NEOCARIDINA EXTERNAL VALIDATION PROTOCOLS
Version: 1.0
Reviewed: 2026-08-12
Status: Protocols ready locally; no external validation claimed

1. reader-comprehension-v1
Capability: reader-comprehension
Validation question: Can a first-colony or considering keeper use one guide to identify a bounded next action and explain the nearest unsupported inference?
Unit of evidence: One moderated task session using one declared guide, one scenario, the participant wording before guidance and the participant wording after guidance.
Recruitment boundary: Invite adults who self-identify as considering or within their first year of Neocaridina keeping. Do not recruit only customers, friends, praise-givers or people who already know the site vocabulary.
Procedure:
  1. Present one realistic decision scenario without teaching the expected answer first.
  2. Record the participant initial action and reason in their own words.
  3. Provide the relevant canonical guide and allow ordinary reading without coaching.
  4. Ask for the next action, what evidence or measurement it uses and what the guide does not establish.
  5. Classify the session against the declared rubric before discussing the preferred interpretation.
  6. Record the exact confusion and revise the governing guide when the same failure mechanism recurs.
Evidence supporting movement: Repeated sessions across more than one recruitment source show that readers can choose the bounded action and name its closest limitation without moderator prompting, and documented edits reduce a previously recurring failure mechanism.
Evidence requiring revision: Readers choose a materially riskier action, repeat a myth the guide intends to correct, cannot locate the required measurement or infer certainty that the guide does not support.
Interpretation limit: A completed or successful session is not proof of expertise, long-term animal outcomes, unique-person reach, trust in the publisher or correctness of the underlying biological claim.
Privacy boundary: Record no name, contact detail, account, IP address or audio by default. Store only declared experience band, recruitment source category, scenario, task wording, rubric outcome and moderator note.
Stop or review rule: Pause recruitment and revise immediately after a consequential safety misunderstanding. Otherwise review after five usable sessions per scenario or when three sessions show the same failure mechanism, whichever occurs first.
Record fields: session date; recruitment source category; experience band; scenario id; guide path; initial action; post-guide action; named limitation; rubric outcome; confusion code; edit decision; reviewer

2. question-coverage-v1
Capability: question-coverage
Validation question: Do real keeper questions reach a useful governed answer, an honest no-match, or a newly named evidence gap without forcing a similar-looking result?
Unit of evidence: One moderated, consented question submission preserved before thematic coding, with its routing result, answer boundary and final editorial status.
Recruitment boundary: Accept voluntary questions from owned and manually participated communities without scraping private groups, buying question lists or treating repeated posts as unique people.
Procedure:
  1. Preserve the submitted question and context inside the private moderation boundary.
  2. Run the shared keeper-language navigator before manual editorial routing.
  3. Classify the result as useful governed answer, clarification required, honest no-match, out of scope or evidence gap.
  4. Publish only moderator-written wording under the existing consent and redaction rules.
  5. Record whether retrieval, guide wording, glossary aliases or the named-gap registry changed.
  6. Rerun the fixed editorial routing benchmark after every retrieval change.
Evidence supporting movement: A sustained set of real questions reaches useful bounded answers or explicit no-match decisions, and recurring failures lead to documented retrieval, guide or gap changes that remain correct under the fixed benchmark.
Evidence requiring revision: Moderators repeatedly override the navigator silently, similar wording routes to an unsafe scope, questions are compressed before their original meaning is preserved or unresolved questions receive confident improvised answers.
Interpretation limit: Question counts measure submitted instances, not unique people, prevalence, biological truth, total community demand or completeness of the guide library.
Privacy boundary: Keep raw submissions, context and relationship disclosures private. Publish only with explicit consent and moderator rewriting; apply the existing withdrawal and redaction procedure.
Stop or review rule: Review after twenty-five usable question instances or when three instances expose the same consequential routing failure. Stop public answering immediately when the available authority cannot support a safe boundary.
Record fields: question id; submitted date; source category; topic; initial route; moderated route; outcome class; answer path; no-match reason; change made; published state; reviewer

