Repeated goal reuse
1 → 0external sensor calls
12 → 4Living ticks
After one reported observation was stored, the same goal closed without another sensor call. Inquiry and K0 circuits were teacher-observed.
Conditions and limitations →03 PROOF
What did we
actually observe?
Each result below names its condition and its limit. A bounded experiment is evidence for that fixture, not a claim of general intelligence or superiority.
TECHNICAL HYPOTHESIS
Keep memory, state, and judgment outside a single model call.
The language model may interpret or express a surface. MIRA tests whether observation, experience, relation, closure, action, and outcome can retain separate ownership.
- 01SurfaceLanguage · document · event
- 02ObservationSource · time · coordinates
- 03ExperiencePreserved and reusable memory
- 04RelationConnections local to a goal
- 05ClosureOPEN · MULTIPLE · SINGLE
- 06OutcomeActual result returns as observation
This architecture is a research hypothesis. Its value must be tested separately for each capability and application.
SELECTED MEASUREMENTS
Narrow results with their boundaries attached.
These values come from the published experiment records. Different fixtures are not combined into one performance score.
Behavior moved with memory
A ≠ Bsame executor, different snapshots
OPENafter transition memory removal
The output changed with the memory snapshot and disappeared with the relevant memory. The transition rows were supplied through observed teacher events.
Conditions and limitations →Temporal order remained distinct
ABABA→B 2 · B→A 1BABAA→B 1 · B→A 2Equal surface mass retained different directional temporal structure after the existing fold path was connected.
Conditions and limitations →Action reconciliation after restart
0duplicate external actions
0invented outcomes
An exact acknowledgement closed one committed action. Missing acknowledgement remained unresolved and did not authorize retry.
Conditions and limitations →WHAT THIS DOES NOT PROVE
The missing comparisons matter.
- No general benchmark currently establishes superiority over existing agent architectures.
- The published results use different fixtures and cannot be collapsed into one success rate.
- Teacher-observed circuits remain a major boundary for claims of self-formation.
- Cross-domain reuse has not been established by the published notes.
- Prototype behavior is not equivalent to product reliability.
SOURCE RECORDS
Read the method before accepting the result.
The detailed notes expose the research question, setup, observations, interpretation, residuals, and next question.
Open research notes