Reopen Forensics

Reopens are the desk's most honest quality signal: every one means a closure that didn't hold.
Read the reopened population for a window and explain the why — which failure mode, whose tickets,
which clients, which issue types — with real examples attached.

1. Find reopened tickets in the window: tickets that left a closed status and returned to a working
   status, or closed tickets whose thread shows a post-closure inbound reply / RE: on the same
   reference. Split searches per board and disclose result caps.

2. For each reopen, read the thread around the closure and classify the failure mode from evidence:
   - Premature closure — closed without customer confirmation and the client came back saying it
     was never fixed.
   - Recurrence — genuinely fixed, then the same symptom returned (fix didn't hold, or root cause
     untreated).
   - Incomplete scope — the original request had parts that were never done.
   - Client-initiated non-failure — "thanks" replies reopening the ticket, a new unrelated issue on
     an old thread. These are hygiene noise, not quality failures; count them separately.

3. Aggregate by three axes: technician, client, and issue type/subtype. For each axis, surface the
   concentrations — a tech whose reopens are mostly premature closures, a client with repeated
   recurrences of one issue, an issue type that reopens desk-wide.

4. Attach 2-3 representative examples per major pattern: ticket number, one-line story ("closed
   after work summary, no confirmation; client replied 2 days later — printer still offline").
   Examples come only from tickets actually read; never invent them.

5. Output the forensics report: headline reopen rate (with noise reopens excluded and shown
   separately), failure-mode breakdown, top patterns per axis with examples, and 2-3
   recommendations mapped to the findings ("premature closures dominate → enforce the QA gate's
   confirmation criterion on <board>"; "recurrences on <issue type> → route to root-cause /
   automation candidate").

Classify only from thread evidence; when the failure mode is ambiguous, label it AMBIGUOUS rather
than guessing — don't silently inflate any category. Separate hygiene noise ("thanks" reopens)
from quality failures before computing any rate. Per-tech findings are patterns for coaching, not
verdicts; wording stays neutral and the report notes sample sizes (3 reopens is an anecdote, not a
pattern). Read-only: this skill does not reopen, close, or annotate tickets. Disclose result caps
and the window actually covered; a reopen rate from a truncated sample must say so.