@nodetool-ai/reliability-harness
v0.8.1
Published
Reliability harness core: journey manifests, RunRecord, normalization, and per-channel stream diff (docs/RELIABILITY_ARCHITECTURE.md §12)
Readme
@nodetool-ai/reliability-harness
Core of NodeTool's reliability harness: journey manifests, the RunRecord
format, output normalization, and per-channel stream diffing.
A journey replays a scripted workload against a driver and records everything
it emitted as a RunRecord. Normalization strips the parts that legitimately
change between runs (ids, timestamps, durations) so two records compare on
behavior alone; the diff is per channel, so a change in log output does not
mask a change in results.
Layout
| Path | Holds |
|---|---|
| core/journey.ts | Journey manifest schema and loading |
| core/record.ts | RunRecord — what one replay captured |
| core/normalize.ts | Run-to-run variance stripped before comparison |
| core/diff.ts | Per-channel stream diff |
| core/golden.ts | Golden-record read/write |
| core/invariants/ | Assertions that must hold across every journey |
| drivers/ | What a journey drives |
| faults/ | Fault injection |
Usage
npm install @nodetool-ai/reliability-harnessThe package also ships a CLI entry (cli.ts) used by the repo's Ring 0
reliability journeys, which nodetool harness gate selects for diffs touching
packages/kernel/.
