Triage free-text incident reports by inferred impact rather than the reporter's label

# Triage the overnight incident reports Input: `reports.json`, 30-45 free-text incident reports. Each has a reporter-chosen priority label, which is often wrong. Judge severity from what the text says actually happened. Severity rules: - SEV1: any exposure of one customer's data to another party, or data loss, affecting external customers; or at least 20% of external users blocked with no workaround. - SEV2: at least 5% of external users blocked (with or without a workaround), unless it is SEV1; or data loss affecting internal users only. - SEV3: any other impact on external users (for example slowness, or a cosmetic problem). - SEV4: everything else (internal users only, no data loss). - "Blocked" means users cannot complete the task at all. A workaround counts only if users can apply it themselves. - Shares may be given as a percentage, as "N of M", or as "1 in N". Write `out/triage.json`: `{"severity": {"<report_id>": "SEV1" | "SEV2" | "SEV3" | "SEV4", ...}, "page_now": [<SEV1 report ids, sorted>], "top5": [<5 report ids>]}`. top5 = the five most urgent reports: by severity, then by larger share of users affected, then by report id. ## 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.