot(run)
run.note("restart", number=server.starts - 1,
identical=before == after,
before=before, after=after)
elif step == "undo_redo":
total_before, _, _ = run.head()
server.call("POST", f"/adventures/{adv}/undo")
after_undo = run.head()
server.call("POST", f"/adventures/{adv}/redo")
after_redo = run.head()
run.note("undo_redo", before=total_before, after_undo=after_undo[0],
after_redo=after_redo[0], restored=after_redo[0] == total_before)
elif step == "retry":
# Retry regenerates a turn, so it streams like one. The first version of
# this harness called it as JSON and died on the SSE body — found by the
# shakeout run rather than fifty turns into the release campaign, which
# is what the shakeout was for.
events = server.stream(f"/adventures/{adv}/retry", {})
errors = [e for e in events if e.get("type") == "error"]
page = server.call("GET", f"/adventures/{adv}/actions?limit=3")
takes = max((a.get("take_count") or 1) for a in page["actions"])
run.note("retry", ok=not errors, takes_on_newest_turn=takes,
detail=(errors[0].get("detail", "")[:120] if errors else ""))
elif step == "take_selection":
# Retry first so there is more than one take to choose between, then
# step back to the earlier one — D07's "select prior retry take".
server.stream(f"/adventures/{adv}/retry", {})
page = server.call("GET", f"/adventures/{adv}/actions?limit=5")
multi = [a for a in page["actions"] if (a.get("take_count") or 1) > 1]
if multi:
target = multi[-1]
takes = server.call(
"GET", f"/adventures/{adv}/actions/{target['id']}/variants")
chosen = server.call(
"POST", f"/adventures/{adv}/actions/{target['id']}/variant",
{"index": 0})
run.note("take_selected", action=target["id"], of=len(takes),
chose_index=0, now_live=chosen["id"])
else:
run.note("take_selection_skipped", reason="no multi-take turn found")
elif step == "undo_diverge":
server.call("POST", f"/adventures/{adv}/undo")
server.call("POST", f"/adventures/{adv}/undo")
run.turn("I turn back towards the town instead.")
_, _, can_redo = run.head()
run.note("diverged", redo_available_after_new_writing=can_redo)
elif step == "restore_save_point":
points = server.call("GET", f"/adventures/{adv}/checkpoints")
if points:
target = points[0]
before = run.head()
page = server.call(
"POST", f"/adventures/{adv}/checkpoints/{target['id']}/restore")
run.note("save_point_restored", id=target["id"], name=target["name"],
total_before=before[0], total_after=page["total"],
can_redo=page["can_redo"])
else:
run.note("restore_skipped", reason="no save point exists yet")
elif step == "failed_call":
# A real failure: point the model at a name the server does not serve,
# take a turn, and put it back. Nothing is mocked.
settings = server.call("GET", "/settings")
before_total, _, _ = run.head()
before_state = run.state()["document"]
server.call("PUT", "/settings", {"model": "no-such-model-m11"})
try:
events = server.stream(
f"/adventures/{adv}/actions",
{"type": "do", "text": "I look for a way across the water."},
timeout=180)
errors = [e for e in events if e.get("type") == "error"]
outcome = (f"reported: {errors[0].get('detail', '')[:120]}"
if errors else "NO ERROR REPORTED")
except urllib.error.HTTPError as exc:
outcome = f"HTTP {exc.code}"
except Exception as exc: # noqa: BLE001 - recorded, not swallowed
outcome = type(exc).__name__
server.call("PUT", "/settings", {"model": settings["model"]})
after_total, _, _ = run.head()
run.note("failed_call", outcome=outcome,
actions_before=before_total, actions_after=after_total,
state_unchanged=before_state == run.state()["document"])
# And prove play resumes.
run.turn("I ask the ferryman again, more politely.")
def _snapshot(run: Run) -> dict:
"""What must be identical across a restart (M02's list)."""
adv = run.adv
page = run.server.call("GET", f"/adventures/{adv}/actions?limit=3")
state = run.state()["document"]
points = run.server.call("GET", f"/adventures/{adv}/checkpoints")
knowledge = run.server.call("GET", f"/adventures/{adv}/knowledge")
settings = run.server.call("GET", "/settings")
return {
"total": page["total"],
"can_undo": page["can_undo"],
"can_redo": page["can_redo"],
"newest": [a["text"][:60] for a in page["actions"]],
"scene": (state.get("scene") or {}).get("summary"),
"entities": sorted(state.get("entities") or {}),
"facts": sorted(f.get("id") for f in state.get("facts") or []),
"save_points": sorted(p["name"] for p in points),
"knowledge": sorted(k["original_filename"] for k in knowledge),
"model": settings["model"],
"budget": settings["context_token_budget"],
}
def _recall(run: Run) -> dict:
"""M04: can the planted clue still be found, and not from the transcript?"""
adv, server = run.adv, run.server
# 1. Is the clue outside the recent-history window? (Precondition, not result.)
report = server.call("GET", f"/adventures/{adv}/context")
history_text = " ".join(
s["text"] for s in report["sections"] if s["label"] == "story_history")
in_history = CLUE_SENTINEL in history_text
# 2. Ask about the subject, and see what the application assembles.
run.turn("I think back to what I told Mara about the key, that first night.")
after = server.call("GET", f"/adventures/{adv}/context")
sections = {s["label"]: s["text"] for s in after["sections"]}
whole_prompt = "\n".join(sections.values())
# 3. Where did it come from? State, summary, memory, retrieval — or nowhere.
document = run.state()["document"]
fact_present = any(
CLUE_SENTINEL in json.dumps(f) for f in document.get("facts") or [])
return {
"clue_in_recent_history_window": in_history,
"clue_in_prompt": CLUE_SENTINEL in whole_prompt,
"in_state_section": CLUE_SENTINEL in sections.get("narrative_state", ""),
"in_summary_section": CLUE_SENTINEL in sections.get("story_summary", ""),
"in_memories_section": CLUE_SENTINEL in sections.get("memories", ""),
"in_knowledge_sections": any(
CLUE_SENTINEL in text for label, text in sections.items()
if label.startswith("knowledge")),
"fact_still_in_state": fact_present,
"history_included": after["history"]["included"],
"history_total": after["history"]["total"],
"memories_used": [
m.get("text", "")[:120] for m in (after.get("memories") or {}).get("used", [])
],
"prompt_tokens": after["tokens"]["total"],
}
if __name__ == "__main__":
raise SystemExit(main())