arm

demo_v1 baseline

One config on smoke_v1, scored one way; its repeats averaged per item.

dataset: smoke_v1
scored against: silver/smoke_v1 silver
runs: 1
headline: overlap_f1 0.416
last run: 2026-10-23 14:23

01metrics

metric	mean	items	better
overlap_f1	0.416	5	higher
cost_usd	$0.000000	5	lower
latency_ms	0.009	5	lower
output_words	34.8	5	neutral

02cost and speed

cost per item: $0.000000
speed per item: p50 0.007 · p95 0.015 ms

Per item, over its items: cost as dollars and tokens (a local model is free but still uses tokens), speed as the median and the slow tail (p95). A goal that names this arm weighs them in its decision.

03runs

run	made	items	cost	how
`demo_v1_20261023T142324Z`	2026-10-23 14:23	5	$0.000000	measured

04config

config demo_v1 · adapter scripts/adapter.py · credentials are never shown

provider: echo
sentences: 3

05deploy sheet

measured by: demo_v1_20261023T142324Z on smoke_v1
against: silver/smoke_v1
model sent
model record: no model record names it
server: not recorded
build: none recorded (a hosted model has no digest)
settings sent: provider: echo
sentences: 3
adapter: scripts/adapter.py 7026de2eaed3
harness: 0.4.0 68dc55b45c82

No capture here says what production ran (the adapter's production_ran).

Command: evh deploy-sheet --config demo_v1 [--capture C]

06conditions

data.items_sha256: d56cca0b6c3cea45f104e938e41b7ebdffa753e99e219bf15e1028f23a98a8e9
deployment.source: params
fingerprint.version: 5
harness.code_sha256: 57daa50689b238eb2f37f19b2fbeb2682f6482c80b982a17a34519a26de88086
harness.version: 0.4.0
library.anthropic: 1.15.0
library.numpy: 2.4.6
library.openai: 3.33.0
library.rouge-score: 0.1.2
machine.cpu_count: 32
machine.platform: Linux/x86_64
machine.python: 3.11.17
machine.release: 7.2.5-3-omarchy
reference.sha256: 53ea94646f17bf2867a8572696b30623563a2d09828989e56cd1d11f8b910ead
scorer.id: scripts/adapter.py
scorer.sha256: 7026de2eaed3f9e9d7f5fab59677f5201f74683d2ae6ed826f7d2d365a64835a