The problem with agreement
Agreement among documents is not the same as independent corroboration. One source can be copied, paraphrased, summarized, quoted by later sources, imported into several tools, and presented as five separate confirmations. Agreement can also arise from shared software defaults, common timestamps, identically biased acquisition methods, or ordinary coordination. The relevant question is not simply how many sources agree, but how many causal pathways are genuinely independent.
This method is intended for analysts reviewing digital evidence, AI-related artifacts, public reporting, or disputed historical accounts. It focuses on provenance and uncertainty. It does not turn dependency into proof of fabrication.
Define one bounded claim
Write the proposition in a form that could be false, such as "the system executed a specified event during a defined interval." Identify the subject, event, time bounds, and the minimum observation needed to support the claim. Do not begin with a narrative such as "the suspect carefully concealed the event"; that combines actions, attribution and intent before the evidence is reviewed.
Build a source dependency register
For each item, preserve or describe its origin; identify the original collector or author; record when it was acquired, published or exported; and distinguish source bytes from transformed copies. Include the tool and method used to interpret it. A CSV starter is available in the Evidence Dependency Worksheet.
Link observations that descend from the same original file, recording, API export, witness briefing, manuscript edition, or common summary. A copied attachment and its PDF transcript should be modeled as two representations of one source unless the transcript includes independently verified new evidence.
Score pathways, not paperwork
Use categorical judgments rather than fabricated probability percentages:
- Independent: collected separately through a source whose relevant evidence could not reasonably be derived from the other source.
- Partly dependent: has a distinct collection pathway but shares a substantial upstream dependency, such as the same database or account.
- Derivative: transformation, quotation, import, or summary of another listed source.
- Unknown: acquisition or editorial lineage is not adequately documented.
Do not assign independence merely because timestamps or publishers differ. Conversely, two recordings made by separate devices can share a clock error, a scripted input, or a misleading premise. Independence is claim-specific rather than absolute.
Conduct an adversarial review
Give a reviewer the raw observation register without the main story attached. Ask the reviewer to identify the strongest independent evidence, the most likely shared dependency, any contradictory data, and the smallest new observation that could overturn the conclusion. Preserve dissent rather than smoothing it away.
Where safe and lawful, test a known-control dataset: supply a mixture of independently acquired records and deliberately duplicated derivatives, then measure whether reviewers correctly identify the dependencies. Report false independence judgments and unknowns as well as correct classifications. A small, carefully documented trial is more useful than an elaborate conceptual diagram advertised as validated science.
Three important limits
Authenticity: A valid signature or matching hash can confirm particular technical properties of bytes or credentials, not the truth of every claim the content makes.
Attribution: A device record, profile, or account identifier may not establish which human controlled the system at a particular moment.
Intent: Even proven removal, alteration or synchronization may have more than one explanation. Intent requires additional facts.
Cross-project application
- Anti-Forensics.com: missing, altered or ambiguous evidence.
- SyntheticForensics.org: AI-related provenance and authenticity claims.
- SyntheticImprinting.com: source-monitoring and proposed questions about familiarity and perceived origin, not a validated diagnostic effect.
The Evidence Integrity Research hub combines these research areas. This protocol is a repeatable proposal for a review method; no empirical validation or accuracy rate is asserted here.
Source and revision record
Original editorial analysis and definitions. For background reference, consult the source catalog.