--- uai: "1.0" type: project-constraints status: active stable_id: iarpg-active-constraints project: iarpg.com release: IARPG-OPS-2 version: "2.0.31-wip" review_state: current-until-superseded --- # Repository Constraints ## Non-Negotiable Constraints - Preserve the fictional intelligence-operations abstraction; do not publish practical real-world wrongdoing instructions. - Do not assign permanent hero, villain, good, or evil status to an authority, country, nationality, culture, religion, or operative role. - Apply the international fairness method: equal evidence burden, explicit uncertainty, rights and oversight review, affected-community context, source-diversity checks, and no national or cultural moral typecasting. Evidence: [fairness methodology](../docs/long-term-memory/reports/equal-standards-across-unequal-records-international-institutional-fairness-rights-and-evidence-audit.md#mandatory-analytical-distinctions-and-rights-safeguards). - Do not use psychiatric diagnosis or distress as a shortcut for guilt, danger, credibility, competence, or alignment. Evidence: [representation audit](../docs/long-term-memory/reports/adversarial-review-of-mental-health-representation-and-serious-game-ethics-psychosis-simulation-module.md#critical-and-major-issue-analysis-and-redesign-specifications). - Generated language may express claims but never owns authoritative game truth, rewards, access, warrants, or mission completion. - Preserve unknown imported fields and original artifacts; never silently overwrite, normalize away, discard, or rewrite invalid input. - Keep only the centered `.menu-strip` sticky. Light is the default theme; a saved dark preference remains local and applies before CSS. - Every completed WIP cycle increments the patch version, produces a complete flat ZIP, external SHA-256 sidecar, release notes, validation evidence, current hot memory, durable updates, and a standalone next prompt. ## Memory Constraints - `.uai/` remains active first-load memory on every startup and handoff. - Hot records are date-free current state: no timestamps, release-history narratives, raw logs, or copied report bodies. - Detailed bodies belong under `docs/long-term-memory/`; every active repository-owned canonical report belongs under `docs/long-term-memory/reports/`. - Hot facts link to durable evidence with repository-relative deep links. Durable documents link back to the hot record that summarizes them. - `.uai/long-term-memory.uai` is pointer-only and must not become a report body store. ## Authority and Conflict Resolution Use this order: explicit repository instructions; verified implementation and tests; active `.uai`; current authoritative durable documents; machine manifests; historical or archived material. Mark unresolved conflict explicitly and preserve useful superseded history. Evidence: [authority audit](../docs/long-term-memory/research/memory-authority-and-conflict-audit.md#authority-resolution). ## Security Boundary Never place passwords, API keys, access tokens, cookies, private keys, recovery codes, or other secret values in memory or reports. A safe inventory may name a credential location and purpose without exposing the credential. ## Planetary Systems Constraints - Preserve the eight source bodies, but do not publish their weapons-effect equations, casualty estimates, cyberattack code, target optimization, or individualized medical guidance. - Do not claim IARPA affiliation or use the fictional expansion of its name. - Do not turn `/scenario-modeling` into a live prediction, medical, weapons, attack, or targeting service. Evidence: [hard boundaries](planetary-systems.uai#hard-boundaries). ## Country Context Composer Constraints 1. Never auto-promote archival World Context facts into current claims. 2. Keep current observations, assumptions, and fictional translations visibly and structurally separate. 3. Require source date, access date, identity, URL, type, scope, confidence, perspective origin, language, and review state for every current claim. 4. Never calculate country development, stability, morality, threat, loyalty, dangerousness, recruitability, targetability, diagnosis, or protected-characteristic scores. 5. Preserve imported unknown extensions and invalid originals without mutation. 6. Keep all work browser-local with no accounts, analytics, remote storage, or model-provider dependency. Evidence: [decisions and recommendations](../docs/long-term-memory/reports/country-context-composer-and-comparative-fairness-report.md#decisions-or-recommendations). ## Evidence Attribution Review-Readiness Constraints - Preserve all twelve source packets, seventy-two adjudication slots, twelve source requirements, twelve wording tests, ninety-six source questions, and two hundred sixteen claim questions without filling reviewer-owned fields. - Do not infer reviewer identity, qualification, language competence, local relationship, affected-community standing, conflict disclosure, consent, findings, requested wording, or adjudication. - Keep claim status, analytic confidence, source reliability, attribution rung, origin chain, translation state, correction history, reviewer finding, and adjudication structurally separate. - Do not score sources, reviewers, communities, organizations, jurisdictions, credibility, susceptibility, threat, loyalty, legitimacy, morality, protected characteristics, recruitability, or targetability. - Original-language and locally authored source requirements are research prompts, not evidence that such review occurred. Model-generated wording tests are not approved translations or editorial judgments. - Keep actor claims, platform findings, intelligence assessments, court allegations, formal judgments, reach, causal effect, symbols, roles, membership, sponsorship, direction, command, metadata, inference, and decision use distinct. - No live reviewer endpoint, account system, remote database, identity-verification service, analytics, automated publication, or confidential-submission storage is authorized. Evidence: [review-readiness protocol](../docs/long-term-memory/reports/evidence-attribution-human-review-readiness-and-translation-quality-protocol.md#safety-and-non-scoring-boundary) and [active routing](evidence-attribution-review-readiness.uai#current-publication-contract). ## Publisher Contact Boundary - Preserve the `/contact` route and global-footer `Contact Us` link. - Keep employer names, client history, and private project details out of the IARPG profile. - Keep public contact values out of active UAI and durable reports; route readers to the public page. - Do not add a contact form or remote intake system without separate privacy, security, retention, abuse, accessibility, legal, and operational review. - A published residential correspondence address does not imply a staffed office, invitation to visit, identity verification, or consent to profiling or harassment. Evidence: [implementation findings](../docs/long-term-memory/reports/publisher-contact-and-public-profile-page-implementation-report.md#findings) and [active routing](publisher-contact.uai#current-publication-contract).