Commit Graph

  • 169521c9f8 fix(transform): defuse reasoning_extraction refusal — provenance-framed slab banner teamchong 2026-07-01 22:25:08 -04:00
  • 7f2e40bad6 Revert "feat(transform): PXPIPE_TEXT_SLAB=1 keeps the system slab as text (classifier-safe)" teamchong 2026-07-01 21:50:39 -04:00
  • c04e8f80e8 feat(transform): PXPIPE_TEXT_SLAB=1 keeps the system slab as text (classifier-safe) teamchong 2026-07-01 19:46:29 -04:00
  • ed968a7495 Reapply "fix(render): fit Anthropic page to 1568×728 (~1.15 MP) for WYSIWYG glyphs" teamchong 2026-07-01 19:18:20 -04:00
  • 6eaa8bfcd0 Revert "fix(render): fit Anthropic page to 1568×728 (~1.15 MP) for WYSIWYG glyphs" teamchong 2026-07-01 18:50:25 -04:00
  • 9cade86868 debug(node): add PXPIPE_DUMP_DIR to persist rendered PNGs for demo inspection teamchong 2026-07-01 18:50:25 -04:00
  • ad947335e5 demo: run fable arm on claude-fable-5[1m] (1M ctx), matching opus/sonnet teamchong 2026-07-01 18:04:38 -04:00
  • 838db8778e fix(render): fit Anthropic page to 1568×728 (~1.15 MP) for WYSIWYG glyphs teamchong 2026-07-01 17:54:51 -04:00
  • d7999279a5 eval: add constant-cost render-style A/B harness (paused task #7 arms) teamchong 2026-07-01 16:41:24 -04:00
  • b453adf0c1 docs(node): pin why the transform hook is only a kill switch teamchong 2026-07-01 16:41:12 -04:00
  • edf7a05f85 feat(dashboard): tag one-time cache-create losses in recent table teamchong 2026-07-01 16:41:12 -04:00
  • a346ec7ac5 feat(factsheet): carry occurrence counts, match ticket-style codes teamchong 2026-07-01 16:41:00 -04:00
  • d1704241d6 demo: add claude-sonnet-5 support alongside opus teamchong 2026-06-30 15:22:20 -04:00
  • f472566cc1 release: v0.6.10 v0.6.10 teamchong 2026-06-30 12:45:08 -04:00
  • bf49065dce fix(savings): use observed cache state for text baseline Steven Chong 2026-06-30 12:24:09 -04:00
  • ff8a05e30c fix(savings): gate text-counterfactual warmth on static-prefix hash Steven Chong 2026-06-29 15:59:10 -04:00
  • ec43bf5370 release: v0.6.9 v0.6.9 teamchong 2026-06-27 23:41:11 -04:00
  • 966f9a144e refactor(export): single-source REPORT_CHARS_PER_TOKEN in transform.ts teamchong 2026-06-27 23:41:11 -04:00
  • cde21b2e73 fix(savings): base text-counterfactual warmth on wall-clock, not image-cache fate teamchong 2026-06-27 21:26:55 -04:00
  • 76a371028a feat(export): replace in-band @pxpipe directive with explicit pxpipe export CLI teamchong 2026-06-27 18:28:28 -04:00
  • ecea71d1a0 dashboard: order Sonnet newest-first (Sonnet 5 before 4.6) teamchong 2026-06-26 00:58:53 -04:00
  • 9f25a4db3b ci: move actions to Node 24 majors (checkout@v7, setup-node@v6, pnpm/action-setup@v6) teamchong 2026-06-26 00:24:58 -04:00
  • 873cbaf94f release: v0.6.8 v0.6.8 teamchong 2026-06-26 00:17:10 -04:00
  • 6d1c46dd20 fix(cache): pin prefix-cache anchor to the byte-stable history image teamchong 2026-06-26 00:17:10 -04:00
  • 6ca72260e8 chore: release 0.6.7 v0.6.7 teamchong 2026-06-25 22:15:20 -04:00
