# Actual full-domain spatial refinement:10us cost/accuracy pilot

Frozen2026-11-13, before new native observations. Full C1 remains unqualified.
Return from same-dx layout sensitivity and flat area-spectrum diagnostics to
the ACTUAL coupled film. Uniform20references are frozen background-space
trajectories atdt.0625/.03125us. New uniform10,N544,radius500um,SDS6mM,
insoluble, variational preparation,physicalcut240um,domain320x160um,initialh6um.
No gravity/DLVO, kinetics, material, force, mobility, tolerance or solver change.
This remains a short released-edge assay, not sustained frame drainage.

Read/verify original uniform20raw hashes; never rerun/overwrite references.
Current native binary must equal frozen layout-phase execution. Verify the
24previously executed native checks through their exact log/binary hashes,
without presenting reuse as new test execution. Snapshot all current native
thin_film source hashes plus helper/runner/README and full commit.

Two new one-step uniform10guards dt.0625/.03125us, observer on,N544. Require
native complete/failures0, force-target ratio<=1.000001, all four global/region
V/S ledgers<=1e-10,energy-budget excess<=1e-7,prep pressure<=1e-8.
If both guards and cost gate pass, run10us at eachdt, saving step1and10us.
Independent SI/GEOS scalar+mechanical audits at saved steps; all accepted
native history/force criteria retained. Compare original moving-cell-overlap
dh/displacement at10us: original2%TIME and10%SPACE thresholds unchanged.
Compare10/20um at EACH matchingdt. Do not use only easier aggregate h.

Only the first of original10/50/200us checkpoints is present in this pilot.
Even if it passes, full original time/spatial qualification is false. The
objective still requires50/200us, actual global spatial convergence and
independent physical flux/traction; no reference is continuum truth. Retain
same-dx sensitivity and earlier manufactured-area FD failures as failures.

Cost predictor=max(frozen prepared_motion guard_uniform10wall5.516s,
two new guard walls)*steps*1.3. This is an engineering overestimate, not
validated scaling. New observer/startup costs may dominate the one-step guard.
Before matrix, require both predicted horizons jointly fit remaining6600s and
each<=5400s. Guards<=120s, each matrix<=5400s,total native<=6600s (<2h).
If a guard/cost gate fails, preserve all partial outputs and record skips;
never restart a timeout. No >2h job, hence no monitor automation for this batch.
Longer50/200us extension must use measured cost and, if>2h, half-hour monitoring.

Use measured10us runtimes to estimate50/200us costs only after completion,
with explicit uncertainty. Do not shorten the full objective to fit this pilot.
No extra physics/tuning or shape-correction forces to manufacture agreement.
All native raw bytes preserved in bounded ZIPs; white Nature red/bluePNG and
vectorPDF. Distinguish numerical prototype, physics and experimental evidence.