External validation methodology
A reproducible protocol for testing usability and demand intent while keeping internal QA, synthetic fixtures, private responses, and unsupported public claims out of the evidence.
Six validation types are not interchangeable
v7.35 primarily enables usability evaluation and demand validation. It preserves inherited software and research-integrity qualification but makes no peer-review, market-validation, or commercial-traction claim.
Software qualification
Does the system behave as specified? Internal automated and browser QA answer this; it is not external demand evidence.
Research integrity validation
Does the system preserve evidence, provenance, relationship, uncertainty, and calculation semantics?
Usability evaluation
Can an outside person understand the product and complete a meaningful research task without developer assistance?
Demand validation
Does an outside person state a real reason to reuse, adopt, integrate, or request the system?
Academic peer review
Formal scholarly review. v7.35 does not claim this.
Market validation
Evidence of willingness to pay, contract, fund, or institutionalize. v7.35 does not claim this.
From voluntary response to public-safe count
- Evaluator explicitly downloads or copies a local response.
- Evaluator voluntarily shares it outside the site.
- Controlled intake validates schema and explicit real/synthetic classification.
- Source, duplicate, privacy, quote, and contact permissions are reviewed.
- An accepted demand record remains private or becomes aggregate/public eligible.
- The deterministic snapshot consumes only eligible real records.
No suggestion becomes a roadmap requirement automatically. Feedback may produce a backlog candidate with a recorded disposition: accept, investigate, defer, reject, already supported, or duplicate.
Counts before percentages
At small N, the status page shows raw counts and states N. It does not publish percentages, confidence claims, composite demand scores, sentiment scores, or cohort details that could mislead or identify participants.
Demand classes D0–D5 describe explicit actions and statements. They do not predict conversion. “Evaluation records” is used instead of “unique users” unless uniqueness is genuinely established.
Submitted content is untrusted and private by default
Raw responses are not copied into public output. Contact information requires explicit follow-up permission and contact-private handling. Public quoting requires explicit permission and later review. Text is length-bounded, control characters are removed, and the public UI never renders submitted HTML.
Synthetic fixtures remain in tests and are excluded from real counts, quotes, pilot interest, themes, and public claims.