# Fuel quality and station reliability **2026-10-14 D deferral — adopted maintainer restriction:** Future D holds private purchase/document verification and attachment features until extreme app consolidation plus thousands of real clients, unless the maintainer explicitly requests earlier named scope. Official ANP history and non-document reports remain independent. Rules below are future target semantics, not permission to collect documents or implement D now. Owner: FQ-00–FQ-17 implementation plan. Existing price, account, feedback, media, trust and profile rules retain their meanings. FQ-12 separately freezes numeric score methodology. This file changes no API or runtime policy. ## Product promise For each canonical station, expose dated official evidence by fuel, distinct enforcement/sanction records, clearly identified consumer experiences and a versioned explanation of available reliability measures. Every station can have a quality profile; lack of observations means insufficient evidence. A user can inspect all safely matched published official cases, open their source, request correction of a station association and attach a related experience. User votes/claims do not validate chemical quality or legal guilt. ## B-BR-FQ-01–16: proposed behavior and acceptance - **B-BR-FQ-01 — Source separation.** GIVEN PMQC, enforcement, sanctions, community reports, ratings and price confidence, WHEN presenting a station, THEN expose their provenance/meaning separately. No user command edits an official finding; PMQC is not an infraction notice. - **B-BR-FQ-02 — Exact historical identity.** GIVEN a source event, WHEN binding it, THEN use a valid full establishment CNPJ and explicit chronology/review. A name/address/nearby point, CNPJ root or different branch cannot establish the binding. Unmatched/ambiguous cases remain in quarantine/review, not lost. - **B-BR-FQ-03 — Lossless samples and assays.** GIVEN several assay rows for a sample, WHEN importing, THEN retain each distinct assay/result/unit/conformity and count the sample independently. Unknown fields/fuels/status stay explicit. Missing or conflicting native identity never silently merges samples. - **B-BR-FQ-04 — Versioned official facts.** GIVEN a revised resource, WHEN publishing, THEN retain hash/source/schema/native-key provenance and append correction/supersession/withdrawal. Identical content is idempotent; changed content cannot overwrite historical evidence or double count a sample. - **B-BR-FQ-05 — Complete-run visibility.** GIVEN a failed/partial import or concurrent replay, WHEN serving, THEN retain the previous complete published revision. Counts reconcile all inputs and all matching dispositions; a URL label alone never proves a period's coverage. - **B-BR-FQ-06 — Honest procedural status.** GIVEN an action/fine/proceeding snapshot, WHEN displaying, THEN preserve actual scope/status/as-of date. Autuation, payment, absence of later records or unverified linkage never means definitive guilt, acquittal or current legal status. - **B-BR-FQ-07 — Public and private authorization.** GIVEN a guest, WHEN reading official public history, THEN allow bounded reads. GIVEN a write/private owner read, THEN require current authenticated account and verified caller binding; reject forged owner/station/event/fuel and revoked or foreign proof. - **B-BR-FQ-08 — Claims versus checked purchases.** GIVEN a consumer report, WHEN assigning evidence level, THEN distinguish user claim, supplied document and independently checked purchase. An image/OCR/valid-looking key/CNPJ alone never verifies a purchase, adulteration or ANP protocol. Failed/unavailable verification stays explicitly unresolved. - **B-BR-FQ-09 — Temporal correlation.** GIVEN a related official case and consumer experience, WHEN linking, THEN verify station/fuel/revision and show dates. Proximity in time does not prove the same batch/tank, cause or diagnosis. - **B-BR-FQ-10 — Independent community signals.** GIVEN related accounts/keys or reused receipts, WHEN aggregating, THEN avoid multiple support for one attribution/purchase. Distinct accounts are not guaranteed distinct humans. Thresholds need frozen evidence; unresolved abuse stays under review. - **B-BR-FQ-11 — Score sufficiency and isolation.** GIVEN too little/stale/ unmapped evidence, WHEN reading a score, THEN return null with reason/counts. Eligible per-fuel scores have frozen sample semantics/window/rounding and version. Payment, rating, brand, profile badge, raw report count and pending non-quality proceedings cannot affect the proposed v1 numeric score. - **B-BR-FQ-12 — Reproducible dated explanations.** GIVEN identical eligible official facts, policy and injected clock, WHEN rebuilding, THEN produce the same score/flags/counts independent of input order. Return as-of, expiry, coverage and input/policy revision. Expired views are stale even if jobs stop. A current favorable score cannot hide a dated nonconformity. - **B-BR-FQ-13 — Fuel scope.