Log in
benchmarksPublic

Commit 54b9920b

· Browse files at this commit

54b9920be8df

Merged on mrmrs-0a9c5c
mrmrs · 1 month ago · 6 files ~6 modified
M README.md
@@ -317,8 +317,9 @@
317317 python3 scripts/task_loop.py --subjects git --git-remote local \
318318 --scenarios agent_task_baseline_v1 --runs 10
319319
320-# Actual Oak HTTP publication attempt whose loopback fixture reads one request
321-# then drops the reply. Use pinned released/current binaries; no live remote is used.
320+# Actual Oak HTTP publication attempt whose loopback fixture verifies the
321+# declared request bytes, then drops the reply. Use pinned released/current
322+# binaries; no live remote is used.
322323 python3 scripts/task_loop.py --subjects oak_installed,oak_local \
323324 --oak-installed-bin /path/to/released-oak \
324325 --oak-local-bin /path/to/current-oak \
@@ -327,8 +328,11 @@
327328 ```
328329
329330 These cells are different evidence classes. Local Git wall time is not a
330-network comparison, and the interruption fixture measures preservation and
331-outcome knowability rather than successful publication latency. Use
331+network comparison. The interruption fixture reports framed-request delivery
332+knowledge separately from the CLI's own outcome claim and verifies exact local
333+head, unpublished state, and payload preservation; it does not measure
334+successful publication latency. Its call counts are scripted command
335+equivalents, not model-observed tool calls. Use
332336 `parallel_contention.py --modes workspace_per_task` for concurrent isolated
333337 tasks instead of duplicating its worker implementation here.
334338