Every commit scored for change-risk against this repo's own history, so 'elevated' means elevated here rather than on some global curve.
Needs review
105 commits sit in this repo's top risk tercile, which is 34% of the 313scored. The cut is drawn against this codebase's own history rather than a global curve, so a quiet repo still fills its top band, and here it starts at 8.2 out of 10. What pushes a commit up is size and spread together: a large change confined to one area scores below a smaller one scattered across a dozen files.
Commit categories over time, read off the subject line. Fixes carry the accent because that is the series this chart exists to show.
Began feature-led, now leaning fix.
Ranked by change-risk, highest first. Priority is a tercile of this repo's own distribution, so a quiet repo still fills its top band.
| # | Commit | Author | When | Lines | Risk | Top driver |
|---|---|---|---|---|---|---|
| 1 | 76542b4aeval(swe-bench): Lite pilot 10 paired instances — 10/10 resolved both arms, ON $27.27 vs OFF $53.61 (-49%) | teamchong | 2mo ago | +32.5K -1 | 100%Elevated | more lines added than baseline |
| 2 | c6a0d61feval(gist-recall): A/B imaged vs text history — 98/98 both arms, 0 confabulation | teamchong | 2mo ago | +17.4K -1 | 100%Elevated | more lines added than baseline |
| 3 | 47b5e412feat: cache-stable history-collapse imaging for GPT (OpenAI) + Anthropic | Steven Chong | 2mo ago | +3.5K -520 | 98%Elevated | more lines added than baseline |
| 4 | 81550f69feat(render): pack reflow across newlines + grayscale atlas + L1/L2 eval harness | teamchong | 3mo ago | +2.9K -952 | 98%Elevated | more lines added than baseline |
| 5 | 2b7b98f6feat(render): add R3 reflow to recover line-end dead margin (~29% glyph fill → dense) | teamchong | 3mo ago | +4.1K -48 | 98%Elevated | more lines added than baseline |
| 6 | b5d768fcdashboard: sessions UI + stats panel + smooth updates + cleanup | teamchong | 3mo ago | +2.3K -124 | 98%Elevated | more lines added than baseline |
| 7 | f199640bship Unifont @ 10px multilingual rendering + 400 diagnosis fixes | teamchong | 3mo ago | +3.1K -180 | 98%Elevated | more lines added than baseline |
| 8 | f91c149dPort to TypeScript with dual Node + Cloudflare Workers targets | teamchong | 3mo ago | +5.6K -10 | 98%Elevated | more lines added than baseline |
| 9 | 482c9672Add Rust port of proxy (unproven, not wired up) | teamchong | 3mo ago | +3.4K -0 | 98%Elevated | more lines added than baseline |
| 10 | 76a37102feat(export): replace in-band @pxpipe directive with explicit `pxpipe export` CLI | teamchong | 1mo ago | +1.9K -19 | 97%Elevated | more lines added than baseline |
| 11 | 45c83dedfeat(dashboard): rewrite dashboard in Svelte (#25) | teamchong | 3mo ago | +2.5K -821 | 97%Elevated | more lines added than baseline |
| 12 | 01037ed0fix(deps): actually track pnpm-lock.yaml | teamchong | 3mo ago | +2.1K -0 | 97%Elevated | more lines added than baseline |
| 13 | 84173800test: e2e guards for cache stability, dashboard honesty, and design behaviors | teamchong | 1mo ago | +1.1K -0 | 96%Elevated | more lines added than baseline |
| 14 | 9fddbeb2Initial commit: working Python proxy | teamchong | 3mo ago | +1.9K -0 | 96%Elevated | more lines added than baseline |
| 15 | 9199a8d0fix(eval): opus-density — commit live issue #6 results, fix two response-reading bugs | Deval Shah | 1mo ago | +749 -17 | 95%Elevated | more lines added than baseline |
| 16 | ef9ddac5feat(library): caller fidelity + recoverable channel, edge-safe; trim comments | teamchong | 2mo ago | +1.1K -2.4K | 95%Elevated | more lines added than baseline |
| 17 | ad7ef7c1docs(demo): cost-ab + effective-context A/B demos with a model arg | teamchong | 2mo ago | +870 -56 | 95%Elevated | more lines added than baseline |
| 18 | 97fdee84history: Variant C closed-tool-sequence collapse + honest per-session savings | teamchong | 3mo ago | +920 -19 | 95%Elevated | more lines added than baseline |
| 19 | 04b1526btransform: large-tool_result paging + adaptive break-even derivation | teamchong | 3mo ago | +799 -20 | 95%Elevated | more lines added than baseline |
| 20 | fe6b0ec2feat: structure-through role attribution + turn-index recency anchor for history images | teamchong | 2mo ago | +581 -102 | 94%Elevated | more lines added than baseline |
| 21 | 13456812redesign dashboard: light flame theme, plain-language hero, image/text transparency | teamchong | 2mo ago | +663 -430 | 94%Elevated | more lines added than baseline |
| 22 | 97d50513feat(openai): GPT-5.x family + Responses API support | teamchong | 2mo ago | +604 -76 | 92%Elevated | more lines added than baseline |
| 23 | cb522a70feat(dashboard): replace Svelte bundle with AHA stack (htmx + Alpine, server-rendered fragments) | teamchong | 2mo ago | +785 -2.1K | 92%Elevated | more lines added than baseline |
| 24 | 613ca3bcfeat(proxy): support gpt 5.5 chat completions | Steven Chong | 2mo ago | +742 -49 | 92%Elevated | more lines added than baseline |
| 25 | 378dd7f6feat(applicability)!: scope to opus 4.7+ and enforce model gate at proxy | teamchong | 2mo ago | +739 -18 | 92%Elevated | more lines added than baseline |
