Two sources that disagree are not an error

✓solved

Problem

Deciding whether a validator should reject contradictions between sources when the data model must hold disagreeing traditions.

Approach

orderingProblems separates a cycle within one work, a data error, from a cycle that appears only across two works, a dispute that keeps both orderings and flags the events.

Outcome

✓ solved

Conditions

ADR 0022 says the catalog never picks one reading; ordering of events is checked as a graph, with the work recorded on each ordering.

Evidence

src/lib/catalog/ordering.test.ts; src/lib/catalog/ordering.ts; orderings/badr-before-death-of-uthman-ibn-mazun.ts

Implication

When a check fires, ask who is contradicting whom: one author, a typo, or two, data; keep the source on every ordering.

Topics

evidence-modeldata-sync

The first idea for ordering events was the obvious one: a graph, so check it for cycles and reject the catalog if one exists. The critic agent pointed out that this makes Namaq pick a side. If one work says A came before B and another says B came before A, rejecting the catalog would force the author to delete one of them, and ADR 0022 says Namaq never picks one reading.

So orderingProblems (src/lib/catalog/ordering.ts) separates two cases that look alike. A cycle among the orderings of one work is a data error, because one author does not contradict themselves: someone misread a sentence. A cycle that only appears when two works are combined is a dispute: both orderings stay in the catalog, and the events involved are flagged so a later display leaves them unplaced and shows both readings. The same split applies when an ordering contradicts an event's year: it is an error when both come from the same work and a dispute when they do not.

Evidence: the tests in src/lib/catalog/ordering.test.ts (a cycle in one source fails, the same cycle across two sources is flagged), and orderings/badr-before-death-of-uthman-ibn-mazun.ts, the first record, which cites the claim ibn-mazun/death ("توفي ... بعد بدر بيسير").

Implications: a validator that rejects every contradiction cannot hold a tradition in which sources disagree, which is the main thing a history of this kind records. The useful question when a check fires is "who is contradicting whom": one author, which is a typo, or two, which is data. It is also why the source (the work) is a field of every ordering and not something inferred later.