A fixed controller waits for local hot spots to settle before it moves on. Flow sees the complete solidification field, balances the blocked condition across the active faces, and advances only after the conservative transfer is verified. The result is coordinated whole-build control instead of a serial queue of isolated regions.
Five complete software builds. Same targets. Same quality checks. Lower is better.
Why it matters: Flow does not make the controller scan faster. It removes serial waiting from the control law by acting on the solidification field as one conservative object.
Cycles are measured controller steps against the fastest admissible fixed schedule for the same modeled build. The final 4,160-region result was held blind before execution. Public evidence receipt.
| build | controlled regions | fixed cycles | Flow cycles | fewer cycles |
|---|---|---|---|---|
| declared | 1,728 | 1,056 | 584 | 44.70% |
| held-out | 2,240 | 1,350 | 728 | 46.07% |
| fresh | 2,816 | 1,683 | 888 | 47.24% |
| blind | 3,456 | 2,059 | 1,065 | 48.28% |
| actuator-envelope blind | 4,160 | 2,473 | 1,259 | 49.09% |
Flow does not assume unlimited actuation. The software searched all 1,715 registered resource configurations and lowered each accepted whole-field action to a finite schedule. It found 561 nominal-feasible and 401 single-resource-fault-feasible configurations. Exact componentwise reduction leaves 12 nominal and 9 single-fault-tolerant nondominated points, each with a constructive subslot schedule. The envelope spans 2–16 lanes, 4–32 global transfer units per subslot, and 64 down to 8 subslots per frame.
All five builds completed with zero modeled lack-of-fusion, over-limit, first-solidification-debt, or energy-ledger violations. Across the repair bank, 184 repair certificates carried 301 matching sweeps, 1,686 conservative face transfers, and 5,407 integer enthalpy units. Every repair reached the exact registered gradient bound of 24 under a strictly decreasing square potential.