decision-conformance / clef-flash/hf-fde727a/mlx-q8-mlx-community/clef-mlx-mlx0.32.3-mlxvlm0.7.4 /

2026-10-08-core-greedy-nothink-3f4f20

Checks: 47/49 passed

Check	Passed	Failures
invariants	33/33	none
batched vs separate	4/6	support/team: differs by 0.0769 support/needs_human: differs by 0.0644
repeat stability	6/6	none
option order	2/2	none
monotonic	2/2	none

Run

Subject: clef-flash/hf-fde727a/mlx-q8-mlx-community/clef-mlx-mlx0.32.3-mlxvlm0.7.4
Suite: decision-conformance · core tier · profile greedy-nothink · mode native
Items: 17 succeeded of 17 attempted (17 planned)
When: 2026-10-08 20:52:44 UTC · 0 min
Hardware: Apple M4 Pro · 24 GB
Harness: bench @ e72ea88 · scorer bench/decision-conformance 1
Suite hash: 554e796a858218e48bcb83da003554769782a4a32d3e061da87510d7f9cea490
Recipe: clef-flash/apple-m/mlx · profile mlx-q8 · inference-engines @ 05019e7
Latency: p50 1.62 s · p95 6.76 s · p99 6.76 s (per request)
Usage: 5,562 input tokens · 0 output tokens · $0.00

Reproduce

npm run bench -- run --recipe clef-flash/apple-m/mlx \
    --recipe-profile mlx-q8 \
    --on <host with a apple-m:m4-pro> \
    --suite decision-conformance --tier core --profile greedy-nothink

Files

exactly as committed under results/