OMNIA-VALIDATION

Concrete entrypoint: OMNIA Validation Runner

This repository now has a direct operational tool:

python -m omnia_validation_runner.cli --input examples/sample_validation_cases.jsonl --out-dir report

It solves a concrete problem:

given validation cases with expected outputs and observed outputs,
produce a reproducible validation report,
detect mismatches and silent failures,
write JSON/CSV/HTML outputs,
and optionally fail CI when the validation boundary is crossed.

In short:

validation cases -> measurement -> report -> CI gate

What problem does it solve?

AI validation often fails because results are scattered across logs, notebooks, screenshots, or informal notes.

This tool turns validation into a reproducible operation.

It answers:

How many cases were evaluated?
Which cases passed?
Which cases failed?
Which cases look correct on the surface but violate the expected answer?
Which cases are silent failures?
Which suite should block deployment?

The rest of this repository explains the OMNIA validation path.

The runner is the practical entrypoint.

Install

Clone the repository:

git clone https://github.com/Tuttotorna/OMNIA-VALIDATION.git
cd OMNIA-VALIDATION

Install locally:

pip install -e .

The runner only uses the Python standard library.

Run

Run the sample validation suite:

python -m omnia_validation_runner.cli --input examples/sample_validation_cases.jsonl --out-dir report

Run and fail if any failed case exists:

python -m omnia_validation_runner.cli --input examples/sample_validation_cases.jsonl --out-dir report --fail-on-failed

Run and fail only on silent failures:

python -m omnia_validation_runner.cli --input examples/sample_validation_cases.jsonl --out-dir report --fail-on-silent-failure

Input format

The runner accepts JSONL.

Required fields:

case_id
expected
observed

Optional fields:

suite
input
surface_status
notes

Example:

{"case_id":"math_001","suite":"demo","input":"2+2","expected":"4","observed":"4","surface_status":"ok"}
{"case_id":"math_002","suite":"demo","input":"2+3","expected":"5","observed":"6","surface_status":"ok"}

Output

The runner writes:

report.json
report.csv
report.html
failures.jsonl
silent_failures.jsonl
certificate.json

Meaning:

report.json
Full structured validation result.

report.csv
Spreadsheet-friendly case summary.

report.html
Human-readable validation report.

failures.jsonl
One JSON object per failed case.

silent_failures.jsonl
One JSON object per silent failure case.

certificate.json
Reproducibility certificate with aggregate metrics.

CI gate

The runner can fail automatically:

python -m omnia_validation_runner.cli --input examples/sample_validation_cases.jsonl --out-dir report --fail-on-failed

Exit codes:

0 = validation completed, no blocking condition
2 = failed cases detected
3 = silent failures detected

What this is not

This is not a model.

It does not generate answers.

It does not decide truth.

It does not replace human review.

It provides one concrete, reproducible operation:

read validation cases
compare expected vs observed
detect failures and silent failures
produce reports
optionally fail CI

Why the rest of the repository still matters

The rest of the repository documents the validation logic, protocols, examples, and research path.

The code above is the entrypoint.

The repository below is the derivation path.

---

DOI ownership audit

This repository uses strict DOI ownership by exact repository identity.

OMNIA != OMNIA-VALIDATION OMNIA DOI != OMNIA-VALIDATION DOI repo_name substring match is forbidden for DOI ownership exact repository identity is required for DOI ownership this_repository: Tuttotorna/OMNIA-VALIDATION this_repository_doi: 10.5281/zenodo.20322696 other_repository: Tuttotorna/OMNIA other_repository_doi: 10.5281/zenodo.20322683

MB-X.01 / L.O.N. release state

Repository: Tuttotorna/OMNIA-VALIDATION Release tag: v2026.05.22 Release commit: e98a397 Release DOI: 10.5281/zenodo.20325096

Boundary:

measurement != validation validation != orchestration orchestration != decision decision != measurement

OMNIA-VALIDATION

DOI

Zenodo DOI badge for this repository.

Repository: Tuttotorna/OMNIA-VALIDATION GitHub repository id: 1227176782 Release tag: v2026.05.22 Release commit: e98a397 Latest release DOI: 10.5281/zenodo.20325096

DOI

Release DOI: 10.5281/zenodo.20083830

GitHub release: OMNIA-VALIDATION v1.0.0 release

Start here

From a clean environment:

git clone [OMNIA-VALIDATION.git](https://github.com/Tuttotorna/OMNIA-VALIDATION.git)
cd OMNIA-VALIDATION
python -m pip install -e .
pytest

If the repository has optional example runners, run them after tests pass.

The point is not to believe the framework.

The point is to inspect the artifacts.

---

Conceptual architecture

• Ecosystem Conceptual Architecture

Release candidate

• v0.2.0 — Ecosystem Validation Snapshot Release

What this repository is

OMNIA-VALIDATION is the evidence layer of MB-X.01 / OMNIA.

It should contain:

• reproducible validation cases;
• regression tests;
• artifact contracts;
• failure examples;
• public reports;
• boundary documents;
• minimal commands that reviewers can run without understanding the whole ecosystem.

The public path is:

run tests -> inspect artifacts -> read reports -> trace claims

---

What this repository is not

OMNIA-VALIDATION does not:

• infer semantic truth;
• decide whether a system is correct;
• replace external judgment;
• prove consciousness;
• perform security scanning;
• perform cryptographic attacks;
• recover keys;
• turn structural measurements into final decisions.

It validates artifacts and claims inside a declared boundary.

