The budget is a real budget
How output gets ranked, cut, kept connected, and re-measured until the reported number is the delivered number.
Most map-for-agents code truncates and hopes. supercarto treats truncation as a correctness problem, with four mechanisms:
Ranking
Named landmarks survive; anonymous intersections are cut first. An intersection named “2nd & Mission” ranks higher than an unnamed junction 100 m away, because an agent can act on a name.
Connectivity
Cut vertices are never removed, so the surviving network stays connected. A map the agent can traverse but not connect is worse than a smaller map it can use — “how do I get from here to there” has no answer on a severed graph.
No stranded names
A landmark the agent can see but cannot route to is removed. It would confidently try to walk to it — the same hallucinated-route failure mode in miniature, introduced by the cutting step itself.
Verification
emitAndVerify re-measures the actual emitted YAML and cuts again
until it fits. The reported budget is the budget delivered, not an
estimate on an intermediate representation. This is what makes the number a
budget rather than a target. Why you can trust the geometry the budget
delivers: the accuracy page shows distance verification
across latitudes.
Honesty
Every drop is recorded in omitted:, aggregated per layer, so the model
can tell “not there” from “not shown”. An agent combining supercarto’s
output with other sources, or reasoning about a pharmacy that does not
appear, can reason about the difference:
omitted:
- {layer: building, count: 259}
- {layer: poi, count: 141}
- {layer: road, count: 441}