Collapse old conversation history into rendered PNG sections so the model
reads a compact image instead of re-billed text, while preserving prompt
caching and tool-call behavior. Measures real vs compressed token/cost.
Core:
- GPT history collapse (openai-history.ts): append-only, o200k token-length
sectioning. Sections seal only at a tool-closed boundary (open call-id set
empty), so a function_call and its function_call_output never split across
the collapse cut. Fixes the OpenAI 400 "No tool call found for function
call output with call_id ..." that hit long Responses-API sessions.
- Anthropic cache contract (history.ts): append-only per-chunk rendering;
cache_control markers are preserved/moved, never added; chunk boundaries
align with caller marker seams for byte-stable prefix caching.
- GPT image budget (openai.ts): detail:'original' for gpt-5.x, flagship
vision-multiplier fix, patch cap; schema-strip preserves real descriptions.
- Savings accounting (openai-savings.ts): cached_tokens + vision-token basis.
Model scope (applicability.ts):
- Default imaged scope = claude-fable-5 + gpt-5.6.
- gpt-5.5 and claude-opus-4-8 stay opt-in: same pipeline, but they degrade
reading dense imaged history (gist drift), so silently imaging them by
default is wrong. Promotion is gated on an OCR/recall threshold.
Dashboard: GPT + Anthropic rendering, per-family model toggles, persisted
metrics, thumbnail-expired session UI, reflow/newline handling.
Tests: cache-alignment (GPT + Anthropic), history sectioning + tool-boundary
invariants, savings, dashboard, sessions/restart-restore. 452 passing.
The history image sits after the slab in prefix order, but the single cache
breakpoint was anchored on the slab. So the largest block — the ~141k-token
history image — only cached when the caller's roaming downstream marker happened
to land after it. When it didn't, the whole history image re-created at the 1.25x
cache_create rate every turn instead of amortizing into 0.1x reads. On a warm
session that turned a real compression win into a net loss (the "411% more"
the dashboard correctly surfaced).
Fix is pure relocation: move pxpipe's single breakpoint from the slab onto the
LAST history image (which sits after the slab), so slab + history cache as one
stable prefix — created once, then read. Marker count is unchanged: pxpipe still
never ADDS a marker (Task #21), it only moves the caller's existing one. When no
slab imaged, the history path keeps its own relocation.
- transform.ts: relocateAnchorToHistoryImage on the collapse-success branch
- history.test.ts: two old "history carries no marker" invariants upgraded to
assert the relocation (last history image carries it; exactly one marker total)
replay() rebuilt RecentRow from the JSONL but never set
session_saved_so_far_delta, img_id, or contextHistory — so after a restart
old rows showed "-" in Saved and had no Details link, while live rows did.
Live update() does all three; replay() now matches.
- replay() sets session_saved_so_far_delta (baselineInputEff - actualInputEff,
same cache-weighted math as live) so the Saved column survives a restart.
- replay() rebuilds the Context Map breakdown from persisted fields
(baseline_tokens, bucket_chars, image_count, usage) with a synthetic
nextImageId, and sets img_id/img_ids so the Details link resolves.
- The PNG ring is in-memory and not restorable; restored entries carry
imageIds:[] and a `restored` flag, and the Details panel shows
"thumbnails expired when the proxy restarted" instead of dead <img>s.
- +2 replay tests (restart-restore.test.ts).
376 tests pass; typecheck + build clean.
Claude-Session: https://claude.ai/code/session_01RV8uiMYNhrAdTUzJ9fkpcp
@napi-rs/canvas was moved dependencies -> devDependencies (edge-safe
packaging) but the lockfile still listed it under dependencies, so CI's
`pnpm install --frozen-lockfile` refused the mismatch. Regenerated with
--lockfile-only; no version changes.
Claude-Session: https://claude.ai/code/session_01RV8uiMYNhrAdTUzJ9fkpcp
The session hero computed (1 - sent / raw_count_tokens_baseline) — the same
cache-blind math just removed from the Details panel. It priced the all-text
baseline at full rate, ignoring that most of it would have been cheap
cache-reads, so a warm net-loss session showed a big "fewer tokens" win up top
while Details/Saved showed a loss. Two panels, one session, opposite signs.
- renderSessionSummaryFragment now reads baselineInputWeighted /
actualInputWeighted (the same cache-weighted pair as Details + the Saved
column) instead of rawActualTokens / rawBaselineTokens.
