agent run agent_run ADR-0040

Agent run 019e4900…

Full provenance for the no-op agent's proposal of agent_proposed_xs_momentum. The row landed on a local BookState via record_provenance; promotion to a live runner still requires a named human approver (ADR-0019).

-0.1000 0.1000

operator

kind: agent
model_artifact: claude-opus-4-7
identifier: claude-opus-4-7/019e4900-0009-7000-8000-000000000009
provenance_id: 019e4900-000a-7000-8000-00000000000a

leaderboard entry

method=agent_run · objective=rank_ic

n_buckets

5

sharpe_oos

0.00

features = [ "momentum_126" , "momentum_to_vol_ratio_20" ]

horizon = 5, rebalance = "daily"

config_id = 8a4c09dc05fb

citation graph — typed-prefix cited_artifacts

2 cited · claude-opus-4-7/019e4900-000…

Top row is this agent's identifier. Middle row: the notes it cited. Bottom row (when present): notes those cited notes themselves build on (the indirect provenance). Every cited slug resolves against the writing manifest at propose-time (ADR-0015) — the agent cannot land a row citing a hallucinated source.

reproduce this run

$ python scripts/regenerate_agent_sample.py

The script holds the pinned seed / UUIDs / clock and rewrites sample_run.json. The Python drift test test_sample_run_matches_live re-runs build_sample on every CI build and asserts bytewise equality against the committed fixture.