3. community-use-v1
Capability: community-use
Validation question: Are bounded answers voluntarily reused in external communities with their limitation, source and correction path still attached?
Unit of evidence: One observable external reuse event with a stable public URL or moderator-verifiable record, the exact artifact used and whether its boundary remained intact.
Recruitment boundary: Count only voluntary third-party reuse or substantive response in communities where manual participation is allowed. Exclude owned reposts, paid placement, automated posting, private surveillance and undisclosed affiliate use.
Procedure:
  1. Record the governed claim reply, card, guide or standard version that travelled.
  2. Preserve the external context without copying private member information.
  3. Classify whether support, limitation, next action and source remained attached or were materially distorted.
  4. Record any substantive correction, follow-up question or moderation response.
  5. Update the portable artifact when the same detachment mechanism recurs.
  6. Keep paid reach and publisher reposts outside the voluntary-reuse evidence set.
Evidence supporting movement: Voluntary reuse occurs across multiple independently operated communities and repeatedly preserves the evidence boundary or produces substantive correction and follow-up rather than bare endorsement.
Evidence requiring revision: Artifacts routinely lose their limitation, are used as diagnosis or universal rules, depend on publisher prompting, or generate unresolved corrections that do not propagate back to the governing page.
Interpretation limit: A reuse event is not a unique person, endorsement, proof of trust, proof that the advice is correct or evidence that the resource is the leading source.
Privacy boundary: Store public URLs, community category and artifact behavior only. Do not retain private-group screenshots, usernames, member profiles or inferred demographics without explicit permission.
Stop or review rule: Review after ten qualifying reuse events across at least three independently operated communities or immediately after one consequential distortion that could cause harm.
Record fields: event date; community category; public reference; artifact id and version; publisher initiated; paid placement; boundary preserved; substantive reply; correction raised; response decision

4. beginner-retention-v1
Capability: beginner-retention
Validation question: Do permissioned beginners return for useful learning actions and report bounded colony outcomes beyond an initial click or signup?
Unit of evidence: One permissioned learning enrolment linked only to delivery, unsubscribe, declared return actions and optional outcome check-ins under the existing retention boundary.
Recruitment boundary: Use the separate beginner-learning consent. Do not convert stock alerts, purchases, social followers or readiness attempts into marketing permission or assume that one email represents one person.
Procedure:
  1. Verify consent, first delivery, lesson sequence and unsubscribe behavior in production.
  2. Define each useful return action before the sprint and exclude opens or pixels as sufficient learning actions.
  3. Count delivered lessons, declared actions and voluntary check-ins as separate event types.
  4. Ask optional bounded outcome questions that allow unknown, stopped, delayed and unsuccessful paths.
  5. Preserve campaign attribution without building a cross-site identity profile.
  6. Review whether guidance or acquisition promises differ between retained and non-returning cohorts without claiming causation.
Evidence supporting movement: Repeated permissioned cohorts receive the sequence, perform declared useful return actions and provide interpretable longer-term check-ins that include unsuccessful or delayed outcomes.
Evidence requiring revision: Delivery or unsubscribe fails, acquisition promises outrun the learning path, apparent retention depends on opens or repeat clicks, or outcome collection excludes people who delayed, stopped or struggled.
Interpretation limit: Email rows, deliveries, attempts and actions are not interchangeable unique-person counts. Return behavior does not prove animal welfare, causal impact, customer value or profitability.
Privacy boundary: Retain only consented delivery and bounded learning events for the declared period. Do not add tracking pixels, cross-site profiles, inferred household data or health information.
Stop or review rule: Stop acquisition if consent, unsubscribe or delivery integrity fails. Review after one complete six-lesson cohort and again after the first thirty permissioned enrolment records have reached their eligible follow-up window.
Record fields: cohort id; campaign attribution; consent date; delivery state by lesson; unsubscribe state; declared useful action; follow-up eligibility; optional outcome class; missingness; review decision

