GeezerVison

Claude Code reports compaction work in cumulative per-model totals even when it omits that work from the turn's usage. This change tracks those totals per process and adds positive differences to usage for turns that receive a compaction boundary. Discarded results advance the baseline to avoid attributing their work to a later turn; public usage events and stored schemas remain unchanged.

Changes

Sequence diagram

sequenceDiagram
    participant CC as Claude Code
    participant E as Engine connection
    participant M as Turn mapper
    participant K as Usage consumer
    CC->>E: compact_boundary
    E->>M: accept(boundary)
    CC->>E: result(usage, modelUsage)
    alt Active turn
        E->>M: accept(result), shared process mark
        M->>M: Compare cumulative totals and update mark
        M->>K: usage including eligible compaction gap
    else Discarded result
        E->>E: markProcessSpend(result)
    end

• Files reviewed: 4 · 83 changed lines · estimated review effort 2/5, about 15 minutes
• Profile: normal · at most 8 inline comments
• Verification: every finding was checked again by a separate pass (1 candidate judged)
• Cost: 324,047 tokens in (221,312 from the provider's cache) · 2,902 out
• Readers: 3 files went to readers instead of the review pass whole: 3 code files by gpt-6-luna (the review pass gets their summaries and the hunks of their risk spots; 1 call, up to 3 at once, 10,308 tokens counted). Their summaries are hints, not findings. 1 code file no reader could read went to the review pass whole (the reader gave no usable answer for this file).
• Stacked on #128: only this pull request's own commits since #128's head 2c95e59.

Performance (perf-attrib, base 2c95e59 against head 1052236)

perf-attrib: igual: nothing above the measured noise

base 2c95e59b7f3a05988396170baea350d411e8a2a4 (2c95e59b) vs cand 10522364eda5d9ff971a71e1f92a8c38dc72d406 (10522364), median of 2 faux-engine session(s) per side; flagged = beyond the noise measured on 18 base-vs-base runs (x1.5).

JS object types that rose (count):

• tui.ready.heap.type.Function: 43276 -> 43281 (+5)
• tui.ready.heap.type.FunctionExecutable: 14159 -> 14164 (+5)
• tui.ready.heap.type.UnlinkedFunctionExecutable: 14146 -> 14151 (+5)
• tui.after.heap.type.Function: 44341 -> 44344 (+3)

JS heap bytes by constructor that rose:

• tui.rest.ctor.object:ModuleRecord.bytes: 5501493 -> 5502260 (+767)
• tui.rest.ctor.native:Blob.bytes: 1738720 -> 1738819 (+99)
• tui.rest.ctor.closure:Function.bytes: 841759 -> 841791 (+32)
• tui.rest.ctor.closure:f.bytes: 70496 -> 70528 (+32)

Modules loaded at start: 230 -> 230, 4629972 -> 4630739 bytes of chunks; chunk sizes only in cand: [1006590, 275870, 109786, 27972, 19371], only in base: [1006596, 275864, 109774, 27978, 18610] (chunk names are hashes: the list says which changed in size, not by what name)

Reference (skill, 2026-11-03, real model): footprint 64 ready / 66 after / peak 80 MB, 2.6 KB per session. This run (faux engine, whole process tree): 55 / 57.5 / peak 73.5 MB, 2055 B per session.