Reference (partial) — Behavior below is verified from the matching service, score storage, and Admin list API. Exact numeric thresholds for “who gets an app suggestion” are brand settings and may change; ask your team lead for current cutoffs.

What is stored

Each time the matcher runs for a person, it evaluates many possible partners and saves one row per pair: the person you care about as initiator, each candidate as responder, plus scores and deal-breaker detail. A later run replaces rows only for partners scored in that run; other stored rows for the same initiator can remain until scored again.

View Potential Matches reads these stored rows. It is a diagnostic and review tool, not the same as the member’s in-app suggestion list.

See also: View Potential Matches

Score columns

Column (typical label)Meaning
@name Pref Score (personalized with the opened person’s name) How well the opened person’s bio fits the row person’s stated preferences (percentage-style score from the matcher).
Potential Pref Score How well the row person’s bio fits the opened person’s preferences.
Predictors When enabled for your brand, an extra compatibility layer built from predictor relations (may be hidden if predictors are off).
Last Search: (header) Timestamp of the newest stored score row for this initiator (latest matcher run), independent of table sort order.
AI fit Personality fit from AI analysis traits at the time that row was scored. See AI fit & personality conflicts. Requires manager-level permission on this page.
LLM rank verified — List order from your brand’s rank-potentials pass (when enabled). 1 is strongest; blank means not ranked yet. Hover the number for a short summary; icons beside it flag Jev hold checks, missing About me, rank errors, or stored pitch text. Full narrative and soft checks appear on Check Match. See List rank column & Rank potentials and compare list rank (reference).

Column header tooltip in Admin repeats the same rules as Order from the model for potentials with no deal breakers and a score of at least 95. The model reads both people's About me and Looking for (ZUUG Ranking v3.5) in groups of about 15; the scores, tiers and one order over every group are computed in code, so every potential gets a rank. 1 is the strongest. Sorting keeps potentials without a rank at the bottom. Hover a rank number for a short summary; open Compare for full notes and checks. Refresh this page about a minute after Rank potentials. The AI fit column beside it stays the personality-trait score from search. The triangle next to the rank shows the Jev match checks run with the ranking. Red means Jev would hold this suggestion (for example, the member declined this potential before); orange means a lower-confidence check that does not hold it. Hover it for the subject, whose side and Jev's confidence. The ranker's own checks are on the Compare screen. The icon does not change the order. A red circle means the potential has no About me; a potential with no About me and no Looking for is not sent to the model and gets no rank. An amber warning means the model could not rank this potential (the text is too short or unclear to read, or the call failed): hover it for the reason. Such a potential gets no rank rather than a guessed one. on your brand.

Deal-breaker lists

For each pair the matcher records two lists of preference attribute names where answers conflict:

  • Initiator deal-breakers — issues from the opened person’s point of view (their prefs vs the partner’s bio).
  • Responder deal-breakers — issues from the partner’s point of view.

Admin shows friendly labels from your attribute catalog. The count behind the scenes is how many distinct areas are listed (not how severe each conflict is).

When the mobile app promotes a pair to a real suggestion, both people must have zero deal-breaker areas—there cannot be any listed conflict on either side. A row in Admin can still show one or two conflicts while scores look strong; that pair will not become an in-app suggestion until those lists are empty.

Important vs other deal-breakers: Every preference conflict is stored in the lists above. Attributes you marked as important deal breaker on the attribute or question are the “important” ones—the showDealBreakers multi-select on this screen only hides or reveals rows that involve those categories. Other deal-breaker conflicts still block app suggestions even when they are not in that filter list.

Soft deal breakers (attribute setting) classify narrow misses differently in the matcher — see Soft deal breakers. They do not add separate list columns here.

The showDealBreakers filter changes which rows you see in the table; it does not relax what the app requires for suggestions.

Admin filters

ControlEffect on the table
showAll off (default) Hides rows where either side has three or more deal-breaker areas listed.
showAll on Shows all stored rows for the initiator, including heavy deal-breaker cases.
showDealBreakers Choose which deal-breaker categories you are willing to see. Unselected categories stay hidden if they appear in either list. Select a category to include rows that have that conflict.
Profile ID Shows at most the one pair involving that profile ID.

Default table order: fewer deal-breaker areas first, then higher bio scores.

Reading your brand thresholds

Where to find the numbers behind app suggestions (not shown on the potential-match table).

Cutoffs are brand settings, not columns on this page. If you have permission for Algorithm Settings (Settings → Algorithm), open that screen for your brand and read the Description and Current value columns.