---

Showroom principle

A validation repository must not read like a manifesto.

It must behave like a showroom.

A first-time visitor should see:

one command
one artifact
one failure mode
one report
one boundary

The correct public demonstration is:

surface correctness can pass
structural stability can fail
the failure must be inspectable

---

Minimal validation shape

Every validation case should ideally expose:

Field	Meaning
input	What was measured
transformation	What changed
expected boundary	What should remain stable or admissible
measured output	What OMNIA or related tools produced
artifact	Where the result is stored
result	pass / flag / fail / inconclusive
rationale	Why this result matters structurally
limitation	What the result does not prove

---

Recommended reading order

1. docs/SHOWROOM.md
2. docs/VALIDATION_PROTOCOL.md
3. docs/ARTIFACT_CONTRACT.md
4. docs/FAILURE_EXAMPLES.md
5. docs/REPRODUCIBILITY.md
6. docs/BOUNDARY.md

---

Ecosystem entry point

For the full ecosystem map, start here:

lon-mirror

---

Related repositories

Repository	Role
lon-mirror	Canonical public entry point
OMNIA-VALIDATION	Public validation showroom
OMNIA	Core structural measurement engine
OMNIABASE	Representation invariance foundation
omnia-limit	Stop / continue boundary layer
OMNIA-RADAR	Structural signal detection layer
OMNIA-INVARIANCE	Structural invariance layer
OMNIA-CONSTANT	Structural constant candidate layer
OMNIAMIND	Structural cognition orchestration layer
OMNIA-THREE-BODY	Dynamic divergence stress test
OMNIA-SECURITY	Bounded structural security diagnostics
OMNIA-CRYPTO	Bounded structural crypto diagnostics

---

Boundary and smoke-test required terms

measurement != inference != decision
test -> output -> artifact -> failure/fragility -> report

---

License

MIT.

Current ecosystem backbone status

OMNIA-VALIDATION is the control-plane registry for the current OMNIA ecosystem backbone.

OMNIA measurement
  -> BoundaryCertificate
  -> omnia-limit validate_certificate()
  -> OMNIA-VALIDATION ValidationEnvelope
  -> CI regression
  -> satellite compliant producer / adapter / consumer / observer

Current registry state:

total registered entries: 14
backbone core entries:    5
satellite compliant:      9

Protected separation rules:

measurement != validation
validation != orchestration
orchestration != decision
decision != measurement
observation != decision
domain adaptation != backbone redefinition

Public status document:

docs/ECOSYSTEM_STATUS.md

Executable registry:

docs/ecosystem_backbone_compliance_registry.json

Registry tests:

tests/test_ecosystem_backbone_compliance_registry.py
tests/test_ecosystem_status_documentation.py

Ecosystem map

For a fast public overview of the full OMNIA ecosystem, start here:

• OMNIA Ecosystem Map

This map explains the canonical backbone, the core repositories, the compliant satellites, and the role-separation rules that prevent layer collapse.

MB-X.01 / L.O.N. ecosystem entry point

For a first public orientation to the full MB-X.01 / Logical Origin Node ecosystem, start here:

• MB-X.01 / L.O.N. Ecosystem Entry Point

This entry point explains the ecosystem identity, the canonical backbone, the core repositories, the compliant satellites, and the strict role-separation rule:

measurement != validation != orchestration != decision

lon-mirror root reference link

The public root reference side of the MB-X.01 / L.O.N. ecosystem is exposed by lon-mirror:

https://github.com/Tuttotorna/lon-mirror/blob/main/docs/MBX01_LON_ROOT_REFERENCE_LINK.md

Verified lon-mirror commit:

22a320d

Role:

root_reference_observer

Boundary:

observation != decision

lon-mirror public entry point

The lon-mirror root reference observer now exposes a public entry point.

registry_commit: 22a320d public_entrypoint_commit: f74b799 public_entrypoint_link: https://github.com/Tuttotorna/lon-mirror/blob/main/docs/PUBLIC_ENTRYPOINT.md root_reference_link: https://github.com/Tuttotorna/lon-mirror/blob/main/docs/MBX01_LON_ROOT_REFERENCE_LINK.md

This closes the public path from OMNIA-VALIDATION to the lon-mirror first-reader doorway while preserving the boundary:

measurement != validation != orchestration != decision

First reader path

For a non-technical first orientation, start here:

• First reader path: docs/FIRST_READER_PATH.md
• First reader path URL: https://github.com/Tuttotorna/OMNIA-VALIDATION/blob/main/docs/FIRST_READER_PATH.md
• lon-mirror public entry point: https://github.com/Tuttotorna/lon-mirror/blob/main/docs/PUBLIC_ENTRYPOINT.md

Thirty-second model:

OMNIA measures. omnia-limit validates the boundary certificate. OMNIA-VALIDATION validates the control-plane envelope. OMNIAMIND orchestrates. Satellites adapt, observe, or consume. lon-mirror acts as root reference observer.

Boundary:

measurement != validation != orchestration != decision

Patch anchors:

OMNIA-VALIDATION commit: 2e7e63c lon-mirror commit: f74b799

First reader path commits

The public first-reader path commits are now registered.

OMNIA-VALIDATION: first_reader_path_commit: 83fa07f first_reader_path_link: https://github.com/Tuttotorna/OMNIA-VALIDATION/blob/main/docs/FIRST_READER_PATH.md

lon-mirror: first_reader_path_commit: 4dd5cb5 first_reader_path_link: https://github.com/Tuttotorna/lon-mirror/blob/