research(GRA-1477): patch-oscillation analysis 2026-06-02

[Gradata]

Summary

• Delivers docs/research/patch-oscillation-2026-06-02.md as required by GRA-1477
• Answers: are self-healing patches converging or oscillating?
• Key finding: zero RULE_PATCHED events — the patch pipeline has never fired; oscillation guard is untriggered
• Identifies first confirmed oscillation signal: never-use-em-dashes hook reverts on 19-day cycle (test teardown artifact, not production)
• Names two persistent failure candidates with freeze/demote recommendations
• Supersedes GRA-1382 baseline (2026-05-23); delta table included

Artifact

docs/research/patch-oscillation-2026-06-02.md

Test plan

• Confirm artifact file exists at the required path
• Confirm freeze/demote candidates are actionable
• No code changes — research doc only

🤖 Generated with Claude Code

[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

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro

Run ID: 3c9a21cf-ba60-4bc4-baad-1f9e5bc2d14f

📥 Commits

Reviewing files that changed from the base of the PR and between be4df94 and 4cd02fe.

📒 Files selected for processing (1)

• Gradata/docs/research/patch-oscillation-2026-06-02.md

---

📝 Walkthrough

• Research document for GRA-1477: Analyzes whether self-healing patches are converging or oscillating, superseding the GRA-1382 baseline from 2026-05-23
• Zero patch activity: No RULE_PATCHED events recorded — the patch pipeline has never fired; oscillation guard remains untriggered
• First oscillation signal identified: never-use-em-dashes hook reverts on a 19-day cycle (May 14 and June 2), but classified as test teardown artifact from pytest, not production behavior
• Two persistent failure rules flagged: Include more detail about detail (9 failures across 4 dates, HIGH oscillation risk) and Verify facts, numbers, and dates (7 failures across 3 dates, LOW risk) identified as leading indicators for when auto_heal() activates
• Seven burst-only rules documented: Single-session content rule failures, ephemeral and low-priority for patching
• Actionable recommendations: Freeze content-churn rules; promote generic meta-corrections to meta_rule tier; filter raw-python platform source from reversion monitoring; mark Verify facts, numbers, dates as first controlled patch candidate
• Event growth since May 23: RULE_FAILURE increased from 46 to 142 (+96); LESSON_CHANGE from 353 to 884 (+531); RULE_GRADUATED from 116 to 360 (+244)
• No code changes — research document only (196 lines added)

Walkthrough

This PR adds a new research document that analyzes patch oscillation and convergence patterns within the self-healing system. The document presents four empirical findings: zero patch-pipeline activity, a detected 19-day oscillation in the hook pathway, persistent multi-session failure rule candidates, and risk assessments for two specific rules. It concludes with operational policy changes, a final verdict, and updated recommendations for patch management.

Changes

Patch Oscillation Analysis Report

Layer / File(s)	Summary
Document setup and analysis methodology Gradata/docs/research/patch-oscillation-2026-06-02.md	Introduces the research context, prior work references, executive summary noting no patcher firing and new hook-path oscillation evidence, and defines the database query strategy and analysis date range.
Empirical findings on patch behavior and oscillation patterns Gradata/docs/research/patch-oscillation-2026-06-02.md	Presents four findings: (1) RULE_PATCHED event counts remain zero, (2) hook-pathway oscillation in never-use-em-dashes over 19 days with pytest teardown interpretation, (3) frequency-based RULE_FAILURE candidates within a 14-day window separated by persistence, and (4) deep compliance analysis of two persistent rules with oscillation risk levels and freeze/monitor recommendations.
Operational policy, verdict, and recommendations Gradata/docs/research/patch-oscillation-2026-06-02.md	Specifies freeze criteria and demotion thresholds for oscillation candidates, states the convergence/oscillation verdict across patch and hook pathways, identifies highest-risk and safest first-patch candidates, updates recommendations including platform_source filtering and meta-rule promotion, and includes a supporting data summary table of event counts and database date range.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~5 minutes

Suggested labels

docs

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch research/gra-1477-patch-oscillation

---

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