Correction & Retraction Notice (July 13, 2026)
Following an adversarial audit, this document has been corrected. We formally retract the claim of "bidirectional closure" (the Π_A+ operator is a lossy, forward readout, not bidirectional). We retract the use of the term "resonance" for the prototype, correcting it to "parameterized local-growth boundary condition." We correct the threshold inversion bug (the formula for θ_b has been inverted to match the prose). Finally, we clarify that the Specificity Receipt measures target compliance, not absolute truth, and that Seed 42 is a single deterministic trace, not a proof of robust morphology classes. The core M1 vertical slice remains verified, but the theoretical claims have been bounded.
Research note | Corrected July 13, 2026
The Resonant Morphology Thesis
Resonant Morphology connects a simple proposition across the Fractalish stack: a process can leave a structured form, that form can be read under constraint, and the adequacy of that reading can be measured before it is trusted.
This places the thesis in both Cognitive Basin and Fractalish AI. It is a machine-cognition architecture note, a public implementation target, and a claim-boundary document.
Fractalish AI
An active AI architecture program
Fractalish AI is presented here as an active AI architecture and implementation program: local-first, governed, replayable machine cognition designed to connect morphology, memory, measurement, and device evidence.
This is not a claim of artificial personhood or proven sentience. Its present scope is architectural and experimental: testable slices, bounded claims, and reproducible evidence.
The closure loop
The thesis organizes a repeatable loop:
- A parameterized local-growth boundary condition constrains local process.
- Local growth proceeds through finite decisions, thresholds, trails, and restrictions.
- The resulting morphology becomes a structured record of what happened.
- A symbolic readout attempts to encode the form without asserting complete recovery.
- A specificity receipt measures what was preserved, what failed, and what governance state follows.
- The Cognitive Basin admits the result as evidence, feedback, warning, or HOLD rather than as an unexamined conclusion.
Corrected threshold and forward readout
The prototype's branching threshold is corrected to:
θ_b = 2.8(T) + 1.5
The original draft used θ_b = 2.8(1 - T) + 1.5, which mathematically lowered the threshold for high coherence and contradicted the prose. The corrected expression raises the threshold for high coherence, suppressing branching and favoring stability.
The readout is lossy and forward-only. The symbol is deterministically recoverable from the discrete ternary vector, but tuple/glyph round-trip does not reconstruct the original simulated morphology. This is a bounded forward fixture, not bidirectional recovery.
The prototype slice
A local Natural Math / Construction A+ prototype, bifurcation_motif_v1.py, reproduces the following deterministic receipt for seed 42:
| Observation | Reported result |
|---|---|
| Segments | 7 |
| Bifurcations | 3 |
| Glyph ID | 31433 |
| PEFP digits | [0, 1, 0, -1, 0, -1, -1, 0, 1] |
| GSR / NGR | 0.8 / 0.2 |
| Debt classification | CAUTION |
| Failed checklist item | trail_density_ok |
The result is useful because the receipt preserves a visible failure and a nonzero debt state. The motif and symbolic readout are therefore carried forward as bounded evidence rather than unrestricted confirmation.
The Specificity Receipt in this prototype measures target compliance against the declared checklist. It does not measure absolute truth, and Seed 42 remains a single deterministic trace rather than proof of robust morphology classes. A CAUTION output in one fixture is not a general solution to hallucination.
Why this belongs in Cognitive Basin
Cognitive Basin is where the result becomes governable. The prototype does not merely generate an output; it produces an evidence posture. The Basin can treat the receipt as structured feedback, preserve the failed condition, route the state into a bounded next action, and prevent an attractive pattern from becoming an unsupported claim.
This is why Resonant Morphology belongs in the Basin lane: it is about admission, memory, state, feedback, and governed recurrence.
Why this belongs in Fractalish AI
Fractalish AI is the broader implementation program that asks whether morphology, memory, measurement, and device evidence can be joined into local machine cognition. Resonant Morphology gives that program a compact test shape: constrain local growth, produce a morphology, read it, score the reading, and carry the result forward under governance.
That is AI work in the direct engineering sense: stateful systems, measured representations, bounded action, repair, replay, and evidence-aware decision loops. It is not a claim that the system has subjective experience.
Claim boundary
This note does not claim sentience, artificial personhood, physical CNTM memory, physical morphology memory, completed persistent intelligence, a universal morphology decoder, medical authority, or regulatory authority.
It does claim that Fractalish AI can be described publicly as an active AI architecture and implementation program, and that the Resonant Morphology loop is specific enough to be built, run, criticized, replayed, and improved.
The Pivot: Natural Math Leaf Morphogenesis
The Resonant Morphology Prototype successfully demonstrated a deterministic generation-and-encoding pipeline. However, to test whether compact, species-specific local growth constraints can generate characteristic biological forms and preserve recoverable information about the rules that produced them, the project is immediately pivoting to the Natural Math Leaf Generator.
The next phase will not claim bidirectional closure. It will execute a rigorous three-stage test: (1) Generate a generic leaf-like structure from local vein branching and blade filling without a pre-drawn outline. (2) Produce distinct rule families (e.g., palmate vs. pinnate) by altering the inherited Natural Math genome. (3) Execute the Inverse Test, feeding only the final graph topology into a classifier to see if it can identify the rule family that produced it. This is one bounded test of whether final topology preserves useful information about the rules that produced it.