From sources to accountable claims

A source may describe an instrument at a particular time, use a historical name, or disagree with another account. The research task is to retain that context while creating useful relationships and views.

Keep sources and assertions distinct#

An attributed source account is evidence that a statement was made. It does not automatically establish that the statement is correct, current or equivalent to another statement. Reconciliation therefore requires explicit decisions and the possibility of leaving a relationship unresolved.

Document the transformation#

Decision records explain interpretative choices. Execution records identify the software, parameters and inputs used in a computational step. Together they make the route to a result inspectable and help distinguish a changed interpretation from a changed processing environment.

Worked example: two specification accounts#

Imagine two catalogues describing the same organ. One reports twelve stops; another lists fourteen. Before calculating a difference, ask what each source counts, which configuration it describes and whether it is independent of the other. The disagreement might reflect an alteration, a counting convention, a copied mistake or an unresolved identity. This example is illustrative, not a claim about a particular reference instrument.

The useful output is not merely a preferred number. It is an attributable statement whose relationship to the two accounts remains inspectable. If the evidence cannot distinguish the alternatives, leaving the result unresolved is an informative outcome. A later source can then change the interpretation without erasing the earlier evidence.

Separate a processing failure from an uncertain claim#

A parser may fail to recognize a date that is plainly present; that is a processing problem. A source may genuinely leave a date unspecified; that is a limitation of the evidence. Recording these separately prevents software improvements from being mistaken for new historical discoveries. It also helps direct review: correct a transformation when the source is clear, and seek additional evidence when it is not.

Assess the source for this particular claim#

PLENUM’s thesis model starts from a proposition, a source fragment, a subject identity, an instrument state and a time scope. It considers provenance integrity, directness, author competence, temporal fit, methodological rigor, location specificity and contextual completeness. The same publication can therefore be strong evidence for one statement and weak evidence for another. An account close to a rebuilding event may still say little about the instrument’s original construction.

Keep source suitability, extraction quality and entity matching separate. An accurately transcribed passage can be attached to the wrong instrument; a correct identity match can point to a weak source. The documented policy combines indicators as a heuristic, not a probability that a historical claim is true. Missing assessment is not positive evidence. When strong accounts disagree, the research record should retain a contested claim rather than hide the conflict inside an average.

Source dependence also matters. Several catalogues repeating one earlier account do not provide several independent confirmations. PLENUM groups shared evidence lineages for assessment while retaining the individual accounts. This is a practical treatment of dependence, not proof of statistical independence between the remaining groups.