Back to dashboard

Match tuning report

Internal read-only review for deciding how the profile-scoped matcher should improve after enough real feedback exists. Nothing here applies rule changes yet.

Collecting evidence

Not ready yet. Collect match ratings first so the tuning report has real evidence.

0Feedback rows0Completed scans500Decisions read10Scans remaining
Rules version: compiled-randall-private-2026-06-12-offshore-hubsConfig source: compiled profilePoor ratings: 0Strong ratings: 0

Guardrails

  • Hard exclusions mean wrong for this profile/search context only.
  • V1 uses approve, reject, edit, and limited strength choices — no percentage sliders.
  • Applying changes requires a preview-impact step and a new matcher config version.
  • Candidate-facing cards stay simple: short reasons, Weird match, and rating feedback.

Batch review and learning queue

Pick one fit verdict for each role, then add rationale chips that explain the choice. Internal matcher states stay diagnostic; the human feedback model is Match, Good, Stretch, or Not a Match.

Candidate poolReturned for reviewPreviously reviewed omitted204Saved verdicts
Review learning summary

59 of 204 saved decisions have no written reason. Blank reasons still inherit logic from similar title-signal clusters.

5 match · 51 good · 28 stretch · 120 not a match
Strategy, product, operations leadershipNot a Match · 13 roles

it says europe in the top of the posting

8 decisions using inherited logic
Producer and production rolesNot a Match · 11 roles

Requires Creative Titles in work experience.

8 decisions using inherited logic
General program managementGood · 9 roles

stretch because data/IT

8 decisions using inherited logic
Producer and production rolesGood · 9 roles

low match because hybrid, not remote. Open AI is specifically a company that does allow for remote on case-by-case so it's ok to show these but stretch because hybrid.

8 decisions using inherited logic
General program managementNot a Match · 9 roles

no IT

7 decisions using inherited logic
Marketing rolesStretch · 9 roles

I am realizing that my resume does have a product title in it. so these are correctly filtered. i have not experienced great feedback from aplying to product roles but for this exercise, we will keep these coming since it is a bit of a match.

7 decisions using inherited logic
Marketing rolesNot a Match · 8 roles

I have already reported that marketing managers and any marketing focused role is not suitable so this should not have been shown again.

4 decisions using inherited logic
Strategy, product, operations leadershipGood · 7 roles

Inherited from repeated similar good-verdict decisions without a new written reason.

7 decisions using inherited logic

Preview matcher changes

Select draft signals from manual review, then preview how recent decision evidence would move before any matcher config can be saved.

14Current included14Preview included0Added visible0Removed visible0Need live replay
Added

No visible additions in the latest preview.

Removed or downgraded

No removals or downgrades in the latest preview.

Warnings
  • 8 saved decisions cite only scrape/source issues; they feed source QA and were excluded from matcher drafts.
  • This is preview-only. It does not write a matcher config, mutate historical decisions, or change live scan behavior.
  • Matcher apply/rollback storage is not enabled in V1.

Profile-scoped exclusions

Wrong for this profile/search context only. These are not universal judgments about a role.

0 suggestions

No evidence-backed suggestions yet.

Title families

Potential recall improvements for valid target-role families the matcher may be missing.

0 suggestions

No evidence-backed suggestions yet.

Positive signals

Signals that can boost a match, but cannot bypass the title-family gate.

0 suggestions

No evidence-backed suggestions yet.

Negative signals

Signals that should reduce confidence without becoming absolute exclusions.

0 suggestions

No evidence-backed suggestions yet.

Threshold presets

Small preset changes to bucket strictness. No raw percentage sliders in V1.

0 suggestions

No evidence-backed suggestions yet.