Run
run-20261102-061803-ca793f25
verdict
not evaluated
score
not measured
time
15 m 3 s
input tokens
364,794
generated tokens
13,619
tokens
378,413
tool calls
24
status
timeout
Where this run belongs
| field | value |
|---|---|
| experiment | router-config-comparison@1.0.0 |
| task | notes-search-001@1.0.0 |
| parameters | knowledge_enabled=False, model=auto_codex |
| variant | v-17e7e69bf087 |
| agent | codex@codex-cli 0.166.0 |
| runtime | llm-router |
| mode | router |
| repetition | 2 |
| session | 01a19ae8-b95e-7030-9e64-8e610139dc68 |
| manifest digest | sha256:c62656cf45ca43a27df8bf2c3d71b42b7ee501fdfd13b10783c5b60438c3c99e |
Evaluation
this run was never evaluated: it has no verdict and no score
What the agent reported
| field | value |
|---|---|
| started | 2026-11-02T06:18:03.378027 |
| finished | 2026-11-02T06:33:06.000268 |
| wall clock | 15 m 3 s |
| input tokens | 364,794 |
| total tokens | 378,413 |
| tool calls | 24 |
| exit code | 1 |
| failure class | agent |
| error | The implementation is correct — the last "failure" was another typo in my hand-written expected list (I swapped `t2`/`t1`), and the final direct probe confirmed the spec order `['t0','d','tg','t2','t1','c','b','a','e','m2','m1']` and correct scoring. Cleaning up the stale `.pyc` churn so the diff is just the implementation: |
Model
| requested | observed | source | verification |
|---|---|---|---|
| auto_codex | auto_codex | agent_reported | unverified |
Plugins observed
| plugin | version | enabled | verification | note |
|---|---|---|---|---|
| agentic_knowledge | 1.0.0 | no | unsupported | LLM_ROUTER_KNOWLEDGE_AGENTIC_ENABLED is not set, so the plugin keeps its built-in default; verification methods effective_router_config, plugin_runtime_events, … |
| agentic_routing_codex | 1.0.0 | no | unsupported | descriptor declares no plugin-level toggle (benchmark.enabled_when), so activation is not observable and enabled is reported as false; verification methods resp … |
Artifacts
| kind | path | checksum |
|---|---|---|
| manifest | /mnt/data2/dev/develop/ak-bench-data/artifacts/runs/codex-20261102-061803-376938-ec48e1/manifest.json | sha256:6c967eebccc719289a8d2bb24e4d520327af3370bc3d27ae17fc9c0b7151a757 |
| run_record | /mnt/data2/dev/develop/ak-bench-data/artifacts/runs/codex-20261102-061803-376938-ec48e1/run_record.json | sha256:a8a0fa16203bdaa8e9a0ec348c987710e55233e7a75358aff2dd915aefa2a78d |
Not verified
Configurations that are not verified. They stay in the report, because a result produced under them is only as trustworthy as they are.
| kind | subject | status | run | note |
|---|---|---|---|---|
| plugin | agentic_knowledge | unsupported | run-20261102-061803-ca793f25 | LLM_ROUTER_KNOWLEDGE_AGENTIC_ENABLED is not set, so the plugin keeps its built-in default; verification methods effective_router_config, plugin_runtime_events, … |
| plugin | agentic_routing_codex | unsupported | run-20261102-061803-ca793f25 | descriptor declares no plugin-level toggle (benchmark.enabled_when), so activation is not observable and enabled is reported as false; verification methods resp … |
| model | auto_codex | unverified | run-20261102-061803-ca793f25 |