Resources and fidelity

A page can refer to the same URL from several document locations, under different request headers, or from different frames. Offprint records each render-affecting location as a resource reference instead of treating URL equality as content identity.

Reference, record, and outcome

A resource reference is one HTML, CSS, SVG, or embedded-document location that addresses render-affecting content. Each reference receives a stable ID and one resource record in the artifact manifest. The public record carries the resource ID, frame ID, URL, outcome, and retrieval provenance. It does not carry the exact node, attribute, stylesheet range, or rendering role.

The record contains:

• Resource and frame IDs
• Redacted requested URL and full-URL digest
• Terminal resource outcome
• Retrieval provenance when bytes were obtained

The current capture pipeline produces two outcomes:

• embedded records a content digest, media type, and decoded byte count.
• failed records a stable error code, redacted message, and retryability.

The public schema also contains external and omitted variants for format evolution. Canonical safe-static HTML rejects external rendering resources, and the current production capture path does not emit either reserved variant.

Missing-resource policy

The default warn policy keeps a failed resource record, emits a structured warning, and rewrites the reference to an inert fallback for its rendering role. The artifact can remain self-contained while visual fidelity is reduced.

Use fail when every discovered reference must be embedded:

offprint capture https://example.com \
  --missing-resources fail \
  --output example.html

This distinction separates two properties:

• Self-contained means reopening the artifact needs no external rendering resource.
• Complete resource acquisition means every discovered resource was embedded.

Read the receipt's aggregate summary for a quick decision. Inspect the artifact manifest to identify the failed resource ID, frame, URL, and error. Repeated references to one URL in a frame cannot be mapped back to their exact source locations from the public artifact record.

Retrieval provenance

An embedded resource records how Offprint obtained its bytes:

• Inline data: decoding
• Response observed during navigation
• Fetch inside the owning browser context
• Read from the owning frame for blob: data
• Read from an explicitly allowed local file

Provenance can also include the final redacted URL, final URL digest, response status, redirect chain, and number of received bytes. Observed responses preserve authenticated or service-worker-produced bytes when request identity is unambiguous.

Recursive discovery

Offprint discovers references in rendered HTML and continues into stylesheets, SVG resources, and embedded HTML. Resource-recursion depth bounds nested CSS and SVG retrieval. Frame depth bounds frames and recursively embedded HTML. Resource limits count references rather than unique URLs.

The content store hashes bytes while receiving them and deduplicates stored content by digest. totalResourceBytes counts bytes accepted before digest deduplication. The receipt's embeddedBytes value counts unique embedded digests. Encoding can still place the same data URL at every reference, so embeddedBytes is not an artifact-size estimate.

Use capture limits to control these budgets and authenticated pages when resource identity depends on headers or cookies.