One engine for every step between raw evidence and a report you can defend.
Six stages, each with an output a reviewer can open. Deterministic where it matters, labeled where it is simulated.
- 01 · Ingest & provenance
Every file is fingerprinted
SHA-256 of the original bytes, content-type checks that catch a renamed executable, storage by hash, duplicates recognized.
- 02 · Time normalization
Clocks aligned, originals kept
Source time never changes. Normalized time is stored beside it with its rule. Offsets are inferred by matching moments that several sources recorded.
- 03 · Conflict rules
Contradictions with a name
RULE-DUP-001 copied records, RULE-TIME-001/002 declared and inferred clock offsets, RULE-CLAIM-001 unsupported identity. Each states what it cannot tell.
- 04 · Claims & hypotheses
Four states, no fake odds
Supported, Contradicted, Inferred, Unresolved, each citing evidence. Competing explanations compared side by side.
- 05 · Review & audit
Humans decide, on the record
Decisions need a reason and land in an append-only audit trail with who, when and why. They survive re-analysis.
- 06 · Report
Exports that keep the doubt
Markdown and JSON with sources, both times, rules, hypotheses, unresolved questions and limitations.
Private engine, public interface.
The browser only talks to the website. The evidence engine runs beside it and is never exposed to the internet. No AI model decides an outcome.
- Rules: deterministic, versioned IDs
- Store: resettable demo state
- Files: content-addressed