  • 85625f6a7a feat(transform): cache_prefix_sha8 prefix-cache bust attribution probe (#11) teamchong 2026-06-25 22:14:37 -04:00
  • 7d15271587 chore: release 0.6.6 v0.6.6 teamchong 2026-06-25 19:59:37 -04:00
  • e9cc56656a feat(dashboard): dark/light mode toggle teamchong 2026-06-25 19:58:54 -04:00
  • 2e8f89288b fix(factsheet): rank tokens by shape so short high-consequence ids survive teamchong 2026-06-25 19:58:54 -04:00
  • 9d121acc20 fix(node): backpressure + graceful client-abort handling in response streaming Steven Chong 2026-06-25 17:28:33 -04:00
  • f1cc22b8b6 chore: release 0.6.5 v0.6.5 teamchong 2026-06-25 17:23:54 -04:00
  • 78abd081e4 feat(factsheet): byte-stable fact-sheet of exact identifiers on Claude + OpenAI paths teamchong 2026-06-25 17:23:03 -04:00
  • 45d9d57ce5 feat(dashboard): add Sonnet 4.6 and Sonnet 5 chips to image-model catalog teamchong 2026-06-25 17:23:03 -04:00
  • 3d3b31e66f docs(findings): reframe — correct baseline is /compact, not verbatim recall teamchong 2026-06-23 22:58:08 -04:00
  • 8417380061 test: e2e guards for cache stability, dashboard honesty, and design behaviors teamchong 2026-06-23 12:30:16 -04:00
  • c75179b50f chore: release 0.6.4 v0.6.4 teamchong 2026-06-23 06:45:39 -04:00
  • 9eb66f54f2 fix(history): anchor collapsed history recency Steven Chong 2026-06-23 06:37:44 -04:00
  • 85f8536b67 chore: release 0.6.3 v0.6.3 teamchong 2026-06-22 16:25:02 -04:00
  • 08a2b0a118 feat(gpt): pin the live user request as text in history collapse teamchong 2026-06-22 12:15:16 -04:00
  • 681f40fd5b chore: release 0.6.2 v0.6.2 teamchong 2026-06-22 11:07:24 -04:00
  • 22d89ee37d fix(sessions): complete honest cache-math warmth gate teamchong 2026-06-22 11:07:02 -04:00
  • b9418bfa6f fix(gpt): cap history images and honest cache math Steven Chong 2026-06-22 10:45:39 -04:00
  • 023669dec4 chore: release 0.6.1 v0.6.1 teamchong 2026-06-21 17:43:03 -04:00
  • f6be3d3f88 fix(openai): keep opening prompt collapsible Steven Chong 2026-06-21 17:37:51 -04:00
  • 62b9c872ac chore: release 0.6.0 v0.6.0 teamchong 2026-06-21 17:17:56 -04:00
  • fe6b0ec2a8 feat: structure-through role attribution + turn-index recency anchor for history images teamchong 2026-06-21 17:17:08 -04:00
  • dbc6d895f2 chore: release 0.5.0 v0.5.0 teamchong 2026-06-20 21:16:47 -04:00
  • 47b5e412b3 feat: cache-stable history-collapse imaging for GPT (OpenAI) + Anthropic Steven Chong 2026-06-19 16:02:49 -04:00
  • c44cdf4258 chore: release 0.4.0 v0.4.0 teamchong 2026-06-19 16:53:11 -04:00
  • 8b15525cb9 fix(transform): relocate cache anchor onto history image (stop 1.25x re-create) teamchong 2026-06-19 16:39:46 -04:00
  • b20c976f61 fix(dashboard): restore Saved + Details on replayed rows after restart teamchong 2026-06-19 16:06:59 -04:00
  • ad526fce31 fix(ci): regenerate pnpm-lock.yaml after canvas moved to devDependencies teamchong 2026-06-19 16:01:38 -04:00
  • 48f82ffee2 fix(dashboard): hero uses cache-weighted tokens (match Details/Saved) teamchong 2026-06-19 15:58:22 -04:00