5. independent-method-review-v1
Capability: independent-method-review
Validation question: Can qualified independent reviewers reproduce the claim boundary and identify consequential omissions or overreach using the public packet?
Unit of evidence: One completed review under one declared lens with relevant experience, relationship disclosure, exact item, cited source, proposed wording and finite publisher response.
Recruitment boundary: Invite reviewers with relevant aquatic, taxonomic, design, statistical or evidence-synthesis experience who did not author the reviewed wording. Do not count anonymous praise, credentials alone or a paid relationship hidden from the public response.
Procedure:
  1. Provide the versioned independent-review packet without privately steering the expected outcome.
  2. Require the lens, relevant experience and relationship disclosure before evaluating the critique.
  3. Require one exact claim, stable source and proposed bounded wording per challenge.
  4. Classify the response as accepted, partly accepted, not accepted with reason or awaiting source access.
  5. Propagate accepted changes through the correction-impact graph and public change history.
  6. Publish the finite response without presenting reviewer status as proof by itself.
Evidence supporting movement: Completed reviews from more than one relevant lens produce reproducible critiques, including contrary or corrective findings, and every accepted change receives a public finite response and propagation audit.
Evidence requiring revision: Reviewers cannot reconstruct the claim trace, relevant sources are missing, conflicts are undisclosed, critiques remain unanswered or accepted changes fail to reach downstream guidance.
Interpretation limit: One reviewer, credential or positive response is not consensus, peer review of the entire site, veterinary endorsement, independence across every lens or proof of top-source status.
Privacy boundary: Publish reviewer identity only with explicit permission. Always publish the relevant experience category and relationship disclosure; keep contact details and private correspondence outside the packet.
Stop or review rule: Pause invitations when a consequential trace or source-access defect prevents review. Reassess after two completed reviewers spanning at least two lenses and after every accepted material correction.
Record fields: review id; packet version; lens; experience statement; relationship disclosure; item id; source; proposed wording; outcome; public response path; propagation completed

6. independent-adoption-v1
Capability: independent-adoption
Validation question: Do independent publishers, moderators or projects cite, adopt or challenge a governed Molt & Moss artifact without publisher control?
Unit of evidence: One independently initiated citation, standard use, correction participation or sustained referral with a stable reference and disclosed relationship category.
Recruitment boundary: Observe public independent use and accept voluntary reports. Exclude paid placements, reciprocal link arrangements, owned accounts, copied spam, affiliate use without disclosure and private-community surveillance.
Procedure:
  1. Record the exact external reference and governed artifact version.
  2. Classify citation, standard use, correction participation or sustained referral separately.
  3. Check whether the source and limitation remain visible and whether the external use adds an independent decision.
  4. Record relevant commercial, authorship or organizational relationships.
  5. Respond to corrections through the ordinary public workflow without asking for endorsement.
  6. Preserve contrary and discontinued uses rather than reporting only favorable adoption.
Evidence supporting movement: Multiple unaffiliated external actors independently cite, use or challenge governed artifacts over time, preserve their boundaries and make decisions not controlled by the publisher.
Evidence requiring revision: Apparent adoption is paid, reciprocal, publisher-directed, boundary-stripped, concentrated in one relationship network or unsupported by a stable inspectable reference.
Interpretation limit: External use is not proof of correctness, consensus, market leadership, unique reach or permanent trust. Adoption can coexist with substantive criticism and later withdrawal.
Privacy boundary: Use public references or voluntarily submitted evidence. Store no private member identity, inferred relationship, private-group content or unpublished correspondence without permission.
Stop or review rule: Review after three qualifying events from unaffiliated actors in at least two external settings, or immediately when a cited artifact is materially distorted or corrected.
Record fields: event date; event type; external setting; public reference; artifact id and version; relationship category; boundary preserved; independent decision; correction or withdrawal; review note