How it works

A validator-backed reasoning stack with explicit boundaries

This page explains the stack in public-facing terms: where noisy inputs live, where the noisy-to-exact transition is made explicit, and where canonical reasoning artifacts come under validator discipline.

The point is not more trace volume. The point is an exact, reviewable transition from approximate evidence into canonical artifacts with stable outputs and compatibility identity.

Planes Operational strata for noisy inputs, explicit normalization, and exact artifacts.
Roles Tsubasa, Liu, and Tobi are stack roles, not alternate names for the planes.
Verification Checks canonical, reproducible, validator-backed artifacts under exact discipline.
planes vs roles canonical pipeline verification distinction

Three layers

Data Plane, Bridge, Control Plane

Planes describe operational strata and the discipline applied there. They are not aliases for named components or product modules.

Data Plane

Noisy, approximate, measured inputs

Measurements, traces, simulations, statistics, and probabilistic outputs live here. This plane is non-canonical by default.

Bridge

Explicit noisy-to-exact boundary

The bridge makes normalization explicit, prevents hidden float-to-decision jumps, and preserves provenance as approximate input crosses into exact reasoning.

Control Plane

Canonical reasoning artifacts

Canonical reasoning artifacts live here under deterministic evaluation structure, decision discipline, and validator-checkable effects.

Planes describe operational strata. Tsubasa, Liu, and Tobi describe stack roles that author or validate artifacts across the disciplined path.

Stack roles

Components are roles in the stack, not names for the planes

The components below are responsibilities in the stack. They are not synonyms for Data Plane, Bridge, or Control Plane.

Tsubasa

Semantic core for canonical artifacts

Canonical reasoning language and semantic layer for exact, control-plane meaning.

Liu

Sugar and authoring layer only

Improves authoring ergonomics without relaxing canonical boundaries or redefining semantics.

Tobi

Reference validator and trust anchor

Validates canonical meaning in implementation with deterministic diagnostics and _h compatibility identity.

Canonical pipeline

From input artifact to compatibility identity

The canonical path is staged and deterministic. Each step makes the transition into validator-checkable form more explicit.

Parse Measured or authored input enters the disciplined path.
NFC normalize Normalize at the noisy-to-exact boundary.
Preserve provenance Keep the transition from approximate input explicit and reviewable.
Alias and desugar Alias to canonical form and desugar into the semantic core.
Canonical Core AST Reduce the artifact into exact semantic structure.
Canonical ASCII Emit stable text for review, diff, and comparison.
TSER serialization Produce deterministic serialized form for downstream handling.
_h compatibility identity Surface compatibility-oriented identity for workflow gates.

Bridge discipline

Why the noisy-to-exact boundary has to stay explicit

If the transition from noisy measurement or probabilistic output into exact decision logic is hidden, trust collapses. The bridge keeps that transition explicit and reviewable instead of burying it inside implementation detail.

That is what makes the downstream artifact easier to compare, reproduce, validate, and gate across CI and reproducibility-sensitive workflows.

Positioning distinction

Verification is not the same thing as observability or evaluation

These are adjacent concepts, but they answer different questions and should not be collapsed into one another.

Observability

Shows what happened in a system, run, or workflow.

Evaluation

Suggests whether behavior seems useful, acceptable, or performant.

Verification

Checks whether the artifact is canonical, reproducible, and validator-backed under discipline.

Workflow fit

Move from architecture into GitHub workflow fit

Keep the next step narrow: use the docs path, review the GitHub-first workflow starter, and contact Organetic only when you need help placing the released validator line inside a real gate.