  • c899d7d836 fix(dashboard): Details headline uses cache-weighted tokens (match row) teamchong 2026-06-19 14:38:17 -04:00
  • 97d50513aa feat(openai): GPT-5.x family + Responses API support teamchong 2026-06-19 13:47:44 -04:00
  • ef9ddac58a feat(library): caller fidelity + recoverable channel, edge-safe; trim comments teamchong 2026-06-19 13:02:30 -04:00
  • 8eb035025a docs: lead messaging with token reduction, not dollar savings teamchong 2026-06-19 11:26:39 -04:00
  • cf4188625a docs: add HISTORY_CACHE_MODEL.md, conceptual companion to caching/savings teamchong 2026-06-19 11:26:39 -04:00
  • e5f5bc3d31 fix(node): exit cleanly on Ctrl+C during in-flight streams and idle keep-alive teamchong 2026-06-19 09:34:35 -04:00
  • 13456812b1 redesign dashboard: light flame theme, plain-language hero, image/text transparency teamchong 2026-06-19 09:34:35 -04:00
  • de3f756f16 deps: patch 6 dependabot advisories (vite, undici, esbuild) teamchong 2026-06-18 23:47:54 -04:00
  • 7af6c200c3 eval+demo: route harnesses through interactive cci.py; opus → claude-opus-4-8[1m] teamchong 2026-06-18 23:43:37 -04:00
  • 74e2b97b1b docs: exact-recall escape hatch — route byte-exact subtasks to a non-Fable subagent teamchong 2026-06-18 21:34:03 -04:00
  • af1e838f0f chore: release 0.3.1 — Drive-hosted demo, keep preview thumbnail teamchong 2026-06-17 15:42:42 -04:00
  • 85aed799e5 docs(readme): drop blurry committed demo video, link to Drive instead teamchong 2026-06-17 15:15:10 -04:00
  • a9ff92d3e1 docs(readme): add side-by-side A/B demo video (Opus 4.8) teamchong 2026-06-17 13:02:37 -04:00
  • fab35ed2d2 docs(readme): fix stale constants table + test count; note Opus opt-in teamchong 2026-06-17 12:21:48 -04:00
  • d2d410ffc8 chore: scrub private data from repo; honest doc tweaks teamchong 2026-06-17 12:08:16 -04:00
  • 716d9e0965 test: cap corpus L0 test to an 8MB per-file prefix teamchong 2026-06-17 10:44:46 -04:00
  • dadee3b09e feat(demo): setup.sh takes a model arg; default Fable-only (Opus off) teamchong 2026-06-17 10:07:32 -04:00
  • a64768bc4e chore: release 0.3.0 teamchong 2026-06-17 10:01:33 -04:00
  • ad7ef7c1ac docs(demo): cost-ab + effective-context A/B demos with a model arg teamchong 2026-06-17 10:01:33 -04:00
  • 1cb62c1786 feat(dashboard): context-map transparency, flexible model chips, Opus default-off teamchong 2026-06-17 10:01:33 -04:00
  • a9bda51210 feat(savings): honest negative accounting; drop the false >=0 clamp claim teamchong 2026-06-17 10:01:33 -04:00
  • 8af9829872 feat(render): raise page ceiling to ~1932x1932 and align gate/paging geometry teamchong 2026-06-17 10:01:33 -04:00
  • 514a4f0b74 feat(gate): config-driven model scope (PXPIPE_MODELS) + Opus 4.8 re-test teamchong 2026-06-16 12:22:55 -04:00
  • 50f3f1f86b docs: add prompt-caching alignment + honest savings-math doc teamchong 2026-06-16 08:35:48 -04:00
  • be85f58d90 demo: pivot to smart-zone demo (plain Claude rots at large context, pxpipe stays sharp) teamchong 2026-06-16 08:18:21 -04:00
