Run ledger

A portable timeline of two recorded runs. Same CLI, same contract, same repo.

The run ledger is the evidence report: outcome, cost, latency, and the ordered steps of the baseline and the candidate. Missing fields stay unknown. Unknown is not a failure and it is not a pass.

node src/cli.mjs fixtures/missing-data-unknown-cost.json --text

Demo from the missing-cost fixture

This block is the engine output for fixtures/missing-data-unknown-cost.json. The candidate has no cost, so cost stays unknown.

conediff report (contract v1)
baseline  gpt-4o (run run-a)
candidate claude-sonnet (run run-b)

Outcome: unchanged ("pass")
Cost: unknown (missing data)
Latency: increased by 200 (+200, 13.3333%)

Latency attribution (per step, ms):
  generate  base 1200 ████████████████████████  cand (removed) ························  removed
  retrieve  base  300 ██████··················  cand  320 ██████··················  +20 slower

Unknown fields: 1 (candidate.cost)

runledgerkit.dev is a name for this feature. It is not a second company and it does not have its own repository. DNS for that alias is not cut over; use this site and the local CLI.

See the tool waterfall Checkout