Invariants

Infra invariants protect repository-state meaning. A dataset, run footprint, hash, override, sweep, or persisted artifact must remain understandable after the command that produced it has disappeared from memory.

Repository-State Flow

flowchart LR
    dataset["dataset registry<br/>raw-IQ metadata"]
    run["run footprint<br/>manifest report history"]
    variation["typed override<br/>sweep parameters"]
    provenance["hash and front-end<br/>provenance"]
    consumers["command receiver nav<br/>and later inspectors"]

    dataset --> run
    variation --> run
    provenance --> run
    run --> consumers

Invariant Families

family	invariant	breaking symptom
datasets	registry entries, sidecars, sample rates, coordinate metadata, and capture paths resolve through one infra interpretation	two commands accept the same dataset with different meaning
run footprints	manifest, report, history, and artifact headers preserve enough context for replay and review	a stored run cannot be explained without the original command invocation
overrides and sweeps	variation is typed, named, and reviewable before execution	a raw string mutation changes receiver or nav behavior without a declared parameter
hashes and provenance	hashes describe repository-facing inputs and environment evidence, not product identity	a later inspector cannot tell which input, front end, or build context shaped the run
boundaries	infra records repository state and consumes product evidence without redefining product science	infra code starts owning receiver lock policy, signal math, or navigation law

Review Gates

• A dataset change names the exact registry, sidecar, or coordinate rule it changes.
• A run-layout change preserves old run interpretability or names the migration boundary explicitly.
• A sweep or override change exposes typed parameters instead of hidden string rewriting.
• A provenance change states whether the evidence identifies inputs, build context, front-end context, or replay context.
• A boundary change routes product meaning back to command, receiver, nav, signal, or core docs.

First Proof Check

Inspect the run layout guide, dataset guide, override guide, and hashing guide. Then inspect dataset registry source, raw-IQ metadata source, run-layout source, provenance source, and infra guardrail tests.