fix(always-on): a resumed transcript pairs turn-1 attachments with a later turn

[pa-arth]

Review findings from #28 and #29, both real.

1. Resumed prefix used the wrong turn (#28, Greptile)

injectedOpen closes at the first main-chain assistant row — 34 lines BEFORE the
cross-file seen dedup — while computeAlwaysOn measures the prefix from
firstMain, the first RETAINED turn. On a resumed transcript those are different
rows: the closing row is a replay owned by an earlier file and gets dropped, so
firstMain lands on a later turn whose cacheRead has the whole replayed
conversation folded in. Turn-1-sized attributed against a much larger measured
INFLATES fixedPrefixTokens, and reconcile() cannot catch it because it only
fires on a NEGATIVE remainder. Inflation is the silent direction.

Closing before the dedup was deliberate and stays — it is what stops a whole
resumed session's attachments folding into "turn 1". The bug was the other half.

Fixed by recording which row closed the window and which row firstMain will
pick, and declining to measure the session when they differ — counted, not
silently dropped. standingContextTokens is untouched: a resumed turn really did
carry that much. Only the BREAKDOWN is unanswerable, and unmeasured now says so
with its own reason rather than borrowing the no-attachments one.

ZERO INSTANCES IN THE AUTHORING CORPUS: 0 of 483 sessions with attachments, of
926. Verified by same-instant runs of origin/main and this branch — 0 of 30
fields moved. So this fixes no number today; it closes a path that would have been
invisible when it did fire.

That also means the case cannot be recorded, and a suite running only real local
data would stay green with the guard deleted. It is built from RECORDED bytes
anyway: a resumed replay is "rows this parse has seen, then rows it has not", so
seeding seen from a prefix of a recorded transcript and re-parsing the whole of
it reproduces the shape exactly, with no invented rows.

Two recorder bugs surfaced doing that, both the fixture-encodes-an-impossible-shape
failure the recorder exists to prevent:

• message.id was filled with CONSTANT filler, so every assistant row shared an id
  and the parser merged them — a "two-turn" recording came back as one turn with one
  prefix. Identity keys are now a memoized bijection: distinctness and intentional
  sharing both survive, no real value is carried. The old comment claimed ids "key
  nothing across fixtures"; they key the dedup and the streamed-row merge.
• the recorder counted assistant ROWS, not turns. Now counts distinct message ids.

2. Corrected rates had no offline coverage (#29, Greptile)

pricingDrift degrades to pass when it cannot fetch — correct, and it means an
offline run proves nothing about the rows it just corrected. pricingPinned.test.ts
is the deterministic companion: drift asks whether the VENDOR moved, this asks
whether OUR table moved underneath us. Only rows with a reason to be pinned are in
it; pinning all 29 would train people to update the expectation to match the code.

3. A stale comment that would have caused the next mis-pricing

Writing that test failed it, on gpt-5.5-pro. The header claimed "the pro tiers
publish no cached-input rate" and are set to cachedInput == input. False since
gpt-5.4-pro/5.5-pro: both publish 3/M, the ordinary 10%, confirmed against
LiteLLM, and the table correctly carries it. A reader trusting that comment would
have "restored" them to equal-input and over-billed every cached token they read.
The real exception set is three rows and is now named with reasons rather than
derived from the word "pro" — gpt-5-pro and gpt-5.2-pro (genuinely no published
rate) and codex-mini-latest (genuinely 25%).

410/410 tests, typecheck and lint clean. Mutation-checked: reverting luna to the
launch tier, "fixing" gpt-5.5-pro to equal-input, and dropping the prefix-turn gate
each fail only their own test. The two-turn fixture asserts its own precondition
(turn 2's prefix 61,014 -> 61,863) so a future re-recording cannot quietly leave the
guard proving nothing.

Co-Authored-By: Claude Opus 5 (1M context) noreply@anthropic.com

🤖 Generated with Claude Code

[greptile-apps]

Greptile Summary

The PR prevents resumed Claude Code transcripts from pairing first-turn attachments with a later retained turn, adds deterministic pricing pins, and improves recorded multi-turn fixture identity handling.

• Tracks the row closing prefix collection and excludes resumed mismatches from always-on component medians.
• Adds a recorded two-turn resumed-transcript fixture and regression coverage.
• Adds offline assertions for selected OpenAI prices and corrects cached-input pricing documentation.
• Extends the fixture recorder to preserve distinct IDs and record multiple assistant turns.

Confidence Score: 4/5

The PR appears safe to merge, with a non-blocking fixture-recorder fidelity issue around tool-result identity references.

The runtime resumed-transcript gate and pricing coverage are coherent, but newly recorded fixtures break the equality between tool-call IDs and tool-result back-references, weakening tests that consume tool result timing or error metadata.

Files Needing Attention: scripts/record-injected-fixture.mjs

Important Files Changed

Filename	Overview
src/adapters/claudeCode.ts	Tracks whether collected attachments and the first retained main-chain usage belong to the same turn.
src/alwaysOn.ts	Excludes resumed prefix mismatches from component medians and reports a dedicated unmeasured reason.
src/injectedPrefix.ts	Extends injected-prefix metadata with the turn-alignment flag and a conservative default.
scripts/record-injected-fixture.mjs	Adds multi-turn recording and synthetic identity preservation, but omits the parser-relevant tool_use_id back-reference.
src/tests/injectedPrefix.test.ts	Reproduces resumed dedup from recorded bytes and verifies that mismatched prefix breakdowns are excluded.
src/tests/pricingPinned.test.ts	Adds deterministic offline checks for historically regressed pricing rows and cached-input exceptions.
src/vendor/pricing.ts	Corrects documentation of GPT-5 cached-input pricing exceptions without changing runtime rates.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Parse assistant row] --> B[Record prefix-closing turn key]
  B --> C{Row retained after dedup?}
  C -- No --> D[Continue to later rows]
  C -- Yes --> E[Record first retained main-turn key]
  E --> F{Keys match?}
  F -- Yes --> G[Measure injected-prefix breakdown]
  F -- No --> H[Exclude breakdown and report resumed reason]
  G --> I[Keep standing-context measurement]
  H --> I

Loading

_{Reviews (1): Last reviewed commit: "fix(always-on): a resumed transcript pai..." | Re-trigger Greptile}

[pa-arth]

Correction to the last commit message: the suite is 414/414, not 411 — the new join check is an it.each over the four injected-prefix-* fixtures, so it adds 4 tests, not 1. The commit body has the wrong count; amending it needs a force-push, which this machine's presence hook blocks while another session holds a claim, so it is corrected here and in the squash message rather than routed around.