Dual-Reviewer Extraction Adjudication

When a review uses dual-independent extraction (the methodological gold standard), two reviewers extract the same papers independently. Disagreements are inevitable — the question is how to resolve them transparently and auditably. EvidAI's adjudication workflow does exactly that.

When Adjudication Fires

A field is a conflict when:

• Two or more reviewers have extracted the same field for the same paper, AND
• Their extracted values disagree (after normalization for whitespace, case, and numeric formatting), AND
• The conflict has not already been resolved by an adjudicator

The engine scans every extraction in the review on each visit to the Extraction phase and computes the live list of pending conflicts.

Where to Resolve Them

The Dual Verification tab

1. Open the review → Data Extraction (/slr/extraction).
2. Switch to the Dual Verification tab (the tab badge shows the count of pending conflicts).
3. Each row shows:
   • Paper title
   • Field name
   • Number of disagreeing values
4. Click Adjudicate on a row to open the adjudication modal.

The Adjudication modal

The modal shows each reviewer's value side-by-side and asks the adjudicator to:

1. Select the winning value — one of the reviewer values, or enter a fresh value.
2. Provide a rationale — free-text explanation (recommended: reference the paper section that supports the decision).
3. Identify themselves — adjudicator user id is captured automatically.

On save:

• The decision is recorded with signer, timestamp, reason, and before/after values.
• An Extraction adjudicated audit event is written to the review's 21 CFR Part 11 chain.
• The conflict disappears from the pending list.

Discovery from the Hub

The SLR hub shows a small indicator on the Extraction phase card when conflicts are pending (e.g. "3 fields need adjudication"). Clicking the indicator deep-links to the Dual Verification tab. The modal only mounts on the extraction page — the hub is a dashboard, not a task surface.

What the Downloaded Bundle Contains

Every export bundle includes adjudication evidence under extraction/:

• extraction/adjudications.json — full list of adjudication decisions with signer, timestamp, paper, field, reviewer values, chosen value, and rationale.
• extraction/adjudications.csv — spreadsheet-friendly row per decision.
• Each paper's extracted-values file carries the resolved value, with the adjudication id pointing back to the decision record.

Why This Is Different From a Free-Text Note

Reviewer notes capture reasoning. Adjudications capture a binding decision that overrides reviewer inputs for the paper-and-field combination. Both are audit-trailed, but adjudications:

• Change downstream data — the resolved value is what appears in synthesis, meta-analysis, and GRADE.
• Are one-shot — re-opening a conflict requires deliberate re-adjudication (which produces a new chain event).
• Are reviewer-keyed — the chosen value is attributable to the adjudicator identity in every export.

Best Practices

• Adjudicate in batches — dedicate a focused session rather than adjudicating one-by-one between other tasks. Fatigue on adjudication produces brittle decisions.
• Reference the paper section in your rationale (e.g. "Table 2 footnote explicitly states n=342 per-protocol"). Future readers of the audit trail must be able to reconstruct the reasoning.
• Use the hub indicator as a pull, not a push — the hub surfaces pending work; the actual resolution happens in context (the phase page) so you have the paper, field metadata, and reviewer answers in view.
• Do not bypass adjudication by overwriting extracted values directly. The chain of evidence must show the conflict and its resolution.

PRISMA-S Alignment

Dual-independent extraction with formal adjudication is an explicit PRISMA-S requirement for search-derived data and a Cochrane MECIR standard for RCT reviews. EvidAI's workflow generates the exact artifacts reviewers and editors expect.