Method

Write each fact once. Show the uncertainty every time.

The timeline and episode reader are views over shared records. Significance describes historical consequence; film coverage describes attention in a particular episode. Neither is a claim of certainty.

Local references

Yes: the project maintains a local research library. Any book, paper, PDF, subtitle, or media file available on the local computer may be consulted, searched, and cited. Local working files remain untracked when they should not be redistributed; Git contains citations, locators, and our analysis instead.

Evidence labels

Evidence is described along separate dimensions: when it was created, how it was transmitted, the source's position, and how securely the text can be read. A later memory is a kind of evidence, not automatically worthless; a near-contemporary deposition can still be interested.

Significance is not screen time

Each event carries a significance tier from I (load-bearing) to V (fine-grained context), a written rationale, and a counterfactual test for the highest tiers. The current tiers are explicitly marked provisional until the named canonical assessment method has been source-checked and stress-tested across both the imperial crisis and the later republic. Provisional tiers must not silently determine label visibility. Episode coverage uses a different vocabulary: set piece, substantial, brief, mentioned, or not covered. A future canonical distribution will be the result of event-by-event judgment, never a quota.

What “verified” means

Provisional is an editorial hypothesis. Caption-verified means spoken wording was found—or systematically not found—in the supplied captions. Every positive label requires an exact cue locator; every claimed absence records both the searched transcript and its search terms. The validator checks locators against the local SRT and reruns absence searches.

Visual evidence is recorded separately. Not checked makes no visual claim. Sequence context means the event belongs to a visually reviewed sequence, but its own image, title, or document was not independently certified. Directly verified requires a named window in a checksum-pinned audit recipe. OCR helps retrieval; the final visual finding remains a bounded human judgment.

Missing is not outside the episode

An absent episode relationship means “not assessed yet.” Outside span means the relationship was assessed and the event falls beyond the episode's date range. Not covered means the event is editorially relevant but absent from the film evidence searched. Episode spans live in one shared record, and validation rejects an outside-span label when the dates actually overlap.

Coverage describes narrative treatment, not historical importance

Set piece means the episode builds a major discrete narrative unit around the subject, normally with setup, development, and consequence. Substantial means sustained or repeated explanation inside a larger sequence. Brief supplies a bounded action, causal link, or consequence. Mentioned names or invokes the event without developing it. One caption start can introduce a long sequence, so caption count does not determine magnitude. Every positive relation carries an inspectable coverage rationale, and older episodes are re-graded whenever this vocabulary changes.

Three different review statuses

An episode essay can be editorially reviewed while some event relations remain provisional. The interface therefore reports editorial review, sequence-cue verification, spoken-relation verification, and direct visual checks separately. A broad “reviewed” label never substitutes for those counts.

Status is workflow; confidence is argument

A claim's status records where it sits in the research and viewing process. Confidence records how strongly the historical proposition is supported. The axes are independent: an argument written before close viewing can have high historical confidence while its description of the film remains provisional. Every argument-first claim also pre-registers what visual or sequence evidence would force revision, narrowing, or rejection. Close viewing is therefore a test rather than a confirmation pass.

Historical voice is a relationship

An actor is not stored as a timeless attribute of a historical person. Each performance record links a performer, historical speaker, episode, cue, and—where known—the document being read. Episode 1 relationships require both an official PBS casting credit and an attribution in the supplied captions.

Human review

Schema checks can catch missing dates, invalid coverage values, positive coverage without a rationale, unknown audit windows, and omission records without a stated editorial role. They cannot decide whether a coverage magnitude is persuasive, whether a claim is monocausal, whether a map erases sovereignty, whether a narrative smuggles inevitability into hindsight, or whether a cited page really supports the sentence attached to it. Major source passes therefore include a proposition-level citation ledger. Each citation use joins a claim to a source, locator, evidentiary relationship, verification result, review date, and finding; no source is declared universally “verified.”

Local availability is checked separately from reviewability. Every obtained source must expose a usable research representation: directly readable text, derived review text, a declared facsimile, or audiovisual material handled through the viewing-audit workflow. This rule applies across source kinds, so a newly added book or journal article cannot silently bypass review merely because its category was not anticipated.