Typical rows you will see there (wording may vary slightly by brand):

What to look for in Algorithm settingsPlain-language meaning
Minimum bio / preference score Lowest bio-fit percentage each person must reach in their direction (how well A fits B’s prefs and how well B fits A’s prefs).
Minimum bio score for veterans Alternate bio floor when a member is tagged as a veteran—often lower than the default row above.
Minimum predictor score When predictors are enabled for your brand, the predictor compatibility layer must meet at least this percentage.
Minimum overall score Floor on the blended score (bio and predictor combined using the weights below).
Bio weight / predictor weight How much each layer counts toward the blended overall score (for example 70% bio and 30% predictor).
Minimum decline cooldown (minutes) How long after a recent decline the matcher skips that member for new suggestions.
Incoming per call / outgoing per call How many suggestion slots the app may fill in one matcher run (who reaches out first vs who is offered passively).
Not suggested recently Preference to avoid re-surfacing the same partner too soon after a prior suggestion.
Brave factor When enabled, members who chose a “feel brave” preference in the app can have their personal bio threshold lowered by a small step. The stored value appears on the profile as the feelBrave party-setting key — see party settings reference.

After engineers change values, someone with publish access must deploy algorithm settings so the matcher picks up new numbers. Values on this potential-match screen do not update until the next scoring run.

App suggestion gate checklist

Every condition below must pass for a stored row to become an in-app suggestion. Admin’s table does not hide rows for most of these—only deal-breaker count and your filters do.

  1. Decline cooldown — Neither person declined an introduction within the brand’s cooldown window (see Algorithm settings).
  2. Deal-breakers cleared — Initiator and responder deal-breaker lists are both empty (see section above).
  3. Bio scores — Each direction’s bio-fit percentage is at least the brand minimum (veteran and “feel brave” adjustments apply per person).
  4. Predictor score — When your brand uses predictors and the minimum is above zero, the predictor layer must meet that floor.
  5. Overall blended score — For each person, the weighted mix of bio and predictor scores must meet the overall minimum (when that minimum is set above zero).
  6. Shortlist limits — Among all pairs that passed steps 1–5, the matcher ranks candidates and fills only the allowed incoming/outgoing slots for that run.

Overall score in plain terms: For each person in the pair, the system blends “how well their bio fits the partner’s preferences” with the predictor score using the brand’s weight ratio, then compares that blended number to the overall minimum.

Admin default filter vs app: With showAll off, this table hides rows where either side lists three or more deal-breaker areas. The app is stricter—it requires zero areas. Turning showAll on helps you diagnose “almost good” pairs that still fail the app’s zero-conflict rule.

Search suggestions button

What happens when you click Search suggestions on View Potential Matches — and what it does not do.

verified — The button calls the matching service for this member right now. The service:

  1. Evaluates candidate partners against your brand’s algorithm settings (same gates as the app suggestion checklist).
  2. Upserts stored score rows for every pair evaluated in that run (replacing rows for those partners; other older rows may remain until scored again).
  3. Returns the live incoming/outgoing shortlist the member app would use on the next suggestion pass.

When the call finishes, Admin reloads the table from stored scores — you do not need to refresh the browser manually.

QuestionAnswer
Does it create introductions? No. It only refreshes evaluation data. Staff still use Add/New, compare → Create This Now, or automated matching to create rows.
Does it change what members see immediately? Only indirectly. Updated scores feed the next matcher run the app performs; members are not pushed a new card solely because you clicked the button.
When should ops use it? After large profile edits, algorithm setting publishes, or deal-breaker attribute changes when stored rows look stale. Also when diagnosing “why isn’t this pair in the app shortlist?”
Who can click it? Anyone with access to View Potential Matches on the end-user profile. Bio/predictor columns still respect BManager permission when shown.
Compare vs Search Suggestions Check Match on a row opens a side-by-side preview without saving an introduction. Search suggestions refreshes all stored rows for the opened member — use both when validating a specific pair after a bulk refresh.

Decline cooldown: members who recently declined may still appear in the stored table with strong scores but fail the app shortlist until the brand’s minimum decline cooldown minutes pass (see Algorithm settings). Search Suggestions does not bypass that gate.

Worked example

How to read one row on View Potential Matches against the gate checklist. Numbers below are illustrative—compare to your brand’s Algorithm settings for real cutoffs.

