Repository Evidence Foundations

bijux-gnss-infra turns declared inputs and runtime outputs into repository evidence that another process can locate, attribute, and inspect. It owns dataset resolution, run identity, layout, provenance, persisted manifests, typed overrides, sweep expansion, and artifact inspection. It does not decide whether a receiver lock or navigation solution is scientifically correct.

Follow The Evidence

flowchart LR
    declaration["dataset, sidecar,<br/>profile, run context"]
    resolution["typed resolution"]
    identity["identity and provenance"]
    execution["domain execution"]
    records["typed records"]
    footprint["manifest, reports,<br/>artifacts, history"]
    review["inspection or replay"]

    declaration --> resolution --> identity --> execution --> records --> footprint --> review

Infrastructure owns the path around domain execution, not the execution itself. It resolves what will be used, prepares where evidence belongs, and preserves what the producing crate returns.

Start From The Repository Question

question	contract
Which facts identify and describe a capture?	Dataset contracts
Where should a run and its evidence live?	Run footprint contracts
Why do two runs have matching or different identities?	Provenance and hashing
What may a profile override or experiment sweep change?	Override and sweep contracts
Can an existing artifact be identified, explained, and validated?	Artifact inspection contracts
Where does repository ownership stop?	Ownership boundary

What Infra Can Establish

Infrastructure can establish that:

• a dataset or sidecar supplied explicit capture metadata;
• a declared run context resolved to a governed footprint;
• configuration, dataset, build, repository, and front-end provenance were recorded according to the run contract;
• artifacts occupy the expected layout and conform to the selected schema and semantic inspection policy;
• typed overrides and sweep dimensions expanded deterministically.

Infrastructure alone cannot establish that a signal was truly present, a tracking loop remained locked, an observation is physically accurate, or a navigation result supports a positioning claim. Those conclusions require evidence from the producing scientific package.

flowchart TD
    manifest["valid run manifest"]
    attributable["evidence is attributable"]
    artifacts["schema-valid artifacts"]
    interpretable["records are structurally interpretable"]
    science["scientific claim"]
    owner["receiver, signal, or navigation proof"]

    manifest --> attributable
    artifacts --> interpretable
    attributable --> science
    interpretable --> science
    owner --> science

The lower two infrastructure conclusions are necessary for durable review, but they are not sufficient for the scientific conclusion.

Investigate A Broken Evidence Chain

symptom	inspect first
The capture cannot be resolved	dataset identity, sidecar completeness, and explicit unregistered-dataset policy
The run appears in an unexpected location	resume target, output override, dataset context, and derived run identity
Two apparently identical runs differ	configuration hash, repository state, build context, dataset hash, and front-end provenance
An artifact exists without clear attribution	artifact header, run report, manifest, and history entry
A valid artifact supports no scientific conclusion	the producing receiver, signal, or navigation evidence
A persisted record is rejected by a newer reader	schema version, read policy, and compatibility rules before field values

Do not repair these symptoms by assembling paths manually, rewriting manifests, or weakening validation. Such changes break the evidence chain while making the repository look superficially complete.

Boundary Decisions

Use the package overview for the concise package role, scope and non-goals for explicit refusals, and dependencies and adjacencies for each handoff. Repository fit explains why infrastructure sits between commands and scientific packages, while change principles states the durability expectations for persisted contracts.

Implementation evidence begins with the infrastructure contracts, dataset guide, run-layout guide, hashing guide, and validation guide.