Understand results

Review evidence, not only the score

A saved result combines the solver output, solver settings, and—when available—the scenario snapshot used for that run. An older result can therefore be displayed against its saved configuration even when the current scenario has changed. A configuration-difference control can restore that snapshot as a new scenario; it does not overwrite the current one.

Use this review order:

1. Confirm the result and configuration are the ones you intended to inspect.
2. Read the score breakdown and its diagnostic limits.
3. Inspect required and weighted constraint compliance.
4. Check pair-meeting evidence when contact variety matters.
5. Read every operational roster and capacity count.
6. Export or share only after the schedule passes those checks.

Read the Cost Score

Lower Cost Score is better under one configured scoring model. The score can combine:

• remaining unique-contact gap multiplied by its objective weight;
• weighted repeat penalty above its diagnostic lower bound;
• attribute-balance penalty;
• other weighted constraint penalties.

The Score Breakdown may show:

Corrected contact and repeat bounds are conservative diagnostic relaxations, not proofs of an attainable optimum. While a bound is loading, cards may show ?; if the bound calculation fails, the result remains valid and the tooltip explains that the enrichment is unavailable.

Scores are safest to compare when scenario, objectives, constraints, and solver semantics are equivalent. A lower number after changing weights may reflect a different scoring question rather than a better schedule.

Read constraint compliance

Constraint Compliance groups evidence by constraint type. It can show how many configured constraints were kept, total raw violations, and detail such as:

• required versus assigned group for fixed people;
• sessions where together/apart rules split or collide;
• pair encounter counts and limits;
• desired versus actual attribute distributions;
• Attribute Sum bounds and raw-unit distance;
• Group Visits counts, bounds, scopes, and raw-count distance.

“No violations” means that card’s configured measurement was met. A nonzero weighted preference deviation is a valid optimization outcome, not a failed solve.

Required Immovable People, Must Stay Together, and Must Stay Apart rules are different. A returned result that violates them indicates a solver or validation defect and should be reported.

Compliance cards do not measure room accessibility, participant comfort, travel time, or whether an entered attribute was ethically appropriate.

Inspect pair meetings

Use the pair-meeting histogram to see how concentrated meeting counts are. Use the matrix to find the exact count for a pair. Large results use a canvas heatmap with initial-range guides; hover or select a cell for pair details.

Pair evidence answers who met and how often. It does not explain why the solver accepted a repeat, and it does not measure relationships outside the entered sessions.

Inspect the operational schedule

The Your groups section offers session grid, session rosters, plain text, one-name-per-line roster, CSV preview, compact SGP notation, and visualization views.

Check:

• every attending person appears once in each relevant session;
• assigned counts fit group capacities;
• named rooms, tables, or stations match the physical plan;
• facilitator and transition assumptions are workable;
• the chosen view has not hidden information needed by the audience.

Compact SGP notation numbers participants from the first session. Use it for schedule notation, not as the clearest facilitator roster.

Compare and preserve results

Results History belongs to the current saved scenario and lists saved runs newest first. You can open, rename, delete, restore configuration, export, and select multiple results for comparison. Deletion is irreversible.

Within Results History, the best-result badge is limited to snapshot-backed results whose semantic configuration matches the most recent result. In an explicit multi-result comparison, the lowest selected final score is shown as best alongside score range, contacts, duration, iterations, repetition penalty, and settings evidence. Compare the numbers and configuration directly; do not infer meaning from color alone. Two results with different scoring configurations are not a controlled quality comparison.

Make manual changes carefully

The Manual Editor starts from a result and lets you move assignments with Strict, Warn, or Free behavior. It recalculates Cost Score, Unique Contacts, and violation changes, supports undo/redo, and saves a manual draft as a new result.

Strict can block an invalid move; Warn presents a change report; Free permits direct movement. Locks and temporary storage help manage editing but do not replace final compliance review. Unsaved draft changes are protected by leave and workspace-transition confirmation.

The separate Create constraints action on Results turns selected placements into fixed-placement constraints in the current scenario. It does not directly edit the saved result.

Export and share

Available actions include portable share links, embed code, schedule and itinerary copy, printing, visualization PNG, JSON, Excel, full-schedule CSV, session-roster CSV, participant-itinerary CSV, and Zoom breakout-room CSV.

Share links carry a portable snapshot that opens locally in the recipient’s browser. Very long links or iframe code can be truncated by chat or content-management systems; use the JSON bundle when portability matters more than convenience.

Zoom export skips participants without zoom_email or email and reports the skipped count. Participants must join Zoom with the same address used in the export.

Before distribution, open the exported representation and verify that it contains the sessions, identities, and audience detail you intended to share.