/seekter-report — how is the search going

Read-only. Data comes from applications/ via scripts/seekter.mjs, plus runs/*.md for per-day source tables.

1. node scripts/seekter.mjs stats (all time) and node scripts/seekter.mjs stats --since <7 days ago>.
2. node scripts/seekter.mjs list --since <period start>, then read the front matter of the rows you need (source, ats, location_fit, fit).
3. Compute, for the period and all time:
   • applications sent, skips, pending hand-offs;
   • replies: interviewing + offer + rejected, as a share of sent; median days from applied to the first log line after it;
   • the breakdowns, each one command (node scripts/seekter.mjs stats --by <field> [--since …] [--min 5]), which print sent, replies and interviews per group:
     • source (freehire, linkedin, board names, manual) and apply_type (company form, Easy Apply, email);
     • ats, the form system the application went through;
     • location_fit (A/B/C). A track with many sends and no replies is a targeting signal;
     • title and seniority, derived from the role ("Senior Product Designer (Growth)" counts as product designer · senior);
     • fit and resume, once they are filled: whether the fit score predicts replies is the question the score exists to answer, and the resume type is what the run uploaded;
     • month, to see whether the funnel is moving. An interview is a record that reached interviewing or offer, even if it was later rejected or closed. Replies include rejections. Rates are over decided applications (a reply, or closed); one still applied is "in flight", has its own column and counts in no rate, so a week of fresh sends does not pull a group down. A group with fewer than 5 decided shows n<5 instead of a rate (--floor changes it): quote its counts, not a percentage. With only a handful of interviews, compare groups on replies and say how many interviews there are rather than quoting interview rates;
   • top skip reasons (group the Reason column of applications/<YYYY-MM>/skipped.md by its first clause).
4. Every record with status: pending is a hand-off (node scripts/seekter.mjs list --status pending; also the "Needs you" table in applications/README.md): list each with its link and the exact action.
5. Write 3–5 plain observations with the number behind each ("C-track: 14 sent, 0 replies in 3 weeks"). Suggest at most two changes, each tied to a file (profile/settings.json query, profile/profile.md rule). Don't change them without the user saying yes.

Save as runs/report-<YYYY-MM-DD>.md and show the user the short version.