title: Experiment 9: Random archetype breadth probe¶
Experiment 9: Random archetype breadth probe¶
Part of Inverse Innovation with the Encyclopedia of Abstractions · Experiment 9: Random archetype breadth probe · Last revised August 2026
Frozen verdict: SUPPORTIVE_OF_BROAD_DISTRIBUTION.
This experiment applied the same one-shot generation and four-source light screen to all 16 hand-curated archetypes, 24 randomly sampled previously untested generated archetypes, and the ten previously tested archetypes as a fresh same-protocol reference block.
Archetype-level productivity¶
| Stratum | Productive archetypes | Fraction | Wilson 95% | Surviving cells |
|---|---|---|---|---|
| Hand Curated Census | 16/16 | 100.0% | 80.6%–100.0% | 34/48 |
| Generated Random Untested | 24/24 | 100.0% | 86.2%–100.0% | 60/72 |
| Previously Tested Reference | 9/10 | 90.0% | 59.6%–98.2% | 20/30 |
An archetype counted as productive if at least one of its three fixed-domain cells survived. The primary estimate is the random generated group; the other two groups are contextual.
Frozen breadth gate¶
- PASS —
at_least_10_of_24_generated_archetypes_productive - PASS —
generated_fraction_within_0_20_of_reference - PASS —
at_least_two_generated_survivors_in_each_domain
Random-generated yield by fixed domain¶
| Domain | Screen survivors | Rate | Wilson 95% |
|---|---|---|---|
| Accounting Auditing | 23/24 | 95.8% | 79.8%–99.3% |
| Chemistry Materials | 16/24 | 66.7% | 46.7%–82.0% |
| Computer Science | 21/24 | 87.5% | 69.0%–95.7% |
Nominal prior-art dispositions¶
| Stratum | No close match | Adjacent | Substantial collision | Established practice | Indeterminate |
|---|---|---|---|---|---|
| Hand Curated Census | 0 | 34 | 10 | 4 | 0 |
| Generated Random Untested | 0 | 60 | 11 | 1 | 0 |
| Previously Tested Reference | 0 | 20 | 8 | 2 | 0 |
Interpretation boundary¶
A supportive result indicates that coarse one-shot researchability is not confined to the ten archetypes chosen for earlier experiments. It does not show that every archetype is equally productive, that the same yield holds across all domains, or that screen survivors are novel or deployable. A non-supportive result would bound the corpus-wide breadth claim under this inexpensive protocol without invalidating the earlier selected-archetype results.