# Patefacta independent review and usability protocol

Prepared September 11, 2026. This is a protocol, not completed participant research.
Human reviewers completed: 0. Usability participants completed: 0.

## Evidence review

An appropriately qualified clinical/regulatory researcher who did not prepare the evidence should inspect the 27 packets in the authorized review workspace. Read retained source passages and current primary documents, then record the exact DIN, ingredient–strength pairs, form, route, study identifier bridge, arm/cohort scope and remaining limitations. Government review citations and trial-product exposure are separate questions.

For each case, select approval only when the stated scope is supported; otherwise record the evidence still needed. The two Adheroza 420 mg cases require an applicable bridge between the 150 mg trial presentation and the Canadian kit. Do not infer that bridge from the same active ingredient or a brand citation. All decisions retain the evidence version, source locators, reviewer identity and time. A new evidence revision requires a new review.

For the 16-pair provisional benchmark, a human assessor should label the primary sources without seeing the machine predictions or AI labels. Record document-level ingredient relevance, qualified linked intervention support, full-formulation evidence and exact-product evidence separately. A second human should adjudicate disagreements. Report reviewer agreement and unresolved cases. Never silently replace the frozen AI labels; publish a dated human-label version and rerun the frozen matcher.

This 16-pair set has already been evaluated and must not be presented as an unseen test for subsequent rule changes. Future development requires a separate untouched set, including positive full-formulation cases, salt/hydrate ambiguities, devices, kits, observational cohorts and naturally retrieved false matches. Search recall needs a separate reference set of known relevant studies with defined source/time coverage; this classification set cannot measure it.

## Usability pilot

Recruit 5–8 consenting adult student or pharma-research users. No participants have been recruited by this update. Use public drug records only. Explain that the tool is a research prototype, then let participants attempt the tasks without coaching:

1. Find marketed PLAN B and record the DIN, strength and catalogue snapshot date.
2. Open its study timeline, compare broad and product-name scopes, and describe what names were searched.
3. Explain whether an ingredient relationship proves the selected Canadian formulation was used.
4. Find a regulatory-history gap and distinguish missing source information from a failed fetch.
5. Change pages and return; explain the source-check date and visited-page coverage.
6. Submit an explicitly labeled test correction using a public source, save the receipt, and retrieve its status. Mark test reports clearly so they are not confused with real evidence corrections.

For each task record participant code, device/browser, start/end time, completion without help, prompts needed, incorrect interpretations and comments. A task is not successful if the participant confuses ingredient evidence with exact-product or efficacy evidence. Report the raw numerator and denominator of task successes; do not invent satisfaction scores. Separate observed user difficulties from the facilitator's interpretation.

## Performance

Measure navigation-to-visible-results in real browsers on mobile and desktop, separately from API latency. Record new/reused connections, cached/uncached source status, query scope, response correctness and every error. Increase concurrent load only through an authorized bounded test plan. Retain failed runs. Do not extrapolate the 18-request September 11 convenience workload to catalogue-wide performance, uptime or 10,000-user capacity.

## Release record

Publish dated methods, denominators, source versions, unresolved cases and corrections. Automated tests establish only the behavior tested; AI-source agreement, human review, usability evidence and clinical accuracy are different outcomes.