Evidence with a boundary.

Each review begins with the organization, current and former domains, the decision being investigated, the company-provided material, the claims selected by the analyst, a date range, and an as-of cutoff. Evidence after the cutoff does not qualify for a time-locked conclusion.

Claims retain their exact source quotation and locator. A normalized research statement is stored separately so editing a research question does not overwrite what the source actually said.

Pastlight prioritizes externally verifiable claims involving dates, public positioning, products, customers, partners, pricing, certifications, locations, and other representations that may have left a public record. Financial results, private contracts, and intent usually require primary evidence outside this workflow.

For archived web research, Pastlight discovers candidate records for the identified domains and relevant page families, samples meaningful versions across the requested period, retrieves those captures, and extracts readable content. Identical or substantially duplicated versions are not treated as independent corroboration.

Retrieval can fail even when an archive index lists a capture. Pastlight preserves partial failures and does not cite an index record as if the page body had been successfully reviewed.

Deterministic text comparison identifies added, removed, and modified passages between comparable captures. Candidate observations remain tied to exact quotations and capture identifiers. Model assistance may help extract or review structured language when enabled, but it is not the source of record and invalid output falls back safely.

A change in wording does not establish why the change occurred. Pastlight describes the available record using cautious language and generates a neutral question that lets management explain definitions, timing, or context.

Before export, cited evidence identifiers are checked against the stored record and the as-of boundary. Findings display the current claim, source material and page when available, historical representation, date or capture range, observed change, potential relevance, original evidence, management question, confidence, and analyst-review status.

Confidence describes the strength and directness of the available source for the stated observation. It is not a probability that management is honest or that the claim is objectively true.

Reviewed exports include only the findings selected by the analyst. Rejected findings, evidence gaps, and private notes remain distinct from the deliverable.

• Use“The available public record indicates…”
• Use“This representation first appeared in the captures reviewed…”
• Use“Earlier captures positioned the company differently…”
• Use“No supporting capture was located within the reviewed scope.”
• AvoidClaims that Pastlight proved a representation true, false, deceptive, or fraudulent.

Archive coverage varies by domain, date, crawler access, page rendering, robots policy, and replay availability. A missing capture is not proof that a page or fact did not exist. A removed statement is not proof that it became false. A customer or partner logo is not proof of contract status or commercial scope.

Historical public evidence should be corroborated with primary documents and management responses before a material decision. Pastlight is analyst-assistance software, not legal, accounting, investment, or investigative advice.

See the practical guides to verifying company claims and using Wayback Machine evidence.