Source and figure crosswalk
Canonical conceptual source
Self-Improving Intelligence: Representation, Discovery, and Recursive Revision. Komorebi P., first consolidated edition v1.0.0, 8 October 2026.
- Section 7, printed pages 8–10: a fixed unknown experiment model; ambiguity of
observation = hidden bit XOR sensor convention; calibration as a separate information channel. Read directly from the manuscript for this companion. - Section 9: transporting observations and an investigation, not only a final repair name.
- Section 24: an old certificate can lose applicability when its premises change.
- Section 31, printed pages 48–49: functional agent roles, recurrent state, persistent operations, acquisition, proposals, admission, and the distinction between state updates and learned changes. Read directly from the manuscript for this companion.
This companion's new analytical construction
The three-bit source, separate A/B observation access, exact eighteen-event schedule, communication receipts, viewer-only environment display, entropy histories, 3D probability projection, and controlled access comparisons are newly authored illustrative material in code/model.py and code/visuals.py.
The source paper does not claim that this exact two-agent schedule was executed previously. The illustration does not inherit a theorem of safe operation under arbitrary environmental change. In particular, the sensor changes between investigations and the reset rule is supplied. Before an agent receives the contradiction or reset notice, its old certainty can be stale.
Plate map
- Functional boundary and perception–action loop. Conceptual diagram.
- Private information and explicit communication. State from event 11.
- Synchronized system display. State from event 12.
- Exact Agent A location-probability history, events 0–17.
- Entropies of the A/B location distributions, events 0–17.
- Probability of report one before observing each report, versus the actual binary report.
- Three expected source-bit coordinates, derived from the full belief vector. Lossy projection explicitly identified.
- Joint source/orientation distribution at event 12.
- Number of joint hypotheses during post-change investigation and calibration.
- Event raster of the exact schedule. No physical-time or hardware-latency claim.
- Matched post-change evidence-access conditions, not a neural training study.
- Common-action requirement and actual stopping decision.
- Executed belief-update loop versus proposed learned-method revision.
- Model assumptions and interpretation guide.
All figure labels and quantitative displays come from the stated model or its declared equations. No external empirical statistics or contemporary AGI performance claims are introduced.