Documentation and demonstration, not assertion.

Five disciplines govern the work. They are stated here because a method that is only described when asked is not a method.

Investigation, not thesis. Work begins from a question — a battle, a figure, a single episode — never from a conclusion in search of support. A hypothesis may be advanced in order to test it. Verdicts come only after the evidence chains are anchored.

The evidentiary firewall. Every factual claim resolves to one of three states: documented, not documented, or not asserted. The historian adjudicates; the software proposes. Three working principles carry most of the weight. Silence is a finding, not a disproof — an empty confirming lane is a real result, and it cannot disprove something that by design would leave no record. Counts are not votes — a high tally of matching lines is not a verdict, and a documented label must name a primary source. And every claim traces to its primary root, because secondary works citing one another can look like independent documentation and are not; where the chain cannot be completed, we say so.

The lexical lens. Searches use period vocabulary, spelling, and OCR artifacts, and period terms are read in their period meaning. The lens cuts both ways: recall, so that what is there is not missed, and anti-anachronism, so that what is not there is not over-read.

The corridor — writing from inside the moment. The narrative stays within what was known and believed at the time, without hindsight. It operationalizes a discipline Barbara Tuchman articulated and exemplified; the framework and its structure are ours.

Demote, don't erase. A claim that fails is relabelled to its honest status, with the source it would predict if it were true. It is not deleted. Failing to prove a thread is itself a recorded result, and a verdict that is accidentally right is still outside the firewall. When evidence changes a finding, the change is recorded rather than overwritten — what the claim used to say, when it changed, and what was wrong with the original reasoning. A register that quietly replaces one verdict with another is asserting the new one; showing the change demonstrates the method.

Provenance

Public-domain status is determined, not assumed, and rests on one of five recognized bases: expired term, federal government work, edict of government, non-renewal, or publication without notice. A determination attaches to a specific volume or issue rather than to a title or a publisher, because serial runs change status mid-run.

Determinations are recorded as they are made. Where a source has not yet been adjudicated, that is its recorded state — and it is the default for material loaded before the rule took its present form.

Public-domain coverage is stated as a floor with a standing audit, never as a point estimate. Part of any corpus of this kind carries no machine-readable imprint date and can only be verified by opening it. A standing audit in 2026 identified a concentrated body of later material in one collection track, and it is held for adjudication imprint by imprint. A uniform purge by date would have removed census volumes and court reports that are federal works and edicts of government and are in the public domain on those bases — so it was not done.

That is what a standing audit looks like when it is real. The alternative — a compliance claim enforced once at load time and never tested — is the failure this discipline exists to prevent.

The original Halifax Resolves adopted April 12, 1776 in Halifax, North Carolina, North Carolina State Archives. Public domain.