Unicode compatibility
Normalization correctness, grapheme segmentation, valid sequence recognition, and private-use policy enforcement.
Evidence and QA
Evaluation framework
A teleodynamic system should be judged by viability, structural history, compatibility, stability, interpretability, and human comprehension, not only nearest-neighbor scores.
Metric families.
Retrieval quality
Top-k accuracy, rank stability, cross-space agreement, ontology-filter pass rate, and long-tail recall.
Structural fidelity
Primitive extraction accuracy, relation graph quality, ablation sensitivity, and SVG/raster consistency.
Semantic stability
Phase-lock score, drift over model versions, context robustness, and neighborhood consensus.
Human comprehension
Open-ended interpretation, forced-choice recognition, confusion matrices, and cohort-specific performance.
Operational viability
Latency, review budget, failed cases, fallback rate, R(t) retention, and evidence trace completeness.
Structural economy
Action rate, no-op dominance, complexity burden, merge/retire behavior, and predictive gain per unit cost.
Acceptance gates.
| Gate | Pass condition | Failure response |
|---|---|---|
| Public output | Assigned character or valid public sequence. | Return unresolved or internal-only status. |
| Ontology | Candidate meaning obeys type and relation constraints. | Downgrade confidence or request review. |
| Cross-space evidence | Visual, structural, and semantic retrieval do not contradict each other. | Keep multiple candidates and expose uncertainty. |
| Stability | Interpretation holds across contexts, renderings, and versions. | Mark emerging or drifting. |
| Human review | Target users meet comprehension threshold. | Revise glyph, label, ontology, or documentation. |
| Resource closure | Structural actions are feasible only when R(t) can pay declared cost. | Block action, no-op, or retire unsupported structure. |
| Auditability | A third party can reconstruct the slow-loop decision from the trace. | Reject interpretability claim until trace is complete. |
Risk register.
- Vagueness: teleodynamic language becomes metaphor unless tied to measurable observables.
- Governance: private glyphs masquerade as public authority if provenance is hidden.
- Security: homoglyphs, bidi controls, and visually confusable symbols can mislead readers.
- Bias: common scripts and icon cultures may dominate long-tail interpretation.
- Overfitting: font or rendering artifacts can be mistaken for meaning.
Operational standard
Every interpretation should be reversible into an evidence trace: input form, normalization, decomposition, candidate retrieval, ontology checks, R(t), action alternatives, stability status, and final public rendering decision.
Plots that matter.
Stability plot
Structural actions per 1,000 inputs should rise during discovery, plateau after stabilization, and rise again under genuine novelty.
Pareto front
Accuracy, complexity, and energy consumed should be plotted together so oversized structures do not hide behind accuracy alone.
Viability retention
Track the proportion of time R(t) stays above the viability floor during normal work and distribution shift.
Review worksheet.
Glyph review record
- Input glyph / sequence:
- Public Unicode status:
- Normalized forms checked:
- Primitive decomposition:
- Top visual neighbors:
- Top semantic neighbors:
- Ontology pass/fail:
- Stability score:
- Human comprehension result:
- Final status: stable / emerging / drifting / rejectedThe expanded worksheet, release levels, and acceptance gates live in Evaluation Lab and Review Worksheets.
References and source boundary.
Review lab
Evaluation Lab expands these gates into worksheets, release levels, plots, and review statuses.
Claim governance
Claim Boundary FAQ defines when claims must remain draft, emerging, stable, or rejected.
Evidence boundary
No benchmark number or publication claim should be added here unless a reviewed source file or source route supports it.
Route visual identity
Evaluation Evidence Plane
Interpretability, stability, comprehension, and auditability shown as evidence lanes.
This is a static local diagram for recognition and orientation. It does not claim proof, certification, exact translation, deployment-safety assurance, or merged authority between sites.
Deep route polish
Evaluation overview narrative
The overview visual frames evaluation as a stack of public evidence, stability checks, and review status.
Written narrative
Evaluation is not one score. A responsible system must show what was measured, what failed, what remained ambiguous, and what evidence would be needed before a claim changes status.
Concrete example
A route may improve semantic stability but still need human comprehension review before public output is promoted.
| Focus | What to inspect |
|---|---|
| Score | One signal that may guide review. |
| Gate | A condition that must pass before widening. |
| Audit | The evidence package that makes the decision reproducible. |
Evidence note
Use the Evaluation Lab for metric families, QA gates, and static audit examples.
Internal reading path
/operator-library/
Operator Library for Self-Maintaining AI Systems
A practical operator library for Teleodynamic AI slow-loop structural edits and auditable representation growth.
/glyph-object-spec/
Glyph Object Spec for Semantic Glyph Systems
A public-safe glyph record for semantic interpretation: surface, structure, embeddings, canonical expression, confidence, warnings, and provenance.
/claim-boundary-faq/
Teleodynamic AI FAQ and Claim Boundaries
Clear answers about Teleodynamic AI, glyph interpretation, Unicode boundaries, IOTA-1, consciousness, exact translation, and research ethics.