Skip to content

Evidence Update Review

Workflow — instantiates Evidentiary Trace Warranting

Revisits evidence relations when sources, context, measurement, or rival explanations change.

An Evidence Update Review is a recurring workflow that re-opens already-recorded evidence relations and asks whether they still hold — because a relation that was sound when logged can quietly rot as its source is redefined, its instrument recalibrated, its context shifts, or a new rival explanation appears. Its defining trait is that it is about time, not judgment quality: it does not build better arguments or grade them, it detects staleness and forces a re-derivation whenever a triggering change lands. The review is the archetype's guard against the most insidious failure of all — evidence that was once good being trusted long after the ground under it moved. Its output is not a new conclusion but a verdict on each relation: still valid, downgraded, or void pending re-work.

Example

An environmental agency has for two years cited a river-monitoring station's readings as evidence that a factory's discharge is within permit limits. The relation — "sensor readings support the claim of compliance" — has been treated as settled. Then the vendor pushes a firmware update that changes how the dissolved-oxygen probe averages its samples, and separately the sampling point is moved 200 meters downstream after a bank collapse.

The update review is what catches this. On its trigger list are provenance changes (the sampling location moved) and measurement changes (the probe's averaging algorithm changed) — both fired. The review re-opens the compliance relation and finds it can no longer be carried forward unexamined: readings before and after the firmware change are not directly comparable, and the relocated point may sit above a tributary that dilutes the signal. The relation is marked void pending re-validation, and the agency stops citing the post-change readings as compliance evidence until the sensor is re-baselined. Without the review, two silent changes would have kept feeding a conclusion that no longer followed from the data.

How it works

The review is a scheduled-or-triggered pass over standing evidence relations:

  • Trigger list — a defined set of change-events that reopen a relation: a source being redefined or moved, an instrument recalibrated or a pipeline altered, a context shift, or a newly surfaced rival explanation. Some fire the review on arrival; others are swept on a cadence.
  • Re-derivation, not re-argument — for each reopened relation the review asks only "does the original link still follow given this change?" It re-checks provenance continuity and measurement validity, and hands off any weighing or aggregation to the mechanisms that own those.
  • Disposition per relation — each is closed as still valid, downgraded (with reason), or void pending re-work, and downstream summaries that cited it are flagged so a stale conclusion cannot outlive its evidence.

The review is deliberately narrow: it is the when-to-look-again machinery, formalizing that an evidence relation is defeasible over time and must be revised as conditions change.[n1]

Tuning parameters

  • Trigger sensitivity — how small a change reopens a relation. Sensitive triggers catch silent drift but generate churn; coarse triggers are cheap but miss slow rot.
  • Cadence vs. event-driven mix — how much is swept on a schedule versus fired by change-events. Event-driven is timely but depends on being notified; scheduled is reliable but lags.
  • Blast-radius tracing — how far downstream a voided relation's dependents are chased and re-flagged. Wide tracing prevents stale citations but is costly.
  • Default disposition — whether an un-reviewed relation past its trigger is presumed valid or presumed suspect. Presume-suspect is safer for high stakes and noisier for routine work.
  • Change-detection coverage — how many source and instrument changes are actually instrumented to fire a trigger; uninstrumented changes are invisible to the review.

When it helps, and when it misleads

Its strength is temporal hygiene: it stops a decision from resting on a dashboard whose definition changed, a citation whose source was retracted, or a sensor reading from before a recalibration. Because it re-derives rather than re-argues, it is cheap enough to run often, and it directly attacks the archetype's "a metric becomes a conclusion" failure by refusing to let a relation coast on its original credibility.

Its failure mode is coverage: the review can only reopen relations for changes it is told about, so an uninstrumented source edit sails straight through, and a poorly-scoped trigger list creates a false sense that everything current has been re-checked. The classic misuse is running the review as a rubber-stamp — reopening relations but reflexively marking them "still valid" without actually re-deriving the link. The guarding discipline is to require a stated reason for every still valid disposition, not just for downgrades, and to periodically audit the trigger list against the changes that actually occurred.

How it implements the components

  • evidence_relation_update_rule — the trigger list and disposition logic are this rule: the defined conditions under which a relation must be revisited and how it is re-closed.
  • provenance_chain — a source being redefined, moved, or retracted is a primary trigger, and the review re-checks provenance continuity when it fires.
  • quality_and_uncertainty_screen — a recalibration, pipeline change, or altered measurement condition is a trigger, and the review re-verifies that readings remain within valid tolerances after the change.

It does not combine the reopened relations into an aggregate judgment (evidence_synthesis_interface — that's Evidence Relation Matrix) nor set the strength bar a relation must clear for a given decision (decision_relevance_threshold — that's Admissibility or Relevance Gate); the review decides only whether a relation is stale and must be re-derived.

Editorial Notes

Form Classification

Form family: Assessment, Review & Assurance

Rationale: On scheduled or change-event triggers, the mechanism re-evaluates standing evidence relations for provenance continuity, measurement validity, and whether the original link still follows.

Nearest alternative: Protocol, Workflow & Routine — Triggers and handoffs make the pass repeatable, but its defining output is a renewed or invalidated evidence-relation finding.

Review outcome: Adjudicated after independent review; high confidence.

Origin Attribution

Primary origin: Medicine & Healthcare

Origin pattern: Cross-disciplinary synthesis

Present-day reach: Multi-domain

Rationale: Living evidence synthesis in medicine established recurring review and update of conclusions as new studies, changed methods, or altered certainty enter the evidence base.

Related originating lineages:

  • Accounting & Auditing — Continuous audit and evidence-reassessment practice provides the professional lineage for reopening support when sources or conditions change.
  • Philosophy — Defeasible epistemology materially supplies the idea that warrants remain reopenable under new defeaters.
  • Statistics & Experimental Design — Revising evidential relations as measurements, sources, and rival explanations change is rooted in iterative scientific inference. Statistical updating and measurement-validity practice materially shaped re-evaluation after changed instruments, data, or rival explanations.

Review resolution: Cochrane's living-systematic-review guidance makes continuous evidence updating an explicit professional workflow. Bayesian statistics supplies its formal logic, while audit contributes review discipline.

Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.

Review outcome: Researched adjudication after independent review; high confidence.

Sources consulted:

Notes

[n1] Bayesian updating is the formal expression of the same idea the review enforces operationally: a belief's warrant is not fixed but conditional on the current evidence, so when the evidence base or its reliability changes, the supported conclusion must be recomputed rather than carried forward unchanged.