# EU PET Observatory — validation and remaining issues

Implementation/verification date: **25 September 2026**.

## Reconciliation

| Worksheet | Imported data rows |
| --- | ---: |
| MEP ranking | 95 |
| Legislation | 58 |
| Proposals and history | 54 |
| MEP amendments | 309 |
| Guidance and decisions | 57 |
| Scope and gaps | 27 |
| MEP evidence | 1,041 |

The four evidence worksheets total **478 rows**. Evidence/original-scoring cell checks found **zero mismatches**. All 95 original current-member results reconcile, including score, competition rank, proposal count and legislative-file count. The original download is byte-identical to the supplied workbook: SHA-256 `f1a4c9f6cbd89944801278eed25a37c2f7472a15bab04911be22fd63eab7746b`.

## Functionality checked

- Twenty-six automated checks cover all worksheets, original results/checksum, current-member eligibility, equal file weighting, deduplication, complete matching groups, zero/null handling, tied ranks, provisional quantities, review consistency, profile metadata, dated links, OR/AND filters, PET distinctions and XLSX/CSV contents.
- All 101 official profile destinations were matched to the correct identity. The official current-member XML had 718 entries; 95 workbook identities matched current members. Six former identities are excluded from personal scores/exports.
- Original results contain 89 scored current MEPs, 26 provisional MEPs and six unscored current members. Personal scoring currently covers the Digital Omnibus and 170 distinct eligible proposals.
- There are 37 explicit PET rows, including four recital-only and one justification-only reference. Encryption and related measures remain distinct from explicit PET terminology.
- Browser checks exercised search, empty states, selectors, explicit-PET filtering, scoring scope, linked profiles/detail panels, review editing/persistence, export actions and responsive phone/tablet/desktop layouts.
- A preview-only review changed amendment 216 from neutral to +1, recalculating Pekka Toveri from 75 to 87.5, and survived reload. It was restored to the original neutral assessment. Preview history is local and is not published to production.
- Generated Excel bytes were reopened to check rows, scores, empty values, sheets and clickable official-profile links. CSV was separately checked for full URLs. UI export actions completed; managed-browser download-event capture was unavailable, so attachment contents were verified directly.

## Priority 1 and 2 update

- Corrected person filters: country and political group cannot come from different co-signatories; MEP identity, demographic criteria and attribution role intersect on one named association. The former Portugal/ECR combination that matched 12 mixed-signatory evidence rows now correctly matches zero.
- Exposed all nine detailed workbook scoring-role labels plus explicit rapporteur/co-rapporteur associations, preserving the distinction from verified amendment authorship.
- Added side-by-side amendment comparison from tables and details. Both sides show source locators, verified proposers, scores/rationales, baseline/proposed summaries, separate exact PET excerpts and dated link checks.
- New tests cover mixed-person regressions, role intersections, export/current-member consistency, same-provision candidates, lossless word comparison, missing text and distinct signatories within deduplicated proposals. Original aggregate scores and evidence are unchanged.
- Browser verification confirmed the Portugal/ECR zero-result case, OR selections, detailed role narrowing, matching MEP cohorts, scope recalculation and export actions. Comparison selectors, summary highlighting, empty states, deduplication notices and links to full evidence worked on desktop and a 390-pixel phone viewport. Corresponding desktop sections align; phone cards stack.
- Matching provision locators use normalised workbook labels, not inferred legal equivalence. Different forms of citation may require browsing the same legislative file manually. Sixteen amendment evidence rows contain verified PET excerpts; complete verbatim baseline/amendment pairs are not supplied, so highlights compare summaries and are explicitly not legal redlines.

## Navigation update

- Added direct Legislation, Legislation in development, Guidelines and MEP positions navigation, retaining Overview, Technologies, Evidence Explorer, Methodology and an explicit Archive.
- All 478 evidence rows reconcile across the four source collections: 58 adopted-legislation rows, 289 pending legislative rows, 57 official-guidance/decision rows and 74 archived legislative rows. Pending material splits into 43 proposal/draft rows and 246 amendment rows; two judgments/enforcement decisions remain distinguishable from other official material.
- Tests verify category boundaries, historical exclusions, filter combinations and matching section metadata in exports. Desktop and 390-pixel mobile browser checks confirmed section/subsection selection, carried filters, grouped filter controls, empty states, amendment comparisons and export actions. No source statuses, original assessments or scoring rules changed.

## Remaining data limitations

1. **160 of 233 document/supporting URLs could not be content-rechecked.** These returned restrictions or successful-but-empty automated responses. Seventy-three document URLs were content-confirmed; no URLs were declared broken. The 101 verified profiles bring the total to 334 distinct URLs. Per-URL dates/issues appear in Methodology and `public/link-verification.json`. Earlier workbook verification remains separate.
2. The dataset is a **targeted snapshot through 19 September 2026**, not a certified complete inventory. All 27 scope notes are retained. Membership/link checks on 25 September do not update legal-status research beyond the cut-off.
3. Individual scored evidence is concentrated in the Digital Omnibus. Fewer than three scored proposals is provisional; confidence is separate. Scores do not measure beliefs, votes or party attitudes.
4. Amendment-date affiliation is not consistently verified. Missing historical affiliation is flagged; today's group is not substituted. Current group/country comes from the dated official check.
5. Marie-Sophie Lanig's verified official profile now displays **Marie-Sophie Christ**. The current name is shown and the workbook name retained as an alias. The same official numeric profile, biographical details and Parliament records establish continuity.
6. Future membership/legal changes need new dated verification. PET deployment alone does not establish lawful international transfer.

The original score scale, eligibility reasons, baseline, proposed change, rationale, confidence, source locator, transfer position and safeguards remain in panels/exports. Missing evidence is not replaced with invented values.