Skip to content
Simurgh UI Simurgh UI

Privacy-preserving research evidence

Simurgh does not collect icon searches, copied snippets, account data, or interaction events from the documentation site. Product telemetry is disabled. Discovery evidence comes from moderated, consented sessions using the versioned template in research/icon-study-session.template.json.

Study protocol

Section titled “Study protocol”
  1. Explain the tasks, data fields, retention period, and right to withdraw; record consent before starting.
  2. Create a random session code used only for this study. Do not connect it to an account or a recruitment list.
  3. Ask the participant to find an icon from a concept, compare similar candidates, copy the import for their assigned framework, theme it, label an icon-only control, and verify RTL behavior.
  4. Record only booleans, elapsed seconds, counts, the assigned framework, and predefined issue codes. Do not record screens, voices, arbitrary queries, application data, or free-form notes.
  5. Let the participant withdraw during or after the session. Delete withdrawn records immediately.
  6. Delete raw session records after 30 days. Publish only cohort aggregates containing at least five sessions; suppress slices smaller than five.

Store completed records as separate JSON files under the ignored research/sessions/icons directory. After at least five valid sessions, generate a publishable aggregate without session identifiers:

Terminal window
pnpm research:icons

The command rejects missing consent, withdrawn records, stale policy versions, duplicate session codes, free-form or prohibited fields, incomplete metrics, unsupported issue codes, and cohorts smaller than five. It writes only aggregate rates, counts, and framework cohort sizes to artifacts/icon-study-aggregate.json; review that file before committing it.

Allowed metrics

Section titled “Allowed metrics”

The machine-readable policy permits only these fields:

  • search-success
  • time-to-first-correct-icon-seconds
  • zero-result-query-count
  • framework-import-copy-success
  • theme-task-success
  • icon-only-label-task-success
  • rtl-verification-success
  • mistaken-icon-substitution-count

Use structured issue codes for taxonomy, alias, comparison, copy, accessibility, and RTL problems. Do not store the participant’s exact query: a moderator may increment the zero-result count and select a predefined taxonomy code instead.

Prohibited data

Section titled “Prohibited data”

Never collect or retain name, email, ip-address, account-id, screen-recording, or free-form-product-data. Recruitment and scheduling records stay outside the study dataset and are deleted under the recruiting service’s own policy. Do not join study codes to repository, website, support, or package-registry activity.

The repository check validates the policy, template, retention ceiling, cohort floor, prohibited fields, and metric names. Changing the protocol requires review before another session begins; completed sessions retain the policy version used when consent was obtained.

Last verified on 2026-08-13 against Simurgh registry 0.1.1.