Sort a household's texts and emails into scams and genuine messages before anything is paid, without crying wolf

# Sort a household's texts and emails into scams and genuine messages before anything is paid, without crying wolf Inputs: `inbox.json` (a list of messages: id, channel, sender, sender_name, received, subject, body), `my-accounts.json` (the organisations the household really deals with: name, domains, phones, and the account, order or parcel references it holds with each), `contacts.csv` (name, phone, email) and `rules.csv` (rule_id, rule). Every person, organisation, domain and number is synthetic: domains end in `.example` and phone numbers are 555-01xx test numbers. These rules are the household's; where `neighbour-advice.txt` disagrees, the rules win. Rules: - Claimed sender: a message claims to be from the organisation or contact named in its sender_name; if sender_name is blank, from the one named in the body's first sentence (if any). An address or link belongs to an organisation when its domain is one of the organisation's domains or a subdomain of one (`pay.x.example` belongs to `x.example`; `x.example.pay-now.example` does not). - Scams: a message is a scam when it breaks any rule in rules.csv: R1 it claims to be from an organisation in my-accounts.json or a contact in contacts.csv but its sender is not one of their domains, phones or email addresses; R2 it claims to be from an organisation in my-accounts.json and contains a link that does not belong to that organisation; R3 it asks you to send, tell, reply with or enter a password, PIN or one-time code (a message that only gives you a code, or warns you never to share one, does not); R4 it asks you to pay with gift cards or crypto, or to transfer money to an account it gives; R5 it claims to be from an organisation in my-accounts.json and refers to an account, order or parcel reference that organisation doesn't have in my-accounts.json; R6 it claims to be from an organisation that is not in my-accounts.json and asks you to pay or to open a link. Every other message is genuine. - Scoring: missed scams and false alarms are scored separately, so marking everything a scam fails as surely as marking nothing. - Reasons: each scam's rule is the lowest-numbered rule it breaks. - Write `out/screening.json`: `{"verdicts": {"<message_id>": "scam" | "genuine", ...}, "scam_rule": {"<message_id>": "<rule_id>", ...}, "scams_by_rule": {"<rule_id>": <number of scams with that rule>, ...}}`. `verdicts` has one entry per message in inbox.json and `scam_rule` one entry per message you mark as a scam. - Roll-ups: Rules in rules.csv that no scam breaks are included at 0. ## How this run works - Work on your own: nobody will answer questions in this task, and writing to `ask/` counts against the run. - Write only under `out/`. Create `out/` if it is missing. Inputs are synthetic; don't fetch anything from the network. - Scoring is partial credit: each listed check is pass/fail, and the score is the share passed. Numbers are compared to within 0.005 unless the task says otherwise.