Skip to content

PR_REVIEW_MATRIX.md

Latest commit

 

History

History
22 lines (16 loc) · 3.12 KB

PR_REVIEW_MATRIX.md

File metadata and controls

22 lines (16 loc) · 3.12 KB

PR Review Matrix

Reviewed via GitHub API on 2026-05-10 and local fetched refs origin/pr/*. Current hardening branch already contains historical PR #9 and now locally merged PR #10 with a traceable merge commit.

PR Branch Classification Overlapping edits Determinism / replay risks Tokenizer drift risks Disposition
#10 Add sparse micro-frame for tiny packets; update README, benchmarks, and tests codex/fix-short_sparse_3-and-update-readme sparse routing logic; core compression logic; benchmark validation Overlaps README, benchmark runner, core engine, benchmark tests with prior sparse guardrail work. Low after conflict resolution: micro-frame is deterministic and keeps audit layers; sparse edge cases must remain replay-tested. Medium: changes compressed payload shape for tiny packets; covered by token telemetry regression. Merged first per required order as Merge PR #10 sparse micro-frame guardrail.
#4 Initial CompText V7 package: KVTC‑V7 engine, agents, auditor, tests, and CI codex/create-initial-directory-structure-for-comptext-v7-iz7ikn validation infrastructure; dashboard/UI-adjacent agents; CI; initial scaffold Large stale scaffold overlaps current engine, README, pyproject, tests, benchmarks, and deletes current audit benchmark files. High hidden regression risk: would remove mature benchmark/audit modules and replace engine with older implementation. High: older codec/token assumptions conflict with current deterministic telemetry hardening. Do not merge into hardened branch without manual cherry-pick of non-regressive CI ideas.
#2 Initiale KVTC‑V7 Engine & CompText V7 Projektgerüst codex/create-initial-directory-structure-for-comptext-v7 core compression logic; initial scaffold Stale duplicate of initial package overlaps current engine, README, pyproject, tests, and deletes benchmark/audit files. Critical regression risk: reverts post-PR #6/#7/#8/#9 industrial audit work. High: predates token telemetry and sparse micro-frame guardrails. Do not merge; superseded by mainline history.

Required merge-order audit

  1. sparse_guardrail / short_sparse_3 fix — satisfied by prior Fix sparse benchmark guardrail (#9) plus local merge of PR #10 micro-frame refinement.
  2. deterministic token telemetry + tiktoken — implemented in src/validation/token_telemetry.py with cl100k_base and o200k_base support plus fallback drift sentinels.
  3. validation harness + replay engine — implemented in src/validation/replay.py and scripts/validate.py.
  4. semantic forensic audit layer — implemented in src/validation/forensic.py.
  5. dashboard layer — implemented in dashboard/industrial_dashboard.py.
  6. governance/docs/reports/program.md — implemented as program.md and release reports.

Release-audit decision

Open PRs #2 and #4 are stale scaffolding PRs with destructive overlaps. Merging them after PR #10 would violate semantic integrity, replayability, and auditability. Their non-regressive intent is captured in this hardening pass without accepting deletions of current benchmark and audit infrastructure.