Records and results

simulate returns a SimResult in memory. It contains recorded channels, metrics, task and node identifiers, and a resolved configuration snapshot.

sim = simulate(:tracking; node=:falandays, ticks=1000, seed=11, window=1000)
outcome = task_outcome(sim)

Use one simulation for diagnosis and exploration. Use an operation plan when the result needs repeated trials and a durable record.

The explicit window acknowledges that this run is shorter than Tracking’s ordinary 2,000-scored-tick minimum. Treat its outcome as a short diagnostic, not study evidence.

The raw outcome always uses the recorded scoring window. A measured anchor applies only at the scored_ticks interval used to measure it. If the intervals differ, normalized and normalized_bound are missing. Check normalization_status and anchor_scored_ticks before reporting a normalised value. Analytic anchors are window-invariant.

Standard record

record-id/
├── record.toml
├── request.toml
├── resolved.toml
├── environment/Manifest.toml
├── seeds.csv
├── data/
├── summary/
├── figures/
├── report/index.html
└── DONE

request.toml preserves the submitted plan. resolved.toml records all defaults and settings that ran. environment/Manifest.toml preserves the exact dependency resolution. seeds.csv is the realised seed ledger.

CSV files are the authoritative tables. summary/summary.json supports other tools. The HTML report presents the same typed result for a reader.

record.toml stores the package and Julia versions, the Manifest SHA-256 digest, Git revision and state, artifact inventory, and artifact checksums. The inventory includes the resolved Manifest and its checksum. A failed write contains FAILED instead of DONE.

Public contribution records

Merge the experiment protocol and software before submitting public run records. Generate those records from a clean Git commit that is reachable from main.

A maintainer replay and a contributor submission are linked roles in one contribution. They do not provide two independent samples. Human review decides whether the contribution enters the accepted-run catalogue.

Accepted records are public and immutable. A text-only contribution should be at most 1 MiB and must not exceed 5 MiB.

Interpretation boundary

A complete record proves that generation finished and preserves the result. It does not promote exploratory work to confirmed evidence.

Always report the outcome key, raw value, normalised value when available, normalisation status, scoring window, independent units, horizon, warm-up, construction, reset, seed policy, failed trials, and relevant controls.

Source: src/core/Recorder.jl, src/records/Records.jl.