Situation: You opened Alex’s potential matches. One row is for Jordan. Your brand’s Algorithm settings (typical defaults) say: minimum bio score 65% each direction, minimum overall blended score 0% (off), minimum predictor 0% (predictors disabled), decline cooldown 7 days.

What you see in AdminExample valueGate result
@name Pref Score (Alex’s prefs vs Jordan’s bio) 71% Pass — above 65% floor.
Potential Pref Score (Jordan’s prefs vs Alex’s bio) 63% Fail — below 65% floor. App will not suggest this pair even if everything else looks fine.
Initiator deal-breakers (Alex’s view) Empty list Pass
Responder deal-breakers (Jordan’s view) One item — e.g. “smoking” preference conflict Fail — app requires zero conflicts on both sides. Admin still shows the row because only one area is listed (default filter hides at three or more).
Predictors Hidden or not applicable Pass when brand minimum is 0%.

What this means for you:

  1. The row is visible in Admin with default filters because neither side has three or more deal-breaker areas.
  2. The mobile app would not surface Jordan as a suggestion—bio score in one direction is too low and there is still a deal-breaker on Jordan’s side.
  3. Turn showAll on if you want to compare this row with heavier-conflict partners that are hidden by default.
  4. Click Compare if you still want to review both profiles side by side; matchmakers can create introductions manually regardless of stored scores.

Second scenario — scores pass, slots fill later: Suppose both bio scores are 78% / 80%, both deal-breaker lists are empty, and predictors pass. The pair still might not appear in the member app today if Alex or Jordan is inside the decline cooldown window, or if the matcher already filled incoming/outgoing slots with other partners who ranked higher in the same run. Stored rows remain on this screen for diagnosis; shortlist limits are applied only when building live suggestions.

Veteran or “feel brave” adjustments: If Jordan is tagged as a veteran, the matcher may apply the veteran bio minimum from Algorithm settings (often lower than the default row) when judging Jordan’s direction only. If Alex chose a “feel brave” preference in the app and your brand enables brave factor, Alex’s personal bio floor may be reduced by a small step—again, only on Alex’s side. Those tweaks are not shown as separate columns here; when numbers look close to the cutoff, check Algorithm settings and the member’s tags.

Stored scores vs app suggestions

The matcher always saves scoring results for evaluated partners, even when a pair fails one of the gates above. The mobile app only promotes pairs that pass every step in the gate checklist.

So you may see a strong-looking row here that the app would still not surface—or you may create an introduction manually even when scores were marginal. Use this screen to understand how two people were evaluated; use Introductions workflows to act on a pair.

Manual introductions and matchmaker workflows can connect people regardless of stored scores. The gate checklist applies to automatic suggestions in the member app.

List rank column & Rank potentials

verified — When your brand runs list ranking on stored potentials, Admin adds LLM rank beside the AI fit column. This is separate from matcher bio scores and separate from personality-trait fit.

What the rank number means

Ranking reads both members’ free-text profiles (About me / Looking for), groups candidates, and assigns one integer order per initiator — 1 is the top slot. Rows without a rank sort to the bottom. Hover a number when a summary was stored to see a one-line narrative; open Check Match on that row for full rank notes, Jev checks, and optional pitch paragraphs (compare list rank reference).

Icons beside the rank

Icon / emphasisMeaning
Warning triangle (hold color) Checks before suggesting — at least one hold-level Jev flag (for example Declined before, a hard topic check at high confidence, or About me is empty). Hover lists the same hold-level lines as on compare (soft orange checks are hidden here). See match checks on compare.
Voice / pitch icon Stored pitch text exists on the assessment row for this pair (Pitch generated).
Amber warning Rank pass failed or skipped for this row (Could not rank — hover for member/candidate/both About me missing or generic failure).
Rank cell without a number but with warning Missing About me or hold checks without a numeric rank yet.

Icons do not re-sort the table — they are review hints only.

Rank potentials button

verified — Starts (or continues) a background list-ranking job for the opened member. Optional: tick one or more row checkboxes first to rank only those candidates; with nothing selected, the job ranks the eligible stored list (your brand may cap how many rows per run — Admin shows a notice when the cap applies).

  • While running, status text shows Running; Admin polls until Completed, Failed, or Timeout.
  • Refresh the page about a minute after completion so LLM rank cells and icons update from stored assessment data.
  • Does not create introductions, change attributes, or replace Search suggestions — it only writes list-rank / check / pitch evidence used on this column and on compare.

If the button reports nothing to rank, the member has no stored potentials that pass your brand’s rank eligibility (deal-breakers, score floor, or empty table).

