L0
PRECONDITION
Ontological data foundation
What is governed and available to the machine?
Function
Output
Modeler / governance
01Working paper · v1.1
Beyond Ontology
Ontology structures what is available to the machine. Computational epistemology structures how the machine is permitted to infer from it.
A governed source of truth is the beginning of inference—not its justification. This framework specifies how candidate evidence is challenged, qualified, reconciled and made decision-grade.
Read the working paperL0governed state
L8auditable artifact
Move the pointer: the machine responds. It does not declare truth.
02The inferential gap
Governed does not mean justified.
T
Structured domain state
Objects, relations, lineage, timing and permissions
E
Justified inference
Persistent, qualified, calibrated and decision-relevant evidence
Ontology answers what exists and how it is connected. It cannot certify whether a pattern survives another sample, another window, or a changed regime.
More governed data can still generate more spurious patterns. A fitted model cannot certify its own epistemic status. The missing function is a disciplined protocol for attempted refutation.
Paper §§2–3, pp. 3–6 · Equation (2)
Pointer field · illustrative
Move across the field—or use the button—to reveal a stable structure inside the noise.
03L0–L9 architecture
One precondition. Nine inferential layers.
L0 makes governed evidence possible. L1–L9 determine what the machine may conclude, how uncertainty survives, and who remains accountable.
L9Realized outcomes feed Layers 2–6; structural revisions still require modeler sign-off.
04Persistence lab
A signal must survive being challenged.
Within one model family, the question is not “did it detect something?” but “does the decision-relevant property persist when reasonable analytical choices change?”
Illustrative simulation · not a paper result
Change the illustrative survival rate to see why qualification is a revisable state, not a permanent label.
Windows
—
Frequencies
—
Resamples
—
Specifications
—
L4 qualification
Conditional evidence
Epistemic noise
Unresolved
Conditional
Persistent
Paper §§5.4–5.5 and 6.1, pp. 11–14 · Equations (7)–(9)
05Cross-family convergence
Agreement is not truth.
Three engines can repeat one shared flaw. Evidential weight depends on heterogeneous, sufficiently independent corroboration—not a raw vote.
ΣStatisticalestimation / time series
⊢Symbolicrules / structural constraints
ƒLearnedrepresentations / ensembles
Illustrative simulation · not a paper result
Raw agreement3 / 3
Effective independent evidence1.50
Neff = 3 / (1 + 2ρ) for this equal-dependence simulation
“Convergence across sufficiently independent model families is evidence of robustness, not proof that a signal is real.”
Paper §6.2, pp. 13–14 · Equations (15)–(17)
06Probability → impact → artifact
A probability is not yet a decision.
Probability, fuzzy membership, stress response and calibration answer different questions. The artifact keeps those distinctions—and the evidence trail—visible.
Illustrative simulation · not a paper result
Move the input to inspect four distinct Layer 6 outputs. Values are deliberately illustrative.
Calibrated event probability58%frequency-honest target
Fuzzy membership0.64degree of resemblance, not probability
Adverse stress case45%scenario sensitivity
Calibration statusReviewdiagnostic, not confidence
Decision-grade inference artifact · illustrative
MODEL REVIEW
Event hypothesis / Horizon H
Calibrated probability58%Interval 44–72%
Expected impactContext requiredMapped by a domain-appropriate impact model
- Supporting evidence
- 2 qualified families
- Conflict retained
- 1 unresolved family
- Robustness
- Conditional
- Stress
- 45% adverse case
- Validity horizon
- H · stale after expiry
- Lineage
- T → q₁…qⱼ → Zₜ → p*
Paper §§5.7–5.9 and 7.1–7.3, pp. 12–16 · Equations (11)–(13), (18)–(21)
07Conditional complexity gate
The smallest defensible architecture wins.
Sophistication has no epistemic value by itself. Start with the simplest adequate baseline; add only the layers that resolve a material problem out of sample.
01
Does a simple admissible baseline adequately solve the decision problem?
02
Does the added machine produce material out-of-sample gain after computation, latency and governance costs?
Architecture decision
Use the baseline.
If a regression adequately solves it, regression is the correct epistemic architecture.
\[\displaystyle \Delta V = \Delta P + w_C\,\Delta C + w_R\,\Delta R + w_I\,\Delta I - w_K\,\Delta K - w_L\,\Delta L\]
Predictive, calibration, robustness and impact gains must exceed complexity and latency costs—and be statistically and economically material.
Paper §9, pp. 19–20 · Equations (29)–(30)
08Limits & research agenda
A framework to test—not a result to believe.
- Inference is more defensible when the decision is framed before model selection.
- Candidate signals should survive admissible perturbations before promotion.
- Cross-family synthesis must discount shared dependence and retain conflict.
- Probability becomes decision-relevant only when tied to impact and an audit trail.
Open empirical program
- 01
Test whether cross-window persistence improves out-of-distribution performance.
- 02
Estimate which perturbation sets best separate structure from sample artifacts.
- 03
Measure how shared data and inductive bias should discount convergence.
- 04
Compare a single model, a homogeneous ensemble, an ontology-centered system and the full machine on held-out data.
The machine may propose a structural revision. It cannot promote that revision by itself. Modeler sign-off remains necessary.
Paper §§7.5, 10–11, pp. 17, 20–2209The governing principle
Not the most elaborate machine. The smallest one that can defend its inference.
Computational epistemology does not replace ontology or the modeler. It defines an auditable path from governed state to qualified evidence, calibrated probability, material impact and revision under responsibility.