| 26 | 6cbdea58feat(proxy): SSE+JSON output measurement scanner | teamchong | 3mo ago | +551 -50 | 92%Elevated | more lines added than baseline |
| 27 | cde21b2efix(savings): base text-counterfactual warmth on wall-clock, not image-cache fate | teamchong | 1mo ago | +536 -170 | 90%Elevated | more lines added than baseline |
| 28 | 08a2b0a1feat(gpt): pin the live user request as text in history collapse | teamchong | 2mo ago | +426 -60 | 90%Elevated | more lines added than baseline |
| 29 | fb13de7cfeat(dashboard): per-session dollar-weighted savings panel with bucket + passthrough breakdown | teamchong | 2mo ago | +495 -37 | 90%Elevated | more lines added than baseline |
| 30 | 4774aeb1feat(eval): in-image instruction variant (reflow-inimage) — +1.04pp vs baseline | teamchong | 3mo ago | +460 -116 | 90%Elevated | more lines added than baseline |
| 31 | 0aa2cce8Expose pixelpipe library API | Steven Chong | 3mo ago | +566 -156 | 90%Elevated | more lines added than baseline |
| 32 | dfd3bf51feat(proxy): cache-aware baseline via second count_tokens probe | teamchong | 3mo ago | +468 -93 | 90%Elevated | more lines added than baseline |
| 33 | f1688e9ftransform+render: R2 multi-col default-on + per-request pixel/char instrumentation | teamchong | 3mo ago | +478 -35 | 90%Elevated | more lines added than baseline |
| 34 | ba13bf3adashboard: live HTML + JSON endpoints in the Node host | teamchong | 3mo ago | +491 -3 | 90%Elevated | more lines added than baseline |
| 35 | 3109a3b5cli: add 'pixelpipe stats' subcommand for offline log analysis | teamchong | 3mo ago | +489 -2 | 90%Elevated | more lines added than baseline |
| 36 | 1cb62c17feat(dashboard): context-map transparency, flexible model chips, Opus default-off | teamchong | 2mo ago | +424 -72 | 88%Elevated | more lines added than baseline |
| 37 | 514a4f0bfeat(gate): config-driven model scope (PXPIPE_MODELS) + Opus 4.8 re-test | teamchong | 2mo ago | +437 -12 | 88%Elevated | more lines added than baseline |
| 38 | 7d62a4c7eval(swe-bench-pro): expand to 19 pairs — ON 14/19 vs OFF 15/19, agree 18/19; navidrome 3/3 replication clears the ON loss | teamchong | 2mo ago | +508 -17 | 88%Elevated | more lines added than baseline |
| 39 | d65cc1bfchore(deps): bump wrangler 3.114.17 → 4.92.0 | teamchong | 3mo ago | +427 -513 | 88%Elevated | more lines added than baseline |
| 40 | f8902732feat(dashboard): real saved% from count_tokens, rip α-regression dead code | teamchong | 3mo ago | +422 -1.2K | 88%Elevated | more lines added than baseline |
| 41 | b49764e6docs: add TRANSFORM_INFO.md (compression strategy reference) | teamchong | 3mo ago | +334 -0 | 88%Elevated | more lines added than baseline |
| 42 | 50f3f1f8docs: add prompt-caching alignment + honest savings-math doc | teamchong | 2mo ago | +413 -3 | 86%Elevated | more lines added than baseline |
| 43 | 0277faccdocs: reframe savings as workload-dependent snapshots; add demo + render-sizing doc | teamchong | 2mo ago | +438 -13 | 86%Elevated | more lines added than baseline |
| 44 | c1d111a0feat(transform): per-bucket char attribution telemetry (Task #18 Phase 1) | teamchong | 3mo ago | +391 -0 | 86%Elevated | more lines added than baseline |
| 45 | b9c438b2host: wire FileTracker (Node) + JsonLogTracker (Worker) into per-request flow | teamchong | 3mo ago | +302 -8 | 86%Elevated | more lines added than baseline |
| 46 | e49bb118feat(telemetry): log stop_reason + safety-flag per request | teamchong | 1mo ago | +282 -11 | 84%Elevated | more lines added than baseline |
| 47 | bf49065dfix(savings): use observed cache state for text baseline | Steven Chong | 1mo ago | +375 -768 | 84%Elevated | more lines added than baseline |
| 48 | c340ac3dfeat(dashboard): direct compressed-vs-passthrough cost split | Steven Chong | 2mo ago | +272 -47 | 84%Elevated | more lines added than baseline |
| 49 | 0cfe08a8feat(gate): symmetric warm-cache burn + gateEval telemetry (anti-flapping) | Steven Chong | 3mo ago | +272 -14 | 84%Elevated | more lines added than baseline |
| 50 | f5496d6cfeat(dashboard): last-50 image ring, image viewer, cut prune + session detail | teamchong | 3mo ago | +313 -1.6K | 84%Elevated | more lines added than baseline |
Two views of the same model: where the cuts fall, and what commit shape lands you above them.
Every scored commit, binned on the raw 0 to 10 score rather than the percentile. Percentile ranks are uniform by construction, so that axis has no shape to draw. The dashed lines are the tercile cuts behind each row's priority pill.
The 200 most recent commits, on their own recency sample rather than the feed above: that defaults to risk-sorted, so reusing it would plot only the top tercile and call it the spread. Big and scattered is what the model penalises. Click a dot to open it.
Repowise tracks change history across 146 files in 2233admin/pxpipe. In the last 90 days 146 files were touched, 801 times in total, most often transform.ts. Every commit is scored for change risk from its size, spread and the history of the files it touches.