- Headline is input-only; dropped the output-on-both-sides lumping (output is
never "sent" and only dampened the %).
- Honest labels: "effective tokens … after normal cache discounts"; still no $
assumptions (0.1x/1.25x are token multipliers, not $/Mtok).
- types.ts: corrected stale field comments (raw fields are math-drawer only).
- +4 hero tests pinning direction to the weighted pair and that output can't
move the headline %.
374 tests pass; typecheck + build clean.
The Details panel divided the raw count_tokens baseline by sent tokens
(cache-blind), over-claiming "X% smaller" even when the cache-aware row
showed a net loss — the two panels could flatly contradict each other.
- dashboard.ts: feed baselineInputEff/actualInputEff/haveBaseline into
ContextMapData (same weighted pair the recent row uses); push after
the eff-tokens are computed so the breakdown matches.
- fragments.ts: headline divides weighted baseline by weighted actual,
says "bigger" on a net loss, with a clarifying sub-line; no savings
claim when the baseline probe did not resolve.
- +4 tests (context-map.test.ts) pinning headline direction to the row.
370 tests pass; typecheck + build clean.
- Generalize isPxpipeSupportedGptModel to the GPT-5 family (gpt-5, 5.x,
-mini/-nano), default-on; [variant] tags stripped.
- OpenAI render switches to the 768px portrait strip (GPT_STRIP_COLS=152)
so the API's mandatory shortest-side->768 downscale never shrinks our
5px glyphs; downscale-free in both tile and patch regimes.
- Add OpenAI vision-token cost model (resolveVisionCost/openAIVisionTokens)
and rewire the Chat Completions profitability gate to use it instead of
the Anthropic 750px/token model.
- Add transformOpenAIResponses for the /v1/responses (Responses API) shape
used by Codex: instructions + input items + flat tools -> rendered
input_image parts; wire it into the proxy (was passthrough before).
- Tests: +12 (gate family, vision-token math incl. downscale path, chat +
responses transforms). 366 passing; Anthropic path unchanged.
README intentionally not updated (GPT OCR fidelity unverified until tested).
- keepSharp: caller predicate pins exact-match-critical blocks (IDs, hashes,
paths) as text, overriding the chars/token heuristic. Reported via
info.keptSharpBlocks. Never forces imaging; throwing predicate = false.
- emitRecoverable: optional recovery map (block -> original text + provenance)
so a stateful caller can re-inject or re-fetch imaged content. The recover
half of the fidelity contract; documented mitigation for lossy recall.
- edge/Workers-safe: process.env read is typeof-guarded; @napi-rs/canvas moved
to devDependencies (atlas is build-time only, runtime PNG encoder is pure-JS).
- comments: trimmed ~1,640 lines of essay/derivation/changelog narration across
src/ (transform.ts -73%), keeping one-line rationale; deep math points to docs/.
- New exports/types in index.ts + library.ts; +12 tests (keep-sharp, recoverable).
Build clean; 354/354 tests pass.
Tokens are the durable invariant; the dollar figure is a pricing-dependent
consequence that evaporates if Anthropic reprices.
- README: headline + intro now lead with input-token reduction; the
~59-70% end-to-end bill is demoted to a downstream effect 'at current
Fable list prices', with the dollar math pointed to the FAQ
- dashboard: stop claiming Claude's replies are 'never compressed' (false
for any session past keepTail) — scope to latest messages + live output
Captures why imaging conversation history stays cache-safe as a session
grows: the quantized collapseChunk boundary, the one-time cache-create
burn + amortization gate, the protectedPrefix slab anchor, and the
relocate-don't-add cache_control rule. Grounded in src/core/history.ts
and transform.ts; prose mirror of the verified code.
server.close() waits for open connections to drain, so it never finishes while
the dashboard tab polls every 2s or while an SSE response is mid-stream - the
first Ctrl+C printed 'shutting down' but the process never exited.
Drop idle keep-alive sockets immediately, force-close anything still streaming
after a 1.5s grace, and make a second Ctrl+C an instant hard exit.
Full rewrite of src/dashboard/fragments.ts (CSS + render functions + page
shell). Server JSON contracts, htmx endpoints, element IDs, and test-pinned
strings are all preserved.
