Evidence References and Citations

Evidence references are what make StretchSearch defensible. Every finding, ranked candidate, and AI-grounded answer can point back to a specific, versioned piece of evidence — so you can always show your work.

What an evidence reference is

An evidence reference records a citable fact and where it came from. Each reference carries:

• Source type — page, seo_finding, ai_run, or manual_note.
• Evidence status — active, stale, superseded, or revoked.
• Origin type — manual, imported, crawler, or generated_grounded.
• Citable flag — whether the reference may be cited in outputs.
• Links to the project, crawl job, and originating source.

Why the statuses matter

Evidence isn't forever. StretchSearch tracks its lifecycle so citations stay honest:

• active — current and citable.
• stale — the underlying source may have changed; treat with caution.
• superseded — replaced by a newer reference.
• revoked — pulled from use entirely (for example, permission withdrawn).

When you build a client-facing audit or ground an AI answer, you want active, citable evidence — and you want a record of what was superseded or revoked, and when.

Browse evidence references

1. Open the Evidence view.
2. Filter by project, crawl job, source type, source id, evidence status, origin type, and whether it's citable.
3. Page through with limit (up to 100) and offset.
4. Open any reference to see its full detail and the source it points to.

Scope to a project or crawl job

Two focused views make it easy to gather the evidence behind a specific engagement:

• Project evidence — all references for a project, filterable as above.
• Crawl-job evidence — references tied to a single crawl job, useful when assembling the citations for one indexing pass.

How evidence connects to the rest of the product

• SEO findings carry evidence (URLs, offending values) so each issue is backed by proof.
• Ranking previews get an evidence boost from the number and confidence of supporting references.
• Retrieval and the AI bridge package evidence chunks so downstream apps can cite them (see Retrieval Context and the AI Evidence Bridge).

A realistic example

Building the Northwind audit, the analyst filters evidence to the project, source_type = page, evidence_status = active, is_citable = true. They get a clean list of citable page-level evidence to attach to each recommendation. A month later a page changes; its evidence flips to stale, prompting a re-crawl before the reference is reused.

Tips

• Filter to active + citable when assembling anything client-facing — it excludes stale and revoked material automatically.
• Use the crawl-job evidence view to snapshot exactly what a single indexing pass supports.
• Watch for stale evidence on long-running engagements — it's your cue to re-crawl before re-citing.
• Distinguish crawler and generated_grounded origins when you need to separate first-party crawl evidence from AI-derived, grounded evidence.

FAQ

Who creates evidence references? They come from several origins — the crawler, imports, manual notes, and grounded generation. As a reviewer you mainly browse and cite them.

Can I cite stale evidence? You can see it, but it's flagged for a reason. Prefer active evidence, and re-verify before relying on anything stale.

What does generated_grounded mean? Evidence produced by an AI step that was grounded in retrieved context — as opposed to a raw crawler capture or a human note.