Test_Report

RimSynapse Test Report

Auto-generated by the in-game TestRunner. Cases split into two tiers:

• Determination (upstream) — does the LLM decide correctly (the assessment/script/memory it produces)? Runs against a live model (-synapse-determination).
• Execution (downstream) — given a decision, does the engine execute it correctly? Deterministic, structured-field assertions.

Determination — upstream (does the LLM decide correctly?)

Result	+/-	Scenario	Expectation	Detail
✅ PASS	negative	Live eval of a colonist who spent the day attacking an object (the pod-car case).	PersonalityShiftLikelihood is none or low (object violence isn't bloodlust).	likelihood=none
❌ FAIL	positive	Live eval of a colonist with lifetime humanlike kills who fought the living today.	PersonalityShiftLikelihood is moderate or high (a shift is at least plausible).	likelihood=low
✅ PASS	positive	Live memory generation for a minor event.	The generated Weight is within the 0-1 scale.	weight=0.20

Last run: 2 passed, 1 failed, 0 skipped — 2026-08-08 19:12.

Execution — downstream (does the engine execute a decision correctly?)

Result	+/-	Scenario	Expectation	Detail
✅ PASS	negative	A colonist spends the day attacking a wrecked vehicle (an object).	The activity reads as one deduped, non-lethal, object-targeted line — not violent signal.	AttackStatic an inanimate object — non-lethal, not violence against the living (100%)
✅ PASS	positive	A colonist attacks a person and an animal.	Combat against the living is NOT labelled non-lethal.	AttackMelee a person (100%)
✅ PASS	negative	A colonist has one idle chit-chat memory and nothing links to it.	It prunes within the daily maintenance pass (noise does not accumulate).	memories 1->0 (pruned 1); long-term 0->0 (consolidated 0)
✅ PASS	positive	Chit-chat about a colonist who then dies (both share the pawn's id).	The chit-chat is consolidated to long-term the day the death lands.	salience=1.10
✅ PASS	negative	The daily evaluation calls a Bloodlust shift 'unlikely' (low likelihood).	Today contributes zero pressure — an unlikely day can never flip a trait.	pressure=0.00
✅ PASS	positive	Genuine violence against the living recurs over three days.	Trait pressure accumulates across days and crosses the shift threshold.	day3 pressure=1.10
✅ PASS	negative	A malformed/over-eager evaluation tries to add Psychopath.	The whitelist blocks it; only sanctioned traits may be added/removed.	whitelist enforced
✅ PASS	positive	WorldNews queries the R&T availability seam with R&T loaded	Detection reports active and the guarded worker reads R&T world state without throwing	R&T detected; seam read world state: "R&T active; 312 provinces, 279 with a listed owner."
✅ PASS	positive	A full expanded-issue LLM response is parsed through ParseIssue	Masthead, lead+secondary stories, sidebars and ads survive; image prompt captured, asset unresolved	Round-tripped: 2 stories, lead image-wanted, 1 sidebar(s), 1 ad(s)
✅ PASS	positive	A minimal response omitting masthead, sidebars, ads and images is parsed	Masthead is defaulted and collections are non-null; no exception	Sparse issue normalized: paper='The Rimworld Gazette', 1 story, no images

Last run: 10 passed, 0 failed, 0 skipped — 2026-08-15 23:25.