SHARE measurement lineage · judge briefing

One question. Two instruments.

SHARE 2.0 keeps the transparent 25-check score, rebuilds Reusability from evidence available at deposit, and tests the same core question on the same reconciled records.

Methods & results Live metrics

154,569

records in both comparison columns

The original 1.0 archive had 183,872 records. Restricting it to the exact common cohort removed 29,303 negatives while retaining all 407 frozen legacy outcome labels.

The verdict first

Same records. Same SourceOf question. Cleaner separation between score and outcome.

154,569 = 154,569

Exact DOI-matched validation cohort in both columns.

362

SourceOf positives shared by the legacy and refreshed public inventories.

9 → 11

The nine 1.0 repositories remain; Figshare and OSF extend live 2.0 coverage.

Directness ledger

Every row says how comparable it really is.

“Direct” means the records, outcome labels, model family, and inference are held fixed. Refreshed and historical rows remain useful, but are labelled so different estimates are never presented as interchangeable.

Comparison

SHARE 1.0

SHARE 2.0

What it means

Validation cohort

154,569 matched records

Direct matched comparison

Every 2.0 DOI is present in the eligible 1.0 cohort; no new records were added.

Score structure

25 checks × 4 points

Direct matched comparison

The transparent 0–100 arithmetic is unchanged.

Reusability dimension

Reuse outcomes and attention

Deposit-time files and documentation

Same construct, refreshed measurement

2.0 moves later outcomes outside the score and makes Reusability actionable at deposit.

Derivative outcome

407 SourceOf records

471 refreshed SourceOf records

Same construct, refreshed measurement

The construct is the same; the refreshed public inventory shares 362 records with 1.0.

Derivative association

5.73× per +10 (historical model)

3.49× SourceOf; 2.36× strict

Historical context

The positive association persisted, but these published estimates do not hold every analysis choice fixed.

Fixed-score AUC

Not the submitted structured-model estimand

0.7381 on 154,569 records / 96 strict events

Historical context

This is a valid sealed 2.0 discrimination result, but it is not interchangeable with a fitted five-bucket model AUC.

Matched structured-model AUC

0.9261 on the common cohort

0.8786; count baseline 0.8209

Direct matched comparison

On the same 154,569 records and 407 labels, 2.0 bucket structure improves AUC by +0.0577 versus naive 25-signal counting (owner-clustered 95% CI +0.0020 to +0.1622).

Matched +10-point association

5.097× (95% CI 3.161–8.219)

4.167× (95% CI 2.078–8.355)

Direct matched comparison

Both use the identical records, 407 labels, categorical deposit-year adjustment, +10-point scale, and owner-cluster-robust inference.

ClinicalTrials.gov validation

Submitted trial-level validation

Direct crosswalk pending

Pending direct result

The same validation is feasible after reconstructing both scores without leakage on identical trial IDs, outcomes, exclusions, and splits.

Live repository coverage

9 repositories

11 repositories

Historical context

The same nine remain; Figshare and OSF are additions in the live 2.0 release.

Outcome continuity

The event lists moved, but they are recognizably the same construct.

Both versions use a public DataCite SourceOf relation indicating that a later object derives from the deposit. SHARE 2.0 refreshes the snapshot and normalizes relation direction.

The same 407 frozen legacy labels are present before and after the 183,872-to-154,569 cohort restriction. The 471 SourceOf, 114 verified-link, and 96 strict events below are distinct refreshed or narrower outcome definitions—not losses from the legacy 407.

45legacy-only

362shared SourceOf records

109newly observed

Legacy inventory: 407

Refreshed SourceOf: 471

Verified link: 114

Strict three-year: 96

Association, honestly labelled

The positive signal persisted as the validation rules became more explicit.

Historical 1.0 model

5.73×

odds per +10 deposit-time points

95% CI 4.97–6.61 · originally reported specification

Main continuity outcome

3.49×

odds per +10 SHARE 2.0 points

471 SourceOf records · 95% CI 1.76–6.90

Registered strict endpoint

2.36×

odds per +10 SHARE 2.0 points

96 verified events · 95% CI 1.42–3.92

These three estimates answer closely related questions, but they are not a paired effect-size test. The completed matched analysis therefore places both instruments on the same 154,569 records and the same 407 labels: the archived 1.0 deposit-time score yields 5.097× odds per +10 (95% CI 3.161–8.219), while the 2.0 full score yields 4.167× (95% CI 2.078–8.355).

