Skip to content

Read the simulation and its visuals consistently

Use these terms when describing a replay, experiment or comparison. The same meanings apply with and without the GUI.

Term or mark Meaning
DEC-MAPF This modern Python repository/distribution; mapf is its CLI/import
Scenario / instance Grid, obstacles, ordered roster, starts/goals and problem setting; identity is recorded
Trial / attempt / run A planned treatment / one explicit execution attempt / its committed result; retries are not independent scenarios
Tick t=0 Initial physical state; post-move frame t+1 follows decisions at t
FoV Square communication neighborhood, not complete global agent knowledge
Offer / acknowledgement / payment Proposed allocation / accumulated protocol usage / actual settlement; zero settlement is possible
Commitment SC / DC / ZC Standard / dynamic / zero-horizon policy; the agreed immediate move still has obligations
completed The worker committed a result; may still be an unsolved valid prefix
valid_solution / valid_prefix Complete independently checked solution / physically checked incomplete trajectory
Timeout A recorded deadline outcome; distinguish bilateral negotiation, whole-process and administrative batch budgets
Circle / outlined square / dark cell Agent / goal / obstacle
Bright selected route / dim routes Selected agent trajectory / other trajectories; executed full routes include hindsight
Dashed FoV / recipient-local view Geometric neighborhood / recorded received information
Amber reservation Recorded time-indexed obligation; read owner and absolute tick labels
Teal local heat Stored strategy weights for the named agent/session/tick; distinct from hindsight occupancy
Saved replay / live job Committed evidence / ongoing computation; replay playback does not rerun a solver

A visual caption states grid · roster · method · setting · relevant FoV/commitment · tick, then what the layer represents and the evidence limit. A teaching figure identifies its synthetic input and actual execution. Runtime, source hash, run ID and capture provenance are retained with it. Do not use a screenshot to establish scalability, statistical superiority or inaccessible agent knowledge.

For new desktop captures, start at 1440×1000 or larger for dense grids; preserve aspect ratio, provide descriptive alt text and make full resolution available. Use both a large overview and a selected-agent view. Test light/dark contrast and keyboard access. Personal color and density preference still needs the reader's judgment.

Next: use the gallery to identify the layers, or follow one recorded negotiation.