# Remaining independent checks

These are prepared tasks, not completed participant or human-review results.

## Independent evidence review

A reviewer who did not prepare the candidate must inspect each of the 25
prepared packets through the private review workspace. Confirm the exact DIN,
all ingredient–strength pairs, form, route, trial identifier, source version
and supporting passage before approving. Record disagreement and its reason.
Do not approve either Adheroza 420 mg packet until a source connects the trial
presentation to the Canadian presentation. The newly published regulatory
citation does not satisfy that requirement by itself.

Freeze the candidate hashes before labels are collected. Report the number
reviewed, unresolved and rejected separately. Compare algorithm decisions with
independent labels; include false product links, missed links and abstentions.
The same 27 development cases cannot establish independent generalization or
registry-wide recall. A held-out set with both matching and nonmatching
presentations is needed for that claim.

## Participant and phone tasks

Use actual iOS Safari and Android Chrome devices, recording OS/browser/version,
connection type and screen size. Ask participants to complete these tasks:

1. Find Plan B DIN 02293854 and distinguish product-specific evidence from
   broader levonorgestrel studies; inspect the search names and registry link.
2. Find Epsolay and explain the difference between its sourced NOC date and
   the blank RDS decision-date field.
3. Find Maviret DIN 02467550 and identify the separately sourced 2017 NOC event,
   the retained bulk date and an activity naming the other DIN 02522470.
4. Open each Adheroza trial citation and identify what remains unverified.
5. Navigate study pages, return to the prior page, change the product, and
   export a research record without mixing products, queries or releases.

Record first-submit-to-results time, task completion, errors, assistance,
misinterpretations, and keyboard/screen-reader barriers. Predefine the sample,
conditions and success criteria before collecting results. Report actual
participant counts and observed device results; CSS-width checks are separate.

## Coverage reconciliation

Keep the denominator as the frozen 1,704-record legacy index. The changing
modern portal total is not reconciled merely by visiting all observed pages.
Retain page/version receipts and seek an authoritative stable inventory or
explanation of the count difference. A 404 or wrong-record response is an
acquisition outcome, not evidence that a regulatory event never occurred.