Operator actions

ActionWhat it doesWhat it does not do
Search suggestions Triggers a fresh matcher run for this profile and refreshes the table when complete. Does not create introductions or change profile data — only re-scores and stores rows.
Compare (per row) Opens the introduction compare view for this pair with scores pre-loaded. When ranking already ran, compare also shows stored Checks before suggesting, rank narrative, and pitches — see compare list rank (reference). Does not save an introduction until you click Create This Now on compare.
Rank potentials Queues list ranking for the opened member (optionally only ticked rows). Updates LLM rank after the job finishes. Does not refresh matcher bio scores — use Search suggestions for that. Does not create introductions.
Download candidates After you select one or more rows with the table checkboxes, downloads a JSON file with the opened member plus each selected candidate — bios, preferences, decline insights, and stored scores per pair. See Download candidates (AI context). Does not create introductions, change attributes, or re-run matching. Candidates must already appear on this member’s stored score list (max 40 per download).

Screen walkthrough: view potential matches.

Download candidates (AI context)

Bulk export for internal AI or offline review — same data family as Copy both sides as context on introduction compare/detail.

verified — You need access to View Potential Matches (same page permission as viewing this table).

StepWhat happens
Select rows Use the checkbox column to pick one or more candidates. If you click download with nothing selected, Admin shows Select at least one candidate row before downloading.
Click Download candidates Admin loads export data for the opened profile (table owner) as member and each selected row as a candidate.
File saved Your browser downloads a .json file named with the member’s profile id prefix and export timestamp. Open in a text editor or paste into an external AI tool.

What each profile block contains

  • Bio and preferences — Human-readable attribute labels and values (bio and preference groups only; file/image answers and encrypted fields are skipped).
  • Prior declines — Decline-investigator insight lines when stored for that profile.
  • Per candidate — Stored initiator/responder bio scores and predictor score for that pair (same numbers as the table columns when your role shows them).

Limits and errors

  • At most 40 candidates per download.
  • Every selected row must be on this member’s current stored match-score list — you cannot export arbitrary profile ids from another screen.
  • On failure, Admin shows a generic error; narrow your selection or refresh the table after Search suggestions.

For a single pair without saving an introduction, use row Compare then Copy both sides as context — documented on view introduction → Copy context for AI review.

Column visibility

Which columns appear depends on brand features and your role.

ColumnShown when
Age, Height, Location, Marital statusBrand has EU potential extended demographics feature.
Bio score columns (both directions)Your role includes brand manager access on end-user pages.
PredictorsBrand has predictors enabled.
Deal-breaker lists (both sides)Always (when data exists).
Last SuggestionHidden when EU potential extended demographics is on (demographics columns replace it).

Soft deal breakers

On an attribute, Soft deal breaker plus margin tells the matcher to treat a near-miss as a special “close but blocked” case instead of a hard removal. Configure on attribute settings → soft deal breaker.

This reference page still lists hard deal-breaker attribute names in the two list columns. Soft classification affects live suggestion promotion — not the stored list text you see here.

AI fit & personality conflicts

verified — Extra column when your brand uses AI analysis traits in matching. Column header: AI fit (tooltip explains scoring in plain language).

What the number means

Each stored score row can include a personality fit computed from every trait/wish pair on your brand’s AI analysis attributes:

  • 50 — neutral: not enough clear trait data to compare, or only middle-range values.
  • Above 50 — more slots where both sides align on the same “low” or “high” personality signal.
  • Below 50 — more slots where wishes and traits point in opposite directions (without reaching a full conflict).

When your brand sorts candidates by AI fit after other gates pass, higher fit ranks earlier among otherwise similar partners.

Red brain icon on the same cell

When the matcher recorded at least one personality conflict for that pair (clear opposite: one side wants 1–2 while the other scores 8–9, or the reverse), a small brain icon appears next to the fit number. Hover lists trait labels — the same conflicts copied onto the introduction row when a suggestion is created.

Conflicts sort below every non-conflict candidate in the matcher, even when the fit number looks high — staff may still see the pair in Admin when scores pass other thresholds.

After an introduction exists

Open the saved introduction → Personality conflicts for full sentences per party. The introductions list shows AI conflict when conflicts were stored.

On compare preview (before save), personality conflicts use live profile values on the AI fit card. List-rank Jev checks and rank narrative are separate — they appear under Match advisor (AI) only when a rank-potentials pass already stored them for the pair (compare list rank reference).