Reference (partial) — Compare preview behavior traced October 2026 from the compare API and Match advisor (AI) card on compare routes. Rows marked verified match Admin today.

Overview

When you open Check Match from potential matches or the add-introduction dialog, Admin loads a read-only preview of two members. If your brand has already run list ranking for that pair, the compare page can show stored evidence under the same card title as Match advisor (AI) — even though no introduction row exists yet.

This is not the same job as Match Advisor on a saved PENDING APPROVAL introduction. Compare reads a shared assessment store (pair-level ranking, Jev match checks, and optional pitch text). Saved introduction detail runs (or re-runs) Match Advisor after create and may show scoring tables, Run match advisor now, and contradiction counts tied to that introduction id.

Workflow guides: view introduction → Compare-only view, create introduction → Compare view sections.

When the section appears

verified — On compare, the Match advisor (AI) card opens when any of the following is present for the initiator ↔ responder pair:

  • List rank detail — narrative summary and/or side counts from the list ranker.
  • Match checks — Jev yes/no flags stored with the ranking pass (see below).
  • Stored pitches — initiator, responder, or matchmaker pitch text from the same assessment row (plus optional hard-contradiction bullets).

If none of those exist yet, the card is hidden on compare — that is normal when ranking has not run for this pair. It does not mean Match Advisor failed on a saved introduction (there is no saved row yet).

The left-nav link to this card on compare appears only when your brand enables Match Advisor insights; the card body can still render when ranking data exists even if that nav link is hidden.

SituationWhat you see on compare
Pair never ranked No Match advisor (AI) block. Use bio scores and attribute tables only; run Rank potentials (or your brand’s equivalent batch) if ops needs list evidence before manual create.
Ranking ran; checks only Checks before suggesting lines (red / orange) without rank narrative.
Ranking ran; full detail Match checks plus LLM list ranking (rank position, summary, side counts, optional tier / lower-rank reason).
Pitches stored on assessment row Pitch paragraphs and optional Hard contradictions list; timestamp and stale warning when profiles changed after pitches were generated.

Match checks (Jev)

Checks before suggesting — automated yes/no questions about the pair (location, education, earlier decline, text that contradicts a profile, empty About me, and similar).

verified — Each line shows a human topic label, which side it applies to (member's side / potential's side), and a confidence percentage when the model scored the question.

Color / emphasisMeaning for ops
Red text Hold-level — includes Declined before when the pair should not be suggested, hard flags at or above the review threshold, or About me is empty on a side.
Orange text Soft band — Jev “yes” score from about 60% up to (but below) that question’s hold threshold (default threshold 70%, so roughly 60%–69%). Shown on compare for context; does not alone trigger the match-scores table warning icon or block automated pending release. Tune thresholds on match check questions → worked examples.
Muted / indented sub-line Ranker disagreed with Jev — the flag is shown for audit with the ranker’s reason; treat as “review both readings.”
Indented quote / field lines under a flag Ranker agreed — supporting quote or profile field the check relied on.

On the potential-matches table, a warning icon uses the same hold-level rules (soft band and disputed flags are suppressed). Hover the icon on a score row for the same flag text as on compare. Full column guide: potential match scores → List rank column.

verified — Hold-level hard flags use Jev’s “yes” score at about 70% or higher unless the ranker marked the flag disputed (then the table icon is suppressed but compare may still show the muted line).

Common Jev topics

Topic labels come from your brand’s match-check catalog or built-in keys. Wording on screen uses the same translation keys as Admin tooltips.

Typical line labelPlain meaning
Declined beforeStop — pair was introduced before and someone declined; do not suggest again without staff override.
Earlier declineSame stop family when stored under the built-in earlier-decline key.
Their words vs. profileFree text on one side contradicts the other profile (About me / Looking for mismatch).
About me is emptyRequired About me text missing on the flagged side.
LocationLocation / geography mismatch between profiles.
EducationEducation level or school story does not align.
JobJob / career mismatch.
Learning plansLearning plans or study paths conflict.
FamilyFamily size / children expectations conflict.
Home typeHome type or community mismatch.
HashkafaReligious outlook / hashkafa mismatch.
PersonalityPersonality description conflict (distinct from AI trait fit column).
OtherCustom catalog topic or legacy fallback label.

Brands may add more catalog topics — those show as custom human labels on the line (from matchCheck General Codes). When unsure, read the ranker sub-line under the flag on compare. Configuration guide: Match check questions (reference).

List rank insight

LLM list ranking — how the list ranker ordered this candidate for the member you are viewing as initiator.

UI elementPlain meaning
List rank: @rank verified — This candidate’s position in the member’s ranked list (1 = top). Blank when ranking ran but did not assign a numeric rank.
Summary paragraph Short narrative of why the ranker placed the pair here (staff-facing, not member comms).
Member wants shown / Candidate wants shown verified — shown / of counts (how many preference or profile facets were visible to the ranker), optional rating label and points when stored.
You may also want to check: bullets Extra plain-text checks and quoted conflicts the ranker recorded (field + value + optional firmness).
Tier: @tier Quality band label when the ranker assigned a tier (wording is brand/model specific).
Lower: @reason Why the ranker scored this candidate below the top tier when a lower-rank reason was stored.
Last updated timestamp When the assessment row was last updated (ranking pass time, not introduction create time).

Stored pitches on compare

verified — When pitch text exists on the assessment row, compare reuses the Match Advisor pitch layout:

  • Suggested message for the initiator / Suggested message for the responder / Matchmaker pitch — suggestion copy for staff (not editable templates).
  • Hard contradictions — bullet list when hard contradictions were stored with the pitch pass (same styling as saved detail when contradictions exist).
  • Pitches may be outdated — one or both profiles changed after they were generated. — warning when either profile changed after pitches were generated; re-run ranking or Match Advisor on the saved row after create if ops needs fresh text.
  • Advisor could not complete — pitch generation failed; compare still shows other ranking evidence when present.

Compare vs saved Match Advisor

TopicCompare previewSaved introduction detail
Requires introduction id No Yes
Match Advisor insights app feature Nav link optional; body may show ranking store data without it Card hidden when feature off
Run match advisor now No Yes (with Edit)
Scoring disagreement / wrong-quote feedback No Yes when scoring table enabled
List rank + Jev checks Yes when assessment exists Not the primary surface — use compare or potential matches before create
Red contradiction count on introductions list N/A (no list row) After advisor runs on saved PENDING APPROVAL row