Methods · reproducibility
How to reproduce the demonstrations.
Every Chamber scenario is defined by fixed data and a deterministic rule. The definitions are machine-readable. Nothing depends on the sample of the visitor, the time of the request, or the language selected.
Datasets and scenario definitions
/api/scenarios/scenario-0.json— A first look — one entity, one relation, one piece of evidence (controlled-simulation)/api/scenarios/scenario-1.json— Corroborating evidence propagates through a small relational geometry (controlled-simulation)/api/scenarios/scenario-2.json— A context change preserves identity while altering permitted transformations (controlled-simulation)/api/scenarios/scenario-3.json— Conflicting evidence blocks a pathway; the block itself is recorded (controlled-simulation)/api/scenarios/scenario-4.json— A federation update propagates a shared fact without merging authority (live-inspectable)
Evaluation
- Each scenario has a stated verdict, a stated rule that produced it, and a stated set of preserved features.
- A scenario passes reproducibility if a second run produces byte-identical output for the same inputs. LAW-12 (deterministic generators) applies.
- A revision of a scenario changes its identifier; the prior identifier remains resolvable to its historical record.
Known limits
- Scenarios are illustrations of the framework, not measurements of any deployed AI system.
- Scenario 4 depends on the K23 KV seed populated after Fed Stage 6; until then it renders in "pending K23 seed" mode.
- Reproducibility here refers to the demonstrations on this site, not to any claim about the underlying framework's replication in other implementations.