** GIVEN evidence for only one fuel, WHEN presenting the station, THEN identify its coverage. Gasoline findings cannot establish diesel quality; a station-wide numeric index needs its own accepted aggregation policy. Every station can expose an unavailable score-state object. - **B-BR-FQ-14 — Privacy and retention.** GIVEN receipts/workshop/lab documents, WHEN storing or rendering, THEN keep them private, minimize identifying fields and apply the established operational media ceiling of 24 hours. Private protocol/CPF/keys/GPS/object references never appear publicly or in logs. Training consent is unrelated. Erasure/restore replay and checked-status degradation are documented before persistence. - **B-BR-FQ-15 — Review and recovery.** GIVEN disputed linkage, report abuse, source correction or erasure, WHEN deciding, THEN append actor/reason/policy, adjust eligibility and enqueue affected rebuilds atomically. Corrected wrong links are removed from the active wrong-station view; appeals cannot rewrite source records or grant station representatives moderation power. - **B-BR-FQ-16 — Measurable complete access.** GIVEN one frozen complete source edition, WHEN traversing the public case API/app, THEN every matched eligible event is reachable exactly once and each has source/date/detail. Record the denominator plus unmatched/quarantined/missing periods. Offline cached content is dated; queued writes are not server-acknowledged validations. ## BUC-FQ-01–09 and required scenarios - **BUC-FQ-01:** Discover/prepare/stage/publish official evidence. Test malformed archive/schema/period/encoding, limits, loss accounting, deduplication, partial failure, retry/concurrency and prior-publication preservation. - **BUC-FQ-02:** Resolve/review station linkage. Test full-CNPJ branch distinctions, leading zeros/alphanumeric values, missing/historical identity, conflicting mappings, succession and corrective replay. - **BUC-FQ-03:** Revise/retract/recompute. Test corrected assays/identity/status, source resource replacement, duplicate annual/monthly data, atomic visibility, stale jobs and rebuild convergence. - **BUC-FQ-04:** Inspect official station/fuel history and case detail. Test guest access, source separation, no sample, partial fuel coverage, stale data, procedural labels, bounded complete pagination and source links. - **BUC-FQ-05:** Submit/edit/withdraw own structured experience and request linkage review. Test signed owner proof, idempotency/replay, invalid targets/text/dates, foreign-owner denial, parallel writes, moderation and source immutability. Android refinement (2026-11-10): the Profile entry "Meus relatos privados" is the owner's history. A report always targets one station and one fuel, so the history never creates a station-less report: when the signed owner list is empty it explains where reports are created and offers one action that opens the station list; an existing report can still be selected there to edit, appeal or withdraw. Guests see only the free sign-in action. Report text, dates and status stay in memory and behind the secure window. - **BUC-FQ-06:** Attach/check/minimize/expire private supporting documents. Test fabricated/reused/foreign evidence, unavailable verifier, false protocol claim, expiry races, every-copy deletion, erasure and restoration. - **BUC-FQ-07:** Detect/review/appeal a community pattern. Test related attribution, account farms, duplicate purchases, conflicting official history, insufficient independence, audit and no numeric official-score contamination. - **BUC-FQ-08:** Export/erase/recover private quality data and command state. Test logout/account/origin isolation, expiry, restored deletions, redacted audit, lost-response convergence and no resurrection of checked/private evidence. - **BUC-FQ-09:** Explain/rebuild available reliability measures. Test hand-calculated sample denominators, no/sparse/stale inputs, exact arithmetic, input reordering, fuel coverage, policy changes and chronological recovery without history loss. Critical SQL/auth/privacy/jobs scenarios use immediate real PostGIS and actual affected storage/transport boundaries; pure rules use deterministic RED → GREEN → REFACTOR. Runtime completion requires those tests, not this target document. ## Proposed UI language and limits UI is localized; the following Portuguese phrases convey the intended meaning: - "Amostra ANP/PMQC: conforme em [data]" or "Ensaio não conforme registrado". - "Ação de fiscalização registrada"; show source-proven procedural status. - "Relato do usuário", "Documento apresentado" and "Compra verificada" only where a recorded independent verification supports the latter. - "Dados insuficientes para calcular a nota"; show fuel/date/coverage reasons. - "Nota baseada nas amostras históricas disponíveis"; link methodology and source denominator. Avoid "combustível garantido", "fraude comprovada" without appropriate official proof, or claims of ANP certification/affiliation. - "Informar divergência neste registro", "Relacionar minha experiência" and "Consultar canal oficial da ANP". No automatic complaint transmission. Concrete wire enums, endpoints, legal/privacy copy, verification providers and score thresholds are frozen by their owning tasks; none exists merely because its target label appears here.