Job Fit Scoring

Job Fit Scoring

Score and rank open roles against the user's profile with an explicit, weighted rubric, so effort goes to the roles most likely to convert, and stretch/missing requirements stay honest and visible.

When to use

• Right after a job-sourcing sweep adds status=found rows
• "Which of these should I actually apply to?", "rank my pipeline", "score this one job I found"
• Re-scoring after the profile changes (new comp floor, new target titles) or a JD is updated

Inputs

• Reads: workspace/profile/profile.md (achievements, skills, hard constraints), workspace/profile/target-roles.md (titles, must-haves, deal-breakers, comp floor, location), workspace/profile/target-companies.md (company-ICP + tiers), workspace/pipeline/tracker.csv (rows to score, defaulting to everything at status=found), and each role's applications/<id>-*.md (the raw JD)
• Needs from user: nothing if profile/ + found rows exist. Optionally a subset to score, or a manual override on a deal-breaker. If profile/ is empty, route to job-search-setup.

Workflow

1. Select rows. Default to all status=found. Accept a user-specified subset or a single role.

2. Read the real JD. Score against the raw JD text in applications/<id>-*.md, not the tracker summary line. If the JD wasn't captured, fetch/paste it first.

3. Check deal-breakers FIRST. If the role trips any deal-breaker in target-roles.md (wrong location/remote, below comp floor when comp is listed, missing work authorization, on-site-only, industry the user excluded, IC-vs-management mismatch), cap fit_score at 39 (priority C) and write the disqualifying reason as the rationale. Do not let strong scores elsewhere paper over a hard no. If comp isn't listed, it can't trip the comp deal-breaker. Note the unknown instead.

4. Score the 8 dimensions (0–100 total). Award each dimension's points on a 0-to-max scale, then sum:

   #	Dimension	Max	What earns full points
   1	Role / title match	20	Title (or a known variant) matches the target, and the core function is the job the user wants
   2	Seniority match	10	Level lines up with the user's band, not a step down and not two steps up
   3	Must-haves met	20	The JD's required qualifications are things the user genuinely has (evidence in profile/master-resume)
   4	Company-ICP fit	15	Company matches the ICP rubric (stage, size, industry, model) and/or sits in Tier A/B
   5	Comp fit	10	Listed band clears the comp floor with room. Full points only if comp is disclosed and comfortably above floor. Unlisted comp = award 5 (unknown), note it
   6	Location / remote fit	10	Matches the user's location/remote rule cleanly
   7	Trigger / timing signals	10	Fresh funding, new exec building a team, product launch, hiring surge, backfill of a role the user is perfect for, all reasons this is a good moment to be their candidate
   8	Stretch / growth	5	The role stretches the user in a direction they want (upside), not a reach that will read as unqualified

   Score to real evidence. If the JP demands 8 years and the user has 4, dimension 3 is not full marks. Say so. Missing requirements lower the score, and they never get rounded up because the role is exciting.

5. Set priority. A ≥ 80 · B 60–79 · C < 60. Deal-breaker caps land in C by construction.

6. Advance status. Move A and B rows from found → researching (they're worth the research/tailoring investment). Leave C at found (or the user can withdraw). Set a sensible next_action (e.g. "company-research + find referral") and next_action_date for A/B rows.

7. Write the rationale. In each role's applications/<id>-*.md, fill the Fit assessment section with a one-to-three-line honest rationale: the score, what matched, what's a stretch, what's missing. Gaps stay visible. This is what the human reads before deciding to apply.

8. Update tracker + print shortlist. Write fit_score, priority, status, next_action, next_action_date back to tracker.csv. Print a ranked table to chat, highest first.

Output

• Writes to:
  • workspace/pipeline/tracker.csv – fit_score, priority, updated status, next_action, next_action_date per row
  • workspace/pipeline/applications/<id>-*.md – one-line-plus rationale in the Fit assessment section
• Format: a ranked chat table of rank | id | company | role | fit | priority | one-line rationale (incl. the biggest gap). Deal-breaker disqualifications are shown with the reason, not silently buried at the bottom.

Rules & quality bar

• Transparent weights, shown on request. The rubric above is the whole scoring model, with no hidden judgment. If the user asks "why 74?", show the per-dimension breakdown.
• Never inflate. A stretch role scores like a stretch role. Missing must-haves cost points every time. Optimistic scoring wastes the user's applications on roles that won't call back.
• Deal-breakers are hard caps, not deductions. They cap at 39 regardless of other strengths. A role that fails work authorization is not a "78 with an asterisk".
• Comp unknown ≠ comp good. Do not award full comp points for an undisclosed band. Award the midpoint and flag it.
• No fabrication of qualifications. Score against what's actually in profile/master-resume. Don't credit a skill the user doesn't have because the JD wants it.
• Re-score, don't accumulate. Re-running overwrites prior scores for the selected rows so the tracker reflects current profile + JD, not a stale pass.

Related skills

• Reads from: job-sourcing (the found rows) and job-search-setup (the profile it scores against).
• Hands off to: company-research and networking-and-referrals (for A/B roles now at researching), then resume-tailor and cover-letter when the user commits to applying. application-tracker owns the shared tracker schema this skill writes to.