Project: Covert Cubicles Build Phase 4 / 7

MODULE 16 / 31 · LAYER 4 · ALMAA_REASONING Reasoning Quality Detector

Operationalises Decision Pathways (D) from the Primary Spine framework. Computes a 0\u2013100 reasoningScore from up to five signals; classifies into scaffold / challenge / no_action. Two scoring modes: running (during play) and final (at end-game).

DEPENDENCIES: SM_Game, LOGIC_Decision
Pending
DRAFT FORMULA \u00b7 NOT PILOT-CALIBRATED
A draft reasoning-scoring formula is currently in effect.
Two scoring modes (DRAFT ADR-036)
Running mode uses only two of the five signals (evidence_consulted_count, justification_coherence) and treats the three accusation_*_logic signals as neutral 0.5 each. Used during play when M18/M19 might want to evaluate scaffold/challenge adaptations on partial data. Running scores are bounded to roughly [25, 75].

Final mode requires all five signals and is called once by END_Accuse with the accusation-derived motive/means/opportunity logic scores. This is the canonical reasoningScore recorded in the session reasoningLog. NOT directly comparable to the running score.

Automated Assertions

#AssertionResultDetail

Interactive Workbench — Score Sandbox (both modes)

Adjust the five signals to see how the running and final scores diverge. The running score (left) ignores the three accusation signals and treats them as 0.5; the final score (right) uses all five.

Signals (raw observation)
During-play signals
3
0.60
Accusation signals (final mode only)
0.50
0.50
0.50
Theoretical presets
Computed scores: running vs final
During play
RUNNING
0.00
no_action
At end-game
FINAL
0.00
no_action
scaffold
no_action
challenge
04070100
\u25BC = final   \u00b7   faded \u25BC = running
Lifecycle test \u2014 evidence \u2192 justifications \u2192 final accusation

Drives the full per-session lifecycle. Running score updates as evidence and justifications accumulate; recordFinalAccusation() commits the canonical record to SM_Game.reasoningLog.

Session
Evidence consultation (deduped per (decision, clue))
Justification coherence (UI-supplied label in [0,1])
End-game (final accusation)
Defensive (should fail)
Evidence consulted total
\u2014
Justifications recorded
\u2014
Session reasoning log
No reasoning records yet.

API Surface — Manual Verification

Open DevTools and try the calls below.