- Light theme with flame-orange accent (no dark mode)
- One plain-language hero answer on top; all cc=1.25 + cr*0.10 formulas and
counterfactual diagnostics moved behind a 'Show the math' drawer
- Image vs Text Breakdown panel: 'Became an image / Stayed as text' tabs plus
an Image + Source inspector showing the verbatim text behind each page
- Recent table gains a color-coded 'Sent as: image/text' column
- Wide tables (#frag-recent, #frag-stats) and the latest-image inspector
(#frag-latest) scroll inside their card instead of pushing the border out
eval:
- Add eval/lib/cci.py: drop-in interactive Claude session shim (pexpect/PTY)
replacing headless `claude -p`. Drives the real TUI on Max-subscription auth
(no ANTHROPIC_API_KEY), scrapes the reply + /context + /cost, and is
argv/stdin/--output-format json compatible with the old call sites.
- Wire every call site to cci.py: gist-recall (run/run2/run3),
needle-haystack (sweep/crux), gsm8k (bench), swe-bench (run_pilot),
swe-bench-pro (rep_nav/run_pro), and eval/lib/anthropic-client.mjs.
- Isolation per call: --setting-sources project --strict-mcp-config
--dangerously-skip-permissions. Recall evals keep --allowedTools Read;
swe-bench keeps the full agentic toolset. Each trial is a fresh process.
demo:
- Resolve the `opus` shorthand to the full claude-opus-4-8[1m] model in
cost-ab and effective-context a.sh/b.sh. setup.sh intentionally keeps the
base id: the proxy scope gate strips [1m] from the request before matching,
so [1m]-on-launch + base-in-scope is the correct pairing.
docs:
- FINDINGS.md: drop the 2026-06-16 Opus 4.8 re-test sections.
Demo clip moved to Google Drive (committed copy was too low-res); README keeps the preview thumbnail linking to the Drive video. Restoring the thumbnail also un-breaks the demo image on the npm 0.3.0 page. Version 0.3.0 -> 0.3.1 + CHANGELOG.
Attach the recorded plain-vs-pxpipe demo (cost-ab fix + effective-context recall) plus a poster thumbnail, embedded in a new Demo section with an honest caption (Opus opt-in; demo 2 shows pxpipe surfacing the lossy-count limit rather than fabricating). Fable headline clip to follow when Fable is back.
The library-use constants table still showed the pre-0.3.0 values (DENSE_CONTENT_CHARS_PER_IMAGE 5000, MAX_HEIGHT_PX 1568), contradicting the 'how it works' section and the shipped code. Updated to 1932 / 92160 + DENSE_CONTENT_COLS 384; test count 323->342; 'Fable-5 only' -> 'Fable-5 by default, Opus opt-in'.
- Remove eval/corpus/{sessions,text-blocks}.json from the repo — they are real captured transcripts (file paths, project names, message content). gitignore eval/corpus/; local copies kept for the eval scripts.
- gen_sweep.mjs: replace a hardcoded /Users/<name> absolute import with a relative path.
- README: 'holds ~92k readable chars' -> 'up to ~92k chars' (don't overclaim legibility at the new page density).
- CHANGELOG: note OCR legibility at the larger page size is not independently re-eval'd.
The real-corpus reflow check read whole transcript files; on a large ~/.claude/projects (multi-hundred-MB files) that blew the 30s timeout and aborted 'pnpm test' (hence prepublishOnly). Read only the first 8MB per file — enough for the per-file sample; a truncated trailing line just fails JSON.parse and is skipped. Corpus test now runs in ~5s.
setup.sh resolves a model arg (default fable) and sets PXPIPE_MODELS to Fable plus the chosen model, so the demo matches the production default (Opus OFF) and 'setup.sh opus' enables Opus end-to-end across setup/a.sh/b.sh. READMEs updated to pass the model to all three.
Both demos default to Fable; pass a model as the first arg (a.sh opus, sonnet,
haiku, or a full claude-... id) to override -- run both columns on the same
model for a fair A/B. Adds eval/ab token-savings scripts.
- "How your context works" panel: per-request token flow (as-text -> real) plus
the exact-char bucket breakdown of what became images and a gallery of every
rendered page, opened via a "view" link on each recent-requests row.
- "compress models" chips are the union of a catalog (Fable 5, Opus 4.8/4.7,
Sonnet 4.6, Haiku 4.5), the PXPIPE_MODELS env scope, and the active scope, so
any env-enabled model stays toggleable (off <-> on); runtime-only override.
- Opus is OFF by default (Fable-5-only scope); opt-in via env or the chips.
