First run

This page runs one world, then many, then the full chain, against the bundled port_alert scenario. The outputs shown are from real runs. Digests depend on the code version, so yours may differ.

1. Run one world

uv run worldline run scenarios/port_alert.yaml --out experiments/out

scenario port_alert  root trace digest sha256:d16ac4392733589d5dbc12c893be62d64a28e73184c8eb8ae0a2bbb18354dd17
  decision duty_officer/escalate_or_monitor at 1500s -> monitor (knew 17 events)
  decision coast_watch/stand_down_if_acknowledged at 2400s -> stand_down (knew 22 events)
  PASS    absence.no_corrupted_in_decision  no decision relied on an event matching {'corrupted': True}
  PASS    decision.consistent_with_truth  all 2 decision(s) match the rule evaluated over truth
  PASS    epistemic.label_isolation  role 'coast_guard' never received an event labelled 'harbour_private'
  PASS    epistemic.no_future_knowledge  no view or decision knows anything before it was received
  PASS    response.alert_ack  all 1 alert events acknowledged within 300000000000ns by harbour_ops

The root world is healthy: every invariant passes. Run it again and the digest is byte-identical, because the clock is virtual and all randomness is seeded.

2. Branch it into many worlds

uv run worldline explore scenarios/port_alert.yaml --budget 150 --seed 42 --out experiments/out

The explorer tries up to 150 fault schedules with each of two strategies. Guided prioritizes deliveries that decision rules depend on, reading only the scenario. Random is the baseline. It prints a comparison table and writes exploration.json, along with traces and verdicts for the root and every failing branch.

3. Reduce a failure to its cause

uv run worldline minimize experiments/out/exploration.json --out experiments/out

[minimize] ...: 3 -> 1 interventions, 3 oracle calls, one-minimal True, witness ['correction-001', 'ais-006', 'patrol-001']

Three interventions were tried, and one explains the failure: delaying correction-001 to harbour_ops past the decision.

4. Or run the whole chain

uv run worldline demo --out experiments/out --budget 150 --seed 42

[explore] scenarios\port_alert.yaml budget 150 seed 42
| strategy | nodes | failing | fail/100 | first decision FAIL idx | wall ms |
| guided   | 69    | 23      | 33.8     | 6                       | 279     |
| random   | 66    | 20      | 30.8     | None                    | 262     |
[minimize] ...: 3 -> 1 interventions, 3 oracle calls, one-minimal True, ...
[genome] genome:94f0ad69dd5483d0: delay the last before decision 'correction' event (corrects dependency)
         seen by the decision role until after the decision, so that decision.consistent_with_truth fails.
[challenge] convoy_lostlink reproduced=True; controls: [('convoy_lostlink_immune', False)]
[report] 15 agent claims, 97 files
[verify] ok=True attested=16 mismatches=0 root sha256:e115b258...
[authorize] AUTHORIZED  (all 5 requirement(s) met by 15 record(s))
[authorize tampered invariant:absence.no_corrupted_in_decision] DENIED
(integrity:record_edited@1: record 1 content does not match its hash)

The demo also runs the unit tests first, because the release policy requires a tests:unit claim.

5. Look at it

cd ui && npm ci && npm run build && npm run preview

The viewer reads experiments/out/ and opens on the Story view.

Next: Read the output.