Scientific manuscript · Closed study

Oracle Primitive Model

Shared procedural computation for transfer to withheld domain–operation combinations.

Protocol
v1.1.4
Primary H1
Supported
Mechanism
Not supported
Study state
Closed · 27 Aug 2026
Chapter 05 of 09

Authoritative sealed-test results

The primary behavioral hypothesis was supported: shared primitives produced a large, seed-consistent recombination advantage while preserving interpolation performance.

Primary finding

A strong behavioral result.
An incomplete causal explanation.

OPM_SHARED preserved near-ceiling performance on combinations withheld from training while the principal domain-generalist control fell to 57.16%. This supports the preregistered behavioral hypothesis, but it does not by itself establish the complete mechanism claim.

+42.62percentage points
95% paired bootstrap interval+30.19 to +49.23 pp

Δgeneralist on the sealed recombination test · 10,000 paired two-level bootstrap replicates · PCG64DXSM seed 99117

5.1 · Primary analysis

What the primary estimate means

The primary estimand was the paired difference between OPM_SHARED and DOMAIN_GENERALIST on the three withheld domain–operation cells. Model seeds and latent worlds were resampled in a two-level bootstrap. The observed difference was 0.426200, and the entire 95% interval remained above the preregistered +0.02 support threshold.

The interpolation non-inferiority conditions also passed. OPM_SHARED achieved 0.999347 on observed combinations, compared with 0.997264 for DOMAIN_GENERALIST and 0.997583 for PROC_UNTIED. The result therefore cannot be explained as trading ordinary interpolation competence for recombination performance.

5.2 · Authoritative outcomes

Sealed-test accuracy by condition.

Values pool the five declared primary seeds. Recombination is the preregistered H1 endpoint; interpolation, renderer, and structural tests provide context.

OPM shared primitivesOPM_SHARED
99.78%
Domain generalistDOMAIN_GENERALIST
57.16%
Procedure untiedPROC_UNTIED
50.00%
Procedure clonePROC_CLONE
50.00%
Pooled sealed-test accuracy across five declared model seeds
ConditionInterpolationRecombinationRendererStructural
OPM shared primitivesOPM_SHARED0.9993470.9978110.9993540.998979
Domain generalistDOMAIN_GENERALIST0.9972640.5716110.5907290.890854
Procedure untiedPROC_UNTIED0.9975830.4999890.7384170.873667
Procedure clonePROC_CLONE0.9978610.5000000.7492920.873854
Performance atlas

One model profile across all four sealed tests.

Interpolation was nearly saturated for every condition. Separation emerged when rules had to transfer to withheld combinations and altered renderings.

Observed combinations

Interpolation

OPM_SHARED
99.93%
D_GENERALIST
99.73%
P_UNTIED
99.76%
P_CLONE
99.79%
Withheld combinations · H1

Recombination

OPM_SHARED
99.78%
D_GENERALIST
57.16%
P_UNTIED
50.00%
P_CLONE
50.00%
Surface robustness

Renderer

OPM_SHARED
99.94%
D_GENERALIST
59.07%
P_UNTIED
73.84%
P_CLONE
74.93%
Structural robustness

Structural

OPM_SHARED
99.90%
D_GENERALIST
89.09%
P_UNTIED
87.37%
P_CLONE
87.39%
Declared-seed stability

Recombination performance for every primary run.

Each mark is one frozen model seed. OPM_SHARED stayed between 99.17% and 99.97%, and every paired seed comparison favored OPM_SHARED.

OPM sharedOPM_SHARED
1101
99.95%
2202
99.17%
3303
99.97%
4404
99.94%
5505
99.88%
Domain generalistDOMAIN_GENERALIST
1101
50.30%
2202
50.41%
3303
51.81%
4404
82.03%
5505
51.26%
Procedure untiedPROC_UNTIED
1101
50.00%
2202
50.00%
3303
49.99%
4404
50.00%
5505
50.00%
Procedure clonePROC_CLONE
1101
50.00%
2202
50.00%
3303
50.00%
4404
50.00%
5505
50.00%
Paired Δgeneralist by seedEvery declared seed favored OPM_SHARED.

The smallest paired advantage was 17.91 points at seed 4404; the pooled lower-bound support threshold was 2 points.

1101
+49.65 pp
2202
+48.76 pp
3303
+48.16 pp
4404
+17.91 pp
5505
+48.62 pp
Calibration

Expected calibration error exposes baseline overconfidence.

Lower is better. OPM_SHARED remained below 0.002 on every split; non-sharing baselines were sharply miscalibrated on recombination and renderer tests.

ConditionInterpolationRecombinationRendererStructural
OPM shared
0.0003ECE
0.0018ECE
0.0003ECE
0.0007ECE
Domain generalist
0.0018ECE
0.4254ECE
0.4085ECE
0.1078ECE
Procedure untied
0.0016ECE
0.4865ECE
0.2591ECE
0.1224ECE
Procedure clone
0.0012ECE
0.4973ECE
0.2485ECE
0.1249ECE

5.3 · Result explorer

Interrogate the sealed metric cube.

Choose a model condition, test split, and analysis dimension. The explorer reads directly from the frozen baseline-metrics package.

Loading the metric cube…

Explore all 168 frozen dimension-level accuracy rows. Download dimension-accuracy.csv for analysis.