The Forensic Narrative Diff
Compare what accounts claim before deciding what their disagreement means
By Evelyna Bellamy
“Comparing accounts requires an event model, a source model and a place for unresolved claims. A constructed gate incident shows how two descriptions can share a sequence while attributing different motives. The proposed diff exposes those differences without calling shared assertions physical invariants or using agreement as proof.”
Editorial essay by Evelyna Bellamy; not externally peer reviewed. 655 words. Estimated at 200 words per minute for the abstract and main prose; references and interactive studies take additional time. Publication details revised 10 October 2026.
Revised 2026-10-11. Complete bounded draft with identified evidence and declared examples. Earlier universal claims and unsupported causal assertions withdrawn.
Two accounts, one tempting shortcut
Imagine two descriptions of the same locked gate. One says an attendant protected equipment; another says an attendant excluded a visitor. It is tempting to strip away motive and call the residue objective fact. That shortcut hides a question the comparison still needs to answer: what supports the residue? Two writers might share a mistake, copy one another, or leave out the same event.
I propose a narrative diff that preserves three layers: a source-backed observation, a claim made by an account, and the editor’s interpretation of their relation. The layers are analytical choices. They do not create a view from outside all narration. Their value is practical: a reader can challenge a specific edge between evidence and inference.
A completely declared incident
The following is teaching fiction, not a report of a real institution. The invented log contains three entries: 14:02, a visitor reaches Gate A; 14:03, an attendant locks Gate A; 14:05, the visitor leaves. Assume for the first comparison that both authors saw this same log, and that no video, policy document, interview or explanation of the attendant’s intention is available.
Account A describes the lock as a safety measure protecting equipment. Account B describes it as an act of exclusion targeting the visitor. Both accounts assert the same order. Neither supplied record establishes motive. The diff can mark their event sequence as shared, their descriptions of purpose as divergent, and their evidence for purpose as absent. Calling either purpose false would go beyond this stipulated record.
Match events without manufacturing equivalence
Use location, participants, action and time as explicit matching fields. If another account says Gate B was locked at 14:04, leave the match unresolved unless a rule explains why those differences can be tolerated. A convenient alignment should not silently rewrite the source. Likewise, “the visitor was unable to enter” is an inference from the locked gate unless the record also identifies other entrances or an attempted entry.
This model allows a useful distinction between omission and contradiction. An account that never mentions the visitor’s departure omits an entry present in the declared log. An account that states the visitor remained at Gate A at 14:05 conflicts with it. An account that says nothing about motive has not thereby endorsed either interpretation. Empty fields need a meaning, rather than a default verdict.
The source graph has its own questions
W3C’s PROV-O specification supplies relations for derivation, revision and primary-source use. Those relations help describe the dependence between an account and a log. They do not certify the log as accurate or make two dependent accounts independent corroboration. In a real investigation, the log’s creator, clock, revision history and collection conditions would become part of the evidence problem.
Now change the constructed case: Account B copied its event sequence from Account A instead of reading the log. Agreement remains, but the support structure has changed. A comparison that counts two matching accounts as two observations would overstate corroboration. The diff should expose that dependency before offering a confidence judgment.
A reader can inspect the disagreement
A useful interface would show the original excerpt, the aligned event, the proposed transformation and its stated reason together. A disputed alignment should remain editable and visibly unresolved. This essay’s invented incident supplies no measured accuracy result and no demonstration that a parser can infer motives. A generated summary would remain an additional account to inspect.
My judgment is that a narrative diff earns trust through restraint. It should reveal where descriptions select, compress and attribute, while preserving the possibility that the editor’s event model is mistaken. The earlier language of a stable physical invariant layer is withdrawn. What survives is a method for keeping the reasoning accessible to the person who disagrees with it.
APA 7th Edition Citations & Primary Sources
Primary specification for entities, activities, derivation and revision. Recording a source relation does not certify a statement’s truth.
Primary source text