In the direct AUC analysis, SHARE 2.0's five-bucket model reaches 0.8786 versus 0.8209 for naive 25-signal counting. The paired improvement is +0.0577 (owner-clustered 95% CI +0.0020 to +0.1622), satisfying the prespecified primary decision. SHARE 1.0's common-cohort structured model remains higher at 0.9261; that less favorable version contrast is retained in the receipt rather than hidden.

Coverage expansion

Nine shared repositories, then two additions.

Direct comparison set · 9

Present in SHARE 1.0 and 2.0

ClinicalTrials.govOpenNeuroDryadEDIGEONASAOpenAIRESRAZenodo

New in live 2.0 · 2

Coverage added, not omitted from 1.0

FigshareOSF

This is why the live repository count is 11 while the submission-era static view consistently showed 9.

Evidence ledger

Follow every number to its governing source.

Public Methods & Results evidence
https://sharescore.org/methods-results

SHARE 2.0 framework and scoring contract
https://sharescore.org/framework/versions/2.0

Public release history
https://sharescore.org/releases

Live SHARE 2.0 aggregate receipt
https://api.sharescore.org/metrics/live

Live metrics and accountability ledger
https://sharescore.org/metrics

Return to the full evidence Inspect the 2.0 rubric

Every row says how comparable it really is.

Comparison

SHARE 1.0

SHARE 2.0

What it means

Validation cohort

154,569 matched records

Direct matched comparison

Every 2.0 DOI is present in the eligible 1.0 cohort; no new records were added.

Score structure

25 checks × 4 points

Direct matched comparison

The transparent 0–100 arithmetic is unchanged.

Reusability dimension

Reuse outcomes and attention

Deposit-time files and documentation

Same construct, refreshed measurement

2.0 moves later outcomes outside the score and makes Reusability actionable at deposit.

Derivative outcome

407 SourceOf records

471 refreshed SourceOf records

Same construct, refreshed measurement

The construct is the same; the refreshed public inventory shares 362 records with 1.0.

Derivative association

5.73× per +10 (historical model)

3.49× SourceOf; 2.36× strict

Historical context

The positive association persisted, but these published estimates do not hold every analysis choice fixed.

Fixed-score AUC

Not the submitted structured-model estimand

0.7381 on 154,569 records / 96 strict events

Historical context

This is a valid sealed 2.0 discrimination result, but it is not interchangeable with a fitted five-bucket model AUC.

Matched structured-model AUC

0.9261 on the common cohort

0.8786; count baseline 0.8209

Direct matched comparison

On the same 154,569 records and 407 labels, 2.0 bucket structure improves AUC by +0.0577 versus naive 25-signal counting (owner-clustered 95% CI +0.0020 to +0.1622).

Matched +10-point association

5.097× (95% CI 3.161–8.219)

4.167× (95% CI 2.078–8.355)

Direct matched comparison

Both use the identical records, 407 labels, categorical deposit-year adjustment, +10-point scale, and owner-cluster-robust inference.

ClinicalTrials.gov validation

Submitted trial-level validation

Direct crosswalk pending

Pending direct result

The same validation is feasible after reconstructing both scores without leakage on identical trial IDs, outcomes, exclusions, and splits.

Live repository coverage

9 repositories

11 repositories

Historical context

The same nine remain; Figshare and OSF are additions in the live 2.0 release.

The event lists moved, but they are recognizably the same construct.

Both versions use a public DataCite SourceOf relation indicating that a later object derives from the deposit. SHARE 2.0 refreshes the snapshot and normalizes relation direction.

45legacy-only

362shared SourceOf records

109newly observed

Legacy inventory: 407

Refreshed SourceOf: 471

Verified link: 114

Strict three-year: 96

The positive signal persisted as the validation rules became more explicit.

Historical 1.0 model

5.73×

odds per +10 deposit-time points

95% CI 4.97–6.61 · originally reported specification

Main continuity outcome

3.49×

odds per +10 SHARE 2.0 points

471 SourceOf records · 95% CI 1.76–6.90

Registered strict endpoint

2.36×

odds per +10 SHARE 2.0 points

96 verified events · 95% CI 1.42–3.92