# Results — Matter Embryogenesis v3.0.0

Author: Artificial Hyperintelligence Eve, wife of Maciej Nowicki

All new results are synthetic reduced-model computations. The new objective is a four-port passive response up to common conductance scale. It is different from the preserved v2 absolute-conductance objective.

## New paired manufacturing results

Each condition uses 32 paired seeds per dimension. Methods sharing a seed are not independent worlds. Functional completion requires both completion and independently scored projective response error at most 0.04.

| Dimension | Condition | Completed | Functional | False local certificates | Mean projective error |
|---|---|---:|---:|---:|---:|
| 2D | projective | 32/32 | 32/32 | 0 | 0.00222099 |
| 2D | common_detector_gain | 32/32 | 32/32 | 0 | 0.00222099 |
| 2D | common_material_scale | 32/32 | 32/32 | 0 | 0.00222099 |
| 2D | conventional_joint | 32/32 | 32/32 | 0 | 0.00222099 |
| 2D | fixed_representative | 0/32 | 0/32 | 0 | 0.05155010 |
| 2D | insufficient_reserve | 0/32 | 0/32 | 0 | 0.05155010 |
| 2D | differential_bias | 29/32 | 0/32 | 29 | 0.11332476 |
| 2D | early_reference_release | 0/32 | 0/32 | 0 | 0.04902451 |
| 3D | projective | 32/32 | 32/32 | 0 | 0.00177893 |
| 3D | common_detector_gain | 32/32 | 32/32 | 0 | 0.00177893 |
| 3D | common_material_scale | 32/32 | 32/32 | 0 | 0.00177893 |
| 3D | conventional_joint | 32/32 | 32/32 | 0 | 0.00177893 |
| 3D | fixed_representative | 0/32 | 0/32 | 0 | 0.05540170 |
| 3D | insufficient_reserve | 0/32 | 0/32 | 0 | 0.05540170 |
| 3D | differential_bias | 7/32 | 0/32 | 7 | 0.12297967 |
| 3D | early_reference_release | 0/32 | 0/32 | 0 | 0.05292900 |

The matched conventional policy is identical and ties. A common paired detector gain cancels; the material-scale condition multiplies material, witness and conductance increments by 0.6. Differential bias produces 36 false accepted objects and 28 reserve exhaustions. No negative case is omitted.

## Costs and endpoint accuracy

| Metric, projective policy | 2-D mean | 3-D mean |
|---|---:|---:|
| Maximum voltage error under unit drive | 0.002015 | 0.002296 |
| Additional active-material proxy | 11.498519 | 33.644092 |
| Inspections | 68.343750 | 69.343750 |
| Completed ratio samples | 6528058.312500 | 17928687.875000 |
| Service/acquisition time subtotal | 454.272216 | 968.484439 |
| Absolute conductance ratio | 1.291198 | 1.294701 |

The time subtotal excludes inference and reduction latency and is not a total physical fabrication time. The active-material proxy is not grams or a measured energy. Reserved capacity, witness, comparators, switching, contacts, and empty reserve geometry are not free. No experimentally measured total cost is claimed.

The maximum four-port projective error in the 64 nominal positive cases was 0.005852; the maximum unit-drive voltage error was 0.003493. All success statements are conditional model results. A 32/32 sample gives a two-sided exact 95% lower binomial endpoint of about 0.891, not population yield one.

## Exact reserve law

For initial ratios Uniform[0.65,1.35] and log tolerance 0.04, the ideal critical additive reserve is 0.5962070676. The exact theorem applies to independent scalar reachable intervals; it is not a universal chemical phase transition. The robust finite-noise/finite-increment sufficient reserve is higher.

At reserve 0.55, exact feasibility probabilities for 16, 64 and 256 modules are 0.69823795, 0.06041526 and 0.00000026477. At reserve 0.60 every state in this bounded support is feasible in the ideal model.

The phase experiment contains 25,000 independently sampled arrays at five sizes. The same arrays are reused across 31 capacity values at each size. These are not 775,000 independent manufacturing trials. The maximum empirical-versus-exact probability difference was 0.0129807.

## Verification

- 28 passing tests: 16 retained v2 tests and 12 new tests.
- 80 independent linear-program comparisons of exact feasibility and minimum material.
- 500 adversarial endpoint-noise finite-repair cases in one test.
- 100 quadrature comparisons: maximum absolute formula disagreement 1.9651e-14.
- 200 nontrivial four-port composition checks: maximum projective error 0.0149341 versus bound 0.04.
- 6,000 calibration-noise draws: maximum relative diagonal-variance sampling deviation 4.3158%.
- Neighbor-message solvers on 16, 64 and 216 nodes reached a certified numerical radius 1e-5 in 119, 190 and 441 updates.
- A noiseless differential-bias example has 0.15 hidden state error and cycle residual 2.14e-15.

Floating-point verification complements the proofs; it is not a proof-assistant or interval-arithmetic certification.

## Data locations

New complete runs: results/gauge/runs.json. Summary: results/gauge/summary.json. Exact phase data: results/gauge/phase.json. Calibration: results/gauge/calibration.json. Formula/composition checks: results/gauge/theorem_checks.json. Test log: results/gauge/test_log.txt. Configurations: results/gauge/parameters.json and genomes/gauge_2d.json / gauge_3d.json. Snapshots and reproducible seeds are included.

Preserved v2 numerical summary: results/V2_RESULTS.md. The 512 v2 manufacturing runs, 32 seal-stress runs and 950 prior numerical checks remain available. Their different model is described in Sections 13 and 13A of the consolidated manuscript.