diff options
| author | Adam Malczewski <[email protected]> | 2026-06-05 17:53:00 +0900 |
|---|---|---|
| committer | Adam Malczewski <[email protected]> | 2026-06-05 17:53:00 +0900 |
| commit | 8c417472e7801369c3dfd004c9c85d7d69372f7c (patch) | |
| tree | 3da8eac532855a14013b502afaf30d669010c6be /packages/kernel/src | |
| parent | 74986a54093f6492cc4420f5917e5215f42a8f89 (diff) | |
| download | dispatch-8c417472e7801369c3dfd004c9c85d7d69372f7c.tar.gz dispatch-8c417472e7801369c3dfd004c9c85d7d69372f7c.zip | |
feat(observability): map DeepSeek nested cache tokens (prompt_tokens_details.cached_tokens) -> Usage.cacheReadTokens
The real flash fixture showed flash reports cache usage in the NESTED prompt_tokens_details.cached_tokens form (384 cached of 665 prompt); the parser only mapped the flat cache_read_tokens form, so cache tokens never surfaced. Now: cacheReadTokens = usage.cache_read_tokens ?? usage.prompt_tokens_details?.cached_tokens (flat wins; cacheWriteTokens flat-only, never fabricated; partial/null *_details safe). No kernel contract change (Usage already has the fields). +5 parser tests + a real-fixture regression (cacheReadTokens === 384).
These counts (+ a future prefix.fingerprint) are the cheap signals for body de-duplication. The broader trace-body storage-growth concern (verbatim body stored per request -> ~O(N^2) for long conversations) is logged DEFERRED in tasks.md; mitigation already designed (D5 volume control + ยง6 retention/rotation), not yet built. 339 tests, typecheck + biome 0/0.
Diffstat (limited to 'packages/kernel/src')
0 files changed, 0 insertions, 0 deletions