fix(graduation): session-count-aware state for backfilled patterns

[Gradata]

Summary

Cherry-picked from a local commit on the main worktree (40f33e5a, authored 2026-04-19 by Oliver) — promoting it to a reviewable PR so it can land cleanly before the v0.6.1 tag.

_patterns_to_graduated_lessons previously teleported every qualifying correction_pattern directly to RULE@0.92, bypassing the INSTINCT→PATTERN→RULE graduation gates. A 2-session pattern ended up with the same standing as a 50-session rule. Live brain showed 18 RULEs / 0 PATTERNs / 2 INSTINCTs — all 18 RULEs came from weak 2-session evidence.

Fix

_state_for_sessions now maps evidence → state:

• 2 sessions → PATTERN @ 0.70
• 3-4 → PATTERN @ 0.80
• 5+ → RULE @ 0.92

Aligns backfill with live graduation thresholds (MIN_APPLICATIONS_FOR_PATTERN=3, MIN_APPLICATIONS_FOR_RULE=5). Backfill still filters single-session patterns (min_sessions=2).

Test plan

• pytest tests/test_rule_pipeline.py — 26 passed in 0.56s
• After merge, re-backfill live brain and confirm state distribution is no longer 18/0/2

Why now

Needed before cutting the v0.6.1 tag — without this, any brain backfilled on 0.6.1 will replicate the over-promotion bug. Blocker for #117.

Generated with Gradata

[greptile-apps]

Your free trial has ended. If you'd like to continue receiving code reviews, you can add a payment method here.

[coderabbitai]

Caution

Review failed

Pull request was closed or merged during review

📝 Walkthrough

Summary

• Fixed over-promotion during backfill: Previously, all backfilled patterns were promoted to RULE state with 0.92 confidence regardless of session count, causing weak 2-session evidence to be treated identically to battle-tested 50-session rules.

• Session-aware state mapping: Introduced _state_for_sessions() function that derives both LessonState and confidence based on distinct session counts:

  • 2 sessions → PATTERN @ 0.70
  • 3–4 sessions → PATTERN @ 0.80
  • 5+ sessions → RULE @ 0.92

• Aligns with live graduation gates: Thresholds now match live system constants (MIN_APPLICATIONS_FOR_PATTERN=3, MIN_APPLICATIONS_FOR_RULE=5).

• Updated _patterns_to_graduated_lessons(): Now computes distinct_sessions per candidate and calls _state_for_sessions() to populate Lesson.state and Lesson.confidence instead of hard-coding RULE/0.92.

• Test coverage: Updated existing test expectations and added new comprehensive test validating all three state tiers map correctly based on session count.

• No public API changes: Modified function is internal; no breaking changes to public interfaces.

• Post-merge manual verification required: Re-backfill live brain to confirm state distribution no longer shows the over-promotion pattern (expected change from 18 RULEs / 0 PATTERNs / 2 INSTINCTs distribution).

Walkthrough

Updated the rule pipeline to derive lesson state and confidence based on the count of distinct sessions rather than using fixed values. Added a tiered threshold function that maps session counts to state transitions and session-aware confidence scores, replacing the previous hard-coded RULE state and 0.92 confidence.

Changes

Cohort / File(s)	Summary
Session-Based State Determination src/gradata/enhancements/rule_pipeline.py	Added _state_for_sessions() function implementing tiered thresholds (2→PATTERN@0.70, 3–4→PATTERN@0.80, 5+→RULE@0.92). Modified _patterns_to_graduated_lessons() to compute distinct sessions and delegate state/confidence assignment to the new function instead of hard-coding RULE with 0.92.
Test Updates tests/test_rule_pipeline.py	Updated test assertions to expect LessonState.PATTERN with ~0.70 confidence for 2-session patterns instead of RULE with ≥0.90. Added new test validating session-count-driven state transitions and confidence outputs across all three threshold categories.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Suggested labels

bug

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: adding session-count awareness to determine lesson state during pattern graduation, replacing the previous hard-coded RULE@0.92 assignment.
Description check	✅ Passed	The description provides clear context for the fix, explaining the prior bug, the new mapping logic, and the test coverage, all directly related to the changeset.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/graduation-session-count-state

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}