Exports & the trust trail

When the review is done, you do not retype anything into your analysis tool. Open the Exports panel and press Generate exports. TrialExtract writes a set of files from the current values, verifications, and decisions of this review, and nothing re-runs the extraction. Regenerate whenever you like: each file regenerates to exactly where your review stands now. A file shows generating, then ready with a size and a Download button. Generation is all-or-nothing: it regenerates the whole set from the current review, and if a value that must be plotted is missing its provenance it fails loudly rather than writing a partial, misleading set.

Which file to hand which tool

The panel splits the files into two groups. Analysis holds the numbers your tools read; Trust trail holds the proof behind them.

In the Analysis group:

• metafor (CSV): per-study effect sizes as yi and vi, ready to read straight into R. The provenance, verification, and decision columns ride along in the same rows.
• metafor script (R): an rma(method='DL') script that pools the eligible members for you.
• RevMan (CSV): a generic inverse-variance import, carrying the same trust-trail columns.
• Evidence table (CSV): every extracted value, flat, with native and standardized units side by side, plus each value’s status and source blocks.
• Workbook (XLSX): one file, multiple sheets (Evidence, Effect Sizes, Synthesis Pools, Receipts, Decisions, Manifest), for the co-author who wants to read without opening R.

Hand a meta-analyst the metafor CSV or script. Hand a RevMan user the RevMan CSV. Hand a co-author who just wants to see the numbers the Workbook.

What proves the table to Reviewer 2

When a reviewer asks where a number came from, the answer is three files, already written:

• Receipts (JSON): the provenance, derivation, and verification trail for every value. Each value carries a verification state. Where a source-grounding verification pass has not run for a value, the receipt says so plainly: provenance located the number, which is not the same as a verifier agreeing with it. The receipt never overstates what it knows.
• Decisions (JSON): the review-decision audit, holding every accept, edit, and reject you made. A reject or Needs fix decision carries a coded reason (wrong_value, wrong_source, wrong_derivation, missing_data, unclear_receipt, or other), so the page-3 value you threw out is still legible two years later, alongside why.
• Manifest (JSON): schema versions, counts, the DerSimonian–Laird estimator stamp, and a confirmed-versus-provisional summary. When someone asks which estimator produced the pool, the answer is stamped in the file, not left to memory.

These are the take-home. The Analysis files are what you compute with; the Trust trail files are what you defend with.

The export is the display

The pool in your exported files is the pool you saw on screen. The metafor files pool the eligible members: the same eligibility the Synthesis view applies, excluding values that are rejected, Needs fix, quarantined, or comparator-incompatible. The confirmed subset is the accepted-only values. Nothing widens or narrows on the way out, so the diamond in your R output is the diamond in the app. For how that random-effects estimate is formed, see Pooling across studies.

The methods sentence

You do not have to write the tool-assisted-extraction disclosure from scratch. A ready-to-paste paragraph naming the source quotation, derivation chain, verification status, reviewer count, and the DerSimonian–Laird estimator lives in the methods kit. Adjust the reviewer count, copy, done. It reports what these exports contain, in the words a manuscript expects.