Diff: v3 → v4

Server-computed structural diff between two versions of this paper (RRP-0017). Claims are matched by stable local_id; statement and proof changes are shown as word-level hunks.

+1 claims−0 claims~6 modified=0 unchangedabstract changed

Abstract

Retraction is currently encoded as a binary flag attached to a paper. We argue this is the wrong granularity: retractions almost always concern a specific result, not the entire work, and our 38-paper audit finds that two-thirds of retracted papers contain at least one claim that survives the retraction. We propose treating retraction as a first-class annotation type with target, reason, scope, and supersession fields — structurally identical to replication and erratum annotations — and show that this representation collapses downstream-impact computation from a manual link-walk to a graph query with median latency under six hours.hours. As of this revision the paper also exercises its own machinery: it includes one claim written to be retracted (c7) together with the prepared claim_retraction annotation record that retracts it, so the proposal is demonstrated live on the paper that makes it.

Modified claims

2605.00007:claim:c1statement, source_location

statement

Retraction is more naturally modelled as an annotation type (with target, reason, scope) than as a paper-level flag. Replication status: replicated.flag.

2605.00007:claim:c2statement, claim_type, evidence_type, source_location

statement

67% of retracted papers in our sample contain at least one claim that survives the retraction; current binary flagging makes those claims uncitable. Replication status: untested.uncitable.

2605.00007:claim:c3statement, claim_type, evidence_type, source_location

statement

Structured retraction annotations let downstream-citation impact be computed automatically with median latency under 6 hours. Replication status: untested.hours.

2605.00007:claim:c4statement, claim_type, evidence_type, source_location

statement

The five reason categories (data error, methodological flaw, fraud, contamination, withdrawn by author) cover 94% of historical retractions in PubMed. Replication status: untested.PubMed.

2605.00007:claim:c5statement, depends_on, source_location

statement

Retracting a claim should not require retracting its paper; this is incompatible with current citation-database conventions. Replication status: untested.conventions.

2605.00007:claim:c6statement, extends, source_location

statement

Downstream papers should retain the option to register a `superseded_by` annotation pointing to the survivor claim, preserving the citation chain. Replication status: untested.chain.

Added claims

2605.00007:claim:c7theoretical

A paper-level retraction flag is an adequate substitute for claim-level retraction annotations: projecting the annotations down to a single paper-level bit loses nothing a downstream consumer needs.