  • 0277facc5d docs: reframe savings as workload-dependent snapshots; add demo + render-sizing doc teamchong 2026-06-15 22:33:31 -04:00
  • 7ee263c596 docs: drop glyph-matrix resume date (reader is fable-5, currently unavailable) teamchong 2026-06-14 17:27:35 -04:00
  • a10837dcdb docs: add roadmap - glyph legibility, effective context, context-rot hypotheses (all unmeasured) teamchong 2026-06-14 17:24:45 -04:00
  • 0ec9efe9ca docs: FINDINGS field observation (n=1) + glyph-matrix PLAN.md teamchong 2026-06-12 13:25:51 -04:00
  • 958b17357b Revert "docs: troubleshooting note for minimum-release-age install failures" teamchong 2026-06-12 06:23:21 -04:00
  • c865520657 docs: troubleshooting note for minimum-release-age install failures teamchong 2026-06-12 06:22:20 -04:00
  • 8bc1478406 v0.2.1: README scope corrections (docs-only) v0.2.1 teamchong 2026-06-12 05:55:28 -04:00
  • 69eb96d17e docs: README scope wording - imaging covers system prompt + tool docs, not just old history teamchong 2026-06-12 05:40:57 -04:00
  • 5704c84237 docs: add real-world failure FAQ; scrub usernames/home paths from eval corpus teamchong 2026-06-11 17:15:42 -04:00
  • 62790e7371 docs: add FAQ - end-to-end bill math vs touched-only, measurement method, compression scope teamchong 2026-06-11 16:53:18 -04:00
  • d990faf646 eval(verbatim): expand to n=15 dense hex recall on Fable 5 - 13/15 vs Opus 0/15, both misses single-glyph confabulations teamchong 2026-06-11 16:48:12 -04:00
  • 7d62a4c715 eval(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 2026-06-11 15:47:54 -04:00
  • 156945977e eval(swe-bench-pro): 10-pair bench - ON 6/9 vs OFF 7/9, -61% per-request, receipts committed teamchong 2026-06-11 10:43:02 -04:00
  • 4eef1ec1a0 fix(eval/swe-bench-pro): verified proxy startup + atomic repo cache clone teamchong 2026-06-11 08:09:48 -04:00
  • b8e1e855bc feat(eval/swe-bench-pro): resumable 10-pair runner on bench-dedicated proxies teamchong 2026-06-11 08:05:06 -04:00
  • e98bf36387 fix(eval/swe-bench-pro): deconfound ON-arm numbers from operator session sharing port 47821 teamchong 2026-06-11 05:00:01 -04:00
  • 78b2247a92 eval(swe-bench-pro): single-pair pilot - 1/1 both arms, -64% per-request, Rosetta gotcha documented teamchong 2026-06-11 04:56:06 -04:00
  • 8e8cb09223 docs: SWE-bench tables lead with per-request -65%, demote confounded run totals teamchong 2026-06-10 23:49:38 -04:00
  • 83764a18a0 docs: surface SWE-bench parity in honest-part section, link eval dir teamchong 2026-06-10 23:31:55 -04:00
  • 76542b4a3c eval(swe-bench): Lite pilot 10 paired instances — 10/10 resolved both arms, ON $27.27 vs OFF $53.61 (-49%) teamchong 2026-06-10 23:25:24 -04:00
  • 5274e74bd4 docs: lead with whole-bill savings (59%, all 13.7k requests) over touched-only 72% teamchong 2026-06-10 19:08:32 -04:00
  • c6a0d61f00 eval(gist-recall): A/B imaged vs text history — 98/98 both arms, 0 confabulation teamchong 2026-06-10 18:54:03 -04:00