Every scenario is a JSON object with version: 1. Numeric strings, non-finite numbers and truthy substitutes for booleans are refused. The engine does not modify input. Extra fields are ignored; use synthetic data in public examples.
min_sessions: integer 1..1000. observations: at most 10000 objects, nonempty zone/session up to 200 chars, kind in passage_obstruction/uneven_surface/missing_sign/other, integer severity 1..3, strict boolean confirmed. Zone labels must be anonymized before public sharing.
Group exact zone/type pairs; count distinct sessions rather than duplicate rows. A finding needs repeated sessions and at least one human-confirmed observation to enter the review queue. Sort by declared severity then session count, using lexical tie breaks. Public summaries omit notes, coordinates, images, people and session IDs.
Two duplicate rows from visit-1 count as one session. The library has two distinct sessions and a human confirmation, so enters review_queue. The hall has two sessions but no human confirmation, so remains needs_evidence.
No computer vision, autonomous inspection or accessibility certification. A robot may accompany the operator or carry a permitted camera; a person enters observations and confirms findings.
PT: A entrada usa JSON na versão 1. As unidades estão nos nomes dos campos. Erros de contrato interrompem a análise. Campos adicionais não viram evidência. A implementação e os exemplos não demonstram desempenho do robô nem benefício social já validado.
A finding is one exact (zone, kind) group. Distinct sessions are counted with a set; duplicate_rows is row_count - distinct_session_count within the group. Maximum entered severity is a prioritization signal, not an externally verified severity assessment. One human-confirmed row establishes the group's confirmation flag, while the distinct-session requirement is evaluated separately.
The sensitivity table evaluates minimum visits 1 through 4 without changing confirmation flags. Lowering a threshold never substitutes for a person confirming the observation. A group can be both unconfirmed and below threshold, so these categories may overlap.
Zone summaries count findings, review items and items needing evidence. Outputs omit session identifiers and arbitrary input fields; synthetic zone labels themselves remain visible and must be anonymized before sharing. No image processing, geographic inference or accessibility certification is performed.
Grouping is O(n) and deterministic sorting is O(g log g), where g is the number of grouped findings. At most 10,000 input observations are accepted.
PT: Um achado agrupa zona e tipo exatos. Linhas repetidas não criam novas visitas. Confirmação e mínimo de visitas são avaliados separadamente; categorias de pendência podem se sobrepor. O relatório não certifica acessibilidade.