yhh 07d2d4bdd0 Show the cache hit rate after the cold start, and the output ratio ...

The session's first executor request pays for the prefix every later one
reuses, so the session-wide hit rate mixes "did the cache work once warm"
with "how much a cold start could reuse". usagereport already records the
cold start, but nothing a live session reads projected it.

The live figures come from the serve ledger, which restarts on /new,
/resume and /fork; usageTally does not, so its ColdStart can belong to an
earlier conversation in the same process. The broadcaster therefore takes
the cold start from the ledger's own population: the first request the
ledger files under the executor (pricing.LedgerSource, the rule Add
already used), split by the same CacheSplit the ledger applies, and reset
with it. SessionUsage carries it as coldStart.

- /status (TUI): "warm cache" row, ledger hit rate with the cold start
  taken out of hit, miss and unknown; "out/in" row, output ÷ input.
- Desktop session overview behind the cache meter: the same two figures,
  computed in state/overview beside cacheRate.
- wire-parity: ColdStartUsage mirrored in port/session.ts.

2026-11-07 09:06:37 +00:00

..

assembly/boot

fix(ci): the Windows shell listing joins paths the way the OS it is computed for does, the save allocation check is skipped under the race detector that defeats regexp's pool, and the context-retrieval benchmark whose premise archived folds ended is removed with the surface helper only it used

2026-11-07 00:39:03 -08:00

base

Show the cache hit rate after the cold start, and the output ratio

2026-11-07 09:06:37 +00:00

contract

Show the cache hit rate after the cold start, and the output ratio

2026-11-07 09:06:37 +00:00

ext

test: skip symlink fixtures where the platform cannot build symlinks

2026-11-03 15:26:16 -08:00

frontend

Show the cache hit rate after the cold start, and the output ratio

2026-11-07 09:06:37 +00:00

model

refactor(provider): remove the summary request flag and what only it served: the Request.Summary field and ErrSummaryToolsUnsupported, the adapters' no-tools branches, the broker's capability probe, and the test doubles and live test written for a summarizer call that no fold makes; the reported-usage effect test now detects a fold by its archive pointer and has a control arm

2026-11-06 23:47:31 -08:00

platform

fix(ci): the Windows shell listing joins paths the way the OS it is computed for does, the save allocation check is skipped under the race detector that defeats regexp's pool, and the context-retrieval benchmark whose premise archived folds ended is removed with the surface helper only it used

2026-11-07 00:39:03 -08:00

runtime

refactor(provider): remove the summary request flag and what only it served: the Request.Summary field and ErrSummaryToolsUnsupported, the adapters' no-tools branches, the broker's capability probe, and the test doubles and live test written for a summarizer call that no fold makes; the reported-usage effect test now detects a fold by its archive pointer and has a control arm

2026-11-06 23:47:31 -08:00

safety

refactor(delegation): replace the task family with spawn_agents

2026-11-02 08:52:45 -08:00

session/control

fix(control): in yolo mode remember no longer asks for approval either, so neither memory tool prompts under the mode that skips every other approval

2026-11-06 21:13:36 -08:00

state

fix(ci): the Windows shell listing joins paths the way the OS it is computed for does, the save allocation check is skipped under the race detector that defeats regexp's pool, and the context-retrieval benchmark whose premise archived folds ended is removed with the surface helper only it used

2026-11-07 00:39:03 -08:00

tools

fix(tool): a refusal's remedy names the setting tool by its capability id, tool:setting, now that the tool sits behind use_capability and a bare "setting" is not a name the model can see

2026-11-06 22:51:42 -08:00