✔ archived Turn 008 fixture passes schema and domain validation (9.743708ms)
✔ an OPEN turn allows eligible players before intake is locked (7.441333ms)
✔ view generation is deterministic (12.110125ms)
✔ one fact change updates relevant derivatives without disturbing unrelated views (11.31275ms)
✔ detects a skipped approval gate (7.1405ms)
✔ detects a stale calendar label (58.148459ms)
✔ detects an over-limit intake package (39.784833ms)
✔ detects an intake checksum mismatch (5.516959ms)
✔ detects a wrong-player delivery destination (16.972375ms)
✔ detects a delivery checksum mismatch (7.395625ms)
✔ detects disclosure of a GM-only fact (10.913ms)
✔ detects unsupported knowledge propagation (24.994208ms)
✔ detects an active rule without matching approval (4.038333ms)
✔ detects elimination without specific GM confirmation (2.840792ms)
✔ detects a third miss that bypasses the elimination gate (1.5325ms)
✔ detects an unapproved state-change source (4.204917ms)
✔ detects excessive substantive GM decisions (3.830584ms)
✔ detects a future-dated pressure state (2.722042ms)
✔ detects duplicate entity IDs (1.91025ms)
✔ Antigravity translates empty operation enums without altering native values or raw evidence (1.836917ms)
✔ Antigravity creates separate tool-free role folders with no duplicated private input or ambient customizations (2.925709ms)
✔ Antigravity rejects all filesystem and non-output tool use (6.895584ms)
✔ Antigravity refuses ambient filesystem grants and unaccountable dollar budgets (1.9155ms)
✔ Antigravity requires agent discovery, fresh stdin session, exact model and completed structured output (2.498875ms)
✔ startup probe discovers a confined agent without issuing a model request (3.98175ms)
(node:31776) ExperimentalWarning: SQLite is an experimental feature and might change at any time
(Use `node --trace-warnings ...` to show where the warning was created)
✔ branch preserves player-only memory and replay excludes future results and makes no player calls (14547.644083ms)
✔ required trope retrieval uses a pinned corpus and records an explicit disposition (3735.945375ms)
✔ AI GM can reject a recommendation and cannot alter baseline mechanics (5546.053333ms)
✔ single-writer lock refuses an active owner and reclaims only a dead owner (480.127667ms)
✔ provider arguments disable tools/context loading and keep shell syntax out of command interpretation (0.360583ms)
✔ subprocess timeout and output bounds terminate calls instead of hanging (277.455542ms)
✔ Claude stream parsing requires a successful final result and retains intermediate evidence (0.688333ms)
✔ provider budget failures preserve billed cost and aggregate all model usage even on nonzero exit (257.917334ms)
✔ native GM values avoid nested JSON escaping and Codex transport normalizes to the same contract (0.405833ms)
(node:31777) ExperimentalWarning: SQLite is an experimental feature and might change at any time
(Use `node --trace-warnings ...` to show where the warning was created)
✔ paired analysis fork preserves exact inputs, ablates GM wire contract and resumes without new calls (10839.509417ms)
✔ positive disposition validates all eight fields, rejects invented IDs, repairs, commits and survives delivery resume (3872.727583ms)
✔ missing causal-fit fields fail deterministically and unsupported escalation fails independent fixture review (15442.888542ms)
✔ private retrieval language in a report is rejected before delivery (1799.873584ms)
✔ aggregate accounting counts inherited starts once and refuses exhausted shared budgets (3938.792291ms)
✔ host-access failures stop after one durable start, while transient failures can retry (1.3425ms)
✔ persisted exhaustion never silently succeeds and unstarted failures do not loop (0.327541ms)
✔ OAuth endpoint socket failures retry, while durable credential failures stop (0.168208ms)
✔ a required preparation repair blocks without spending another reader attempt (0.122542ms)
(node:31779) ExperimentalWarning: SQLite is an experimental feature and might change at any time
(Use `node --trace-warnings ...` to show where the warning was created)
✔ six-turn simultaneous campaign closes, preserves private memory, resumes without new calls, and is rejected by live schema (24406.425584ms)
✔ crash after first mailbox write resumes without duplicate state or additional model calls (3180.032667ms)
✔ third explicit miss requires a specific elimination and gets a final report; future turns omit that player (9458.145708ms)
✔ a separately confirmed causal elimination is allowed during a second absence (3508.921875ms)
✔ new emergent rule activation has its own AI approval and append-only provenance (1859.728958ms)
✔ unconfirmed elimination, prohibited roster edits, and hidden-fact citations fail closed (3302.469917ms)
✔ forged human approval is rejected without altering the source approval (236.541584ms)
✔ oversized submission can be repaired; persistent invalid output consumes bounded attempts (2727.458ms)
✔ review failure redrafts prose, and causal rejection never delivers (5336.8195ms)
✔ final review GM finding re-adjudicates from original state and redrafts before delivery (2882.560375ms)
✔ resume preserves rejected output and supplies the current actionable validation error (395.187917ms)
✔ transport failure between repair attempts preserves actionable validation feedback (2466.936375ms)
✔ unknown mutation fields identify the target and allowed fields for model repair (285.933875ms)
✔ empty trope retrievals receive the only legal deterministic disposition (0.336083ms)
✔ a model global rejection is expanded across exact retrieval IDs (0.164292ms)
✔ role-scoped state keeps canonical entities and removes accumulated audit history (250.281917ms)
✔ role rules retain core mechanics while omitting irrelevant appendices (245.569167ms)
✔ Antigravity oversized role packets fail before a provider call starts (300.645375ms)
✔ call budget and transport failure do not become player absence (1516.237417ms)
✔ tampered input, missing generated views, mailbox, journal, and symlink escapes are detected (4540.476458ms)
✔ score arithmetic follows calibrated limits and is defect-sensitive (0.147167ms)
✔ trope concepts accept abstract phrases and reject names or prose before analysis is accepted (0.299416ms)
✔ vendored state logic differs only by documented simulation adaptations (0.972667ms)
✔ unknown policy, ambiguous models and unaccountable dollar budgets fail configuration (0.39375ms)
✔ hidden source visibility changes are rejected before any writer, with actionable observation repair (2029.869083ms)
✔ pre-draft review catches copied secrets and revises from original state without carrying rejected facts (2183.950084ms)
✔ pre-draft rejection exhausts bounded GM rounds without report calls or delivery (1285.694625ms)
✔ pre-draft review cannot approve with omitted recipient/fact checks or contradictory findings (2296.591583ms)
✔ disclosure docket includes changed wording, confidence and new recipients (1896.303ms)
✔ missing inner knowledge ID is diagnosed and repaired without accepting malformed state (2041.330542ms)
✔ host startup failure preserves one attempt and resumes with remaining attempts under unchanged pins (397.76475ms)
✔ startup preflight blocks all model scheduling, keeps diagnostics and resumes after environment repair (2680.682875ms)
ℹ tests 75
ℹ suites 0
ℹ pass 75
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 82012.123542