Krzysztof kuhy Rudnicki c5cd63fb80 token_audit: measure tool-call batching per API message ... Counting transcript records reports 1.00 tool calls per message no matter what the model did: the harness writes one JSONL record per tool_use block, all sharing message.id. A hand-rolled per-record count read the real 1.9% batching rate as 0.0%, which nearly shipped as a headline finding. Group by message.id instead, and report it, so the number comes from a tested code path with the window handled by find_transcripts rather than an ad-hoc timestamp. Three snapshot fields give next week's run a baseline. Axes was at the 12-attribute cap, so the counters live in their own Batching dataclass with the two ratios as properties. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NiSSDu4gKbp9GGxTTtXvYW		2026-08-28 12:22:21 +02:00
..
archive-billsplit-2026-08-23.json	Archive billsplit and everything coupled to it	2026-08-23 17:17:22 +02:00
archive-bucket-catch-poker-stakes-2026-08-23.json	Archive bucket_catch and poker-stakes to testsAndMisc-archive	2026-08-23 17:12:54 +02:00
ci-mirror-scoped-push-gate-2026-08-23.json	Make the push gate check the push, and make its pytest stage real	2026-08-23 15:12:23 +02:00
endurain-db-dedupe-2026-08-23.json	Dedupe endurain imports against postgres when the API cannot answer	2026-08-23 15:23:08 +02:00
endurain-import-dedupe-and-adb-serial-2026-08-23.json	Stop the endurain importer duplicating RunnerUp's own uploads	2026-08-23 14:36:03 +02:00
endurain-runnerup-webdav-2026-08-23.json	feat(endurain): self-hosted run viewer fed by RunnerUp WebDAV exports	2026-08-23 13:09:18 +02:00
extract-android-guardian-2026-08-23.json	Extract android_guardian; heavy_job_lock goes to its installer	2026-08-23 18:00:58 +02:00
extract-digital-wellbeing-2026-08-23.json	Extract digital-wellbeing; the directory-depth cap now enforces	2026-08-23 20:57:51 +02:00
extract-hosts-blocker-2026-08-23.json	Extract the hosts blocker into its own repo	2026-08-23 20:30:18 +02:00
extract-mtk-root-2026-08-23.json	Extract mtk_root, with the libraries that lived outside it	2026-08-23 17:39:27 +02:00
extract-phone-focus-mode-2026-08-23.json	Extract phone_focus_mode together with focus_policy	2026-08-23 18:32:58 +02:00
extract-system-maintenance-2026-08-23.json	Extract system-maintenance, the isolated subsystem	2026-08-23 17:55:29 +02:00
extract-three-apps-2026-08-23.json	Extract reverse_survivors, kcd2_dice_solver and focus_owner	2026-08-23 17:34:02 +02:00
flatten-i3-and-misc-2026-08-23.json	Flatten i3-configuration and misc/testsAndMisc-bash	2026-08-23 19:02:18 +02:00
flatten-linux-configuration-2026-08-23.json	Dissolve linux_configuration/scripts/ and single_use/	2026-08-23 18:53:30 +02:00
hosts-guard-tests-2026-08-23.json	Clear both hosts-guard libs; teach the instrument done <redirect> (88 -> 86)	2026-08-23 11:50:13 +02:00
installer-cross-repo-fix-2026-08-24.json	Make install_core_system.sh work on a fresh machine	2026-08-24 12:45:21 +02:00
nvidia-config-tests-2026-08-23.json	Test nvidia_config.sh to 100%; the 8-lib campaign is done (85 -> 84)	2026-08-23 13:14:37 +02:00
split-boot-repair-2026-08-23.json	Cap extensionless scripts, and split the one that was hiding	2026-08-23 19:13:40 +02:00
suite-speedup-2026-08-23.json	Take the fixes/lib suite from 110s to 37s, same 645 assertions	2026-08-23 14:00:13 +02:00
template.json	style(prettier): apply markdown/json formatting updates	2026-05-07 22:08:00 +02:00
thorium-repairs-tests-2026-08-23.json	Test thorium_repairs.sh to 100%, lift the gate timeout (89 -> 88)	2026-08-23 11:05:56 +02:00
token-audit-batching-axis-2026-08-28.json	token_audit: measure tool-call batching per API message	2026-08-28 12:22:21 +02:00
ubuntu-perf-fixes-tests-2026-08-23.json	Test ubuntu_perf_fixes.sh to 100%, take it off the allowlist (86 -> 85)	2026-08-23 12:38:40 +02:00
wire-shell-coverage-ratchet-2026-08-23.json	Wire the shell ratchet into pre-commit, and test the coverage tooling	2026-08-23 15:51:29 +02:00