Fixes from the code review:
- contextHistory cap aligned to the recent table (was 30 < 50, so older "view"
links silently showed the latest request's breakdown); an evicted or
unrecorded request now shows an explicit "no longer available" message.
- the session headline phrases a net loss as "N% more tokens" instead of
rendering a garbled "-7% fewer tokens".
Per-turn and per-session savings are the real baseline_eff - actual_eff with no
floor: a net-losing turn (cache_create-heavy image rewrite that costs more than
the text prefix it replaced) now reports the real loss instead of a fabricated
0. Updates baseline.ts and the caller comments that claimed a >=0 clamp the code
never applied -- the negative is intended and covered by tests/sessions.test.ts.
Fable 5 / Opus 4.8 accept up to 2576px / 4784 visual tokens per image, but a
request with >20 images (pxpipe always sends many) is held to the stricter
<=2000px/side rule, so the real ceiling is ~1932x1932 (1928x1928 = 69x69 = 4761
tokens). MAX_HEIGHT_PX 1568->1932; dense tool/history pages now render at
DENSE_CONTENT_COLS=384 / DENSE_CONTENT_CHARS_PER_IMAGE=92160 (1928x1928 full
page) -- fewer image blocks at the same OCR-validated 5x8 cell. The static slab
is unchanged (313 cols / 1573x1280).
Also fixes a regression the code review caught: the break-even gate,
truncateForBudget, and estimateImageCount predicted against the slab geometry
(313 cols / 159 rows) while the dense renderer emits 384 cols / 240 rows, so
large tool_results were truncated far earlier than the 10-image cap required --
silently dropping output that would have rendered. A per-page char budget is
threaded through the prediction chain (defaults to READABLE so the slab and unit
tests are byte-identical) plus a denseGateGeometry helper, so the gate prices the
same page the renderer actually produces.
- applicability: isPxpipeSupportedModel reads PXPIPE_MODELS (comma-sep,
default claude-fable-5 only) and strips bracket/variant tags like [1m];
Opus becomes opt-in with no code change. Default behavior unchanged.
- FINDINGS: Opus 4.8 re-test - arithmetic 98/100 (-2pp), verbatim 6/15
(was 93/100, 0/15): improved but still taxed -> default stays Fable-only.
- FINDINGS + eval/glyph-matrix/sweep: root cause of the read tax is per-glyph
RESOLUTION, not font shape (confusions scale with px and vanish by 10x16).
Opus needs ~4x glyph area to read reliably; the 1568px ceiling makes that
break-even, so there is no profitable Opus operating point.
New docs/CACHING_AND_SAVINGS.md explains the two things that keep coming up:
1. How pxpipe stays aligned with Anthropic's prompt cache when it images the
static prefix: split static/dynamic blocks first (keep drifting <env> etc.
as cheap text, never imaged), then relocate (never add) the caller's
cache_control onto the last image so the breakpoint anchors at the end of
the stable slab. Covers the cache-friendly splice, the symmetric
anti-flapping burn term, and the horizon-amortized history-collapse gate.
2. How savings are measured without double-counting the cache discount: both
the real request and a count_tokens counterfactual of the original body are
priced with identical cache rates at the same moment, so the discount
cancels in the difference and only the token reduction survives.
Also reconcile the priorWarmTokens JSDoc in transform.ts: the burn term is
applied undivided, matching all three gate call sites and the symmetric
priorWarmImageTokens doc. The '/ N' was stale (the per-turn gate has no
horizon to divide by; the horizon scales the lifetime cost, not the burn).
- generate.mjs builds one large context (filler logs + small state-tracking needle)
- README: two-column run (plain vs pxpipe), same prompt, compare answers + /context
- drop the e-commerce /context demo; context/ is generated, gitignored
- FINDINGS.md: dated 2026-06-12 entry logging in-the-wild verbatim
misreads (7/10 exact, ~3.3% per-char, 2/3 silent) alongside flawless
gist-tier recall in the same session. Explicitly anecdotal, n=1,
non-reproducible — motivation only.
- eval/glyph-matrix/PLAN.md: public plan for the paused confusion-matrix
experiment (Task #7) so the pending work is visible outside README,
per the measured-numbers-only README rule.
- ON 47823 / OFF 47824 with own PXPIPE_LOG files: operator session can
never pollute bench measurement (separation by construction)
- resume = skip existing patch files; clean stop on quota errors
- Docker grading documented as proxy-free (no model calls in containers)
- README: mention Pro pilot alongside Lite
teamchong