arm
triage_kimi_k31_pro_v2
One config on orrery_triage_v1_labelled, scored one way; its repeats averaged per item.
- dataset
- orrery_triage_v1_labelled
- scored against
- gold/orrery_triage_v1_labelled gold
- goal
- Triage decisions on Orrery's error signals
- runs
- 1
- headline
- triage_cost 0.723
- last run
- 2026-10-31 15:51
- re-scored
- the goals read its re-score, runs-rescored/5f2c057cdaa5/runs/orrery_triage_v1_labelled/triage_kimi_k31_pro_v2
01metrics
| metric | mean | items | better |
|---|---|---|---|
| triage_cost | 0.723 | 65 | lower |
| correct | 0.585 | 65 | higher |
| cost_usd | $0.012041 | 65 | lower |
| deferred | 0.000 | 65 | lower |
| escalated | 0.354 | 65 | lower |
| false_alarm | 0.710 | 31 | lower |
| false_dismiss | 0.147 | 34 | lower |
| latency_ms | 21532.6 | 65 | lower |
| model_correct | 0.523 | 65 | higher |
| model_escalated | 0.000 | 65 | lower |
| model_false_alarm | 0.355 | 31 | lower |
| model_false_dismiss | 0.500 | 34 | lower |
| model_triage_cost | 1.477 | 65 | lower |
| n_probes | 2.015 | 65 | higher |
| no_decision | 0.000 | 65 | lower |
| table_miss | 0.000 | 65 | lower |
| tokens_in | 4250.6 | 65 | lower |
| tokens_out | 354.0 | 65 | lower |
| triage_cost_p10 | 1.108 | 65 | lower |
| triage_cost_p3 | 0.569 | 65 | lower |
02cost and speed
- cost per item
- $0.012041 · 4,605 tok
- speed per item
- p50 22,561 · p95 34,524 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 |
|---|---|---|---|---|
triage_kimi_k31_pro_v2_20261031T152828Z | 2026-10-31 15:51 | 65 | $0.782642 | measured |
04config
config triage_kimi_k31_pro_v2 · adapter examples/triage-fleet/adapter.py · deployment hosted · credentials are never shown
- max_tokens
1024- model
eval-kimi-k31-pro- provider
gateway- reasoning_effort
none- usd_per_mtok_in
2.0- usd_per_mtok_out
10.0
05conditions
- data.items_sha256
d428aca49adc8e762dcbbcad6c9a1dba1ac4e4e3315ff26025af975faba1176a- deployment.machine
homelab LiteLLM (the fleet's gateway)- deployment.model
eval-kimi-k31-pro- deployment.source
record- fingerprint.version
5- harness.code_sha256
216b9a5df01c9986c4f863acb6f4d79c9796becf8a6944ba1e82f747844cd244- harness.version
0.5.0- library.anthropic
2.3.0- library.numpy
2.5.4- library.openai
3.41.0- library.rouge-score
0.1.2- machine.cpu_count
32- machine.platform
Linux/x86_64- machine.python
3.14.7- machine.release
7.2.5-3-omarchy- reference.sha256
f617ad3bb9bb9f5feb8ffe54e0ef1b9b05dfb316fac13f63a925f9d35fb191cd- scorer.id
examples/triage-fleet/adapter.py- scorer.sha256
3d3a6afdb978f2db2d6c6f732d5842d09c3df0eca24ef724171a162b4b937527
06the same config, scored other ways
scored against silverscored by a judgescored against a label set, not a reference setunmarked rows are scored against gold
| arm | scored against | headline | mean |
|---|---|---|---|
| triage_kimi_k31_pro_v2 · rescored 3d3a6afd | gold/orrery_triage_v1_labelled gold | triage_cost | 0.723 |
| triage_kimi_k31_pro_v2 · rescored 5f2c057c | gold/orrery_triage_v1_labelled gold | triage_cost | 0.723 |
07rounds
- Round r3 of Triage decisions on Orrery's error signals, closed: one of its arms.