experiment: validate behaviour selection against real assessments

docs/current-handoff.md

@@ -4,18 +4,18 @@
 
 ## 1. Where We Left It
 
-- Engine experiments are paused after Experiment 25B;
-- UI experiments are paused;
+- Engine experiments resumed with a passive validation;
+- UI experiments remain paused;
 - Knowledge-management experiments are complete;
-- Experiment 38 validated that a cold session recovers project state from three documents alone;
-- The working tree was clean at commit `544573a` (Exp 37).
+- Experiment 39 tested the existing Behaviour Selection module against real Investigation State Assessment outputs across three scenarios;
+- Acknowledge dominates (71% of selections) because it fires first when health=healthy, blocking Summarise/Pause/Clarify even in concluding or stalled states.
 
 > This handoff describes the latest stopping point only. When work moves on, replace stale current-work details rather than appending another historical note. Historical experiment and commit information belongs in `docs/design-evolution-log.md`.
 
 ## 2. What Is True Now
 
 - Main active engine path: deterministic reasoning pipeline (scenario reconstruction, graph update, unknown selection, question formulation, turn orchestration).
-- Passive experimental classifiers from Experiments 18–25B remain isolated diagnostic layers; none control the user-facing investigation.
+- Passive experimental classifiers from Experiments 18–25B remain isolated diagnostic layers; none control the user-facing investigation. Behaviour Selection was passively evaluated against real assessment outputs in Experiment 39 — it produced all valid behaviours but with skewed distribution (Acknowledge 71%).
 - Keyword and phrase-based scope detection remains provisional scaffolding.
 - `docs/current-project-state.md` is the main entry point for active project state.
 - `docs/task-context-packs.md` chooses the minimum context documents for each work type.
@@ -34,9 +34,13 @@ Experiment 38 tested whether a genuinely cold session (no prior conversation con
 
 **Commit:** pending (experiment: validate cold-start project recovery) — to be committed this session.
 
+Experiment 39 resumed reasoning experiments with a passive validation of Behaviour Selection against real Investigation State Assessment outputs. Seven turns across three scenarios were evaluated. Acknowledge dominated (71%) because it fires at priority 1 whenever health=healthy, even in terminal and stalled states where Summarise or Pause would be more useful. The assessor→selector contract aligns cleanly; no transformation is needed between pipeline stages. All five behaviours remain reachable but some never appear in typical scenarios (Clarify requires too_broad health which few fixtures produce). Status pending Rob's review.
+
 ## 5. What Remains Open
 
-- Whether tasks crossing pack boundaries can still stay concise without loading excessive context (not tested by this experiment).
+- Whether the Acknowledge→Summarise priority conflict can be resolved without breaking existing rule conditions;
+- Whether behaviour selection produces different patterns in investigation domains other than supplier procurement, market entry, and product comparison;
+- Which rules (Acknowledge, Clarify, Summarise, Pause) fire most frequently across a larger set of scenarios.
 
 ### When This Knowledge-Management Phase Is Complete
 
@@ -82,8 +86,8 @@ Answer before continuing:
 
 ---
 
-*Created by Experiment 34. Updated by Experiment 38. Branch: `feature/user-workspace-ux-v0.7`.*
+*Created by Experiment 34. Updated by Experiments 38, 39. Branch: `feature/user-workspace-ux-v0.7`.*
 
-### Return-to-Work Note (Experiment 38)
+### Return-to-Work Note (Experiment 39)
 
-Experiment 38 was performed from a cold session with no prior conversation context and confirmed that the reduced context system works as designed. A genuinely fresh session recovered the complete project state—active capabilities, passive diagnostic layers, paused work items, and context-pack routing—from just three documents without reading the full design-evolution log or any archived material. All seven knowledge-management completion criteria are confirmed met. The handoff itself was found to be accurate with no stale or ambiguous statements; one structural update was made (resolving the open item about handoff longevity). Engine and UI experiments remain paused pending Rob's review. Branch: `feature/user-workspace-ux-v0.7`. First file to inspect when returning: `docs/current-handoff.md`, then `docs/current-project-state.md` and the Engine Experiment pack per the routing table. The next action is discussion, not automatic implementation.
+Experiment 39 resumed reasoning experiments with a passive validation of Behaviour Selection against real Investigation State Assessment outputs across three scenarios (long investigation, contradictory evidence, short early). No user-facing engine behaviour changed. The selector produced all valid behaviours but with skewed distribution: Acknowledge dominated at 71% because it fires first when health=healthy, even in terminal/concluding states where Summarise would be more useful — this is an acknowledged priority conflict, not a contract mismatch. The assessor→selector pipeline aligns cleanly. Clarify and Pause never fired (no too_broad health produced; other rules blocked). Status pending Rob's review. Branch: `feature/user-workspace-ux-v0.7`. First file to inspect when returning: `docs/current-handoff.md`, then `docs/design-evolution-log.md` entry for Experiment 39, and `tests/behaviour-selection.real-assessment.test.js` for evaluation details. The next action is discussion about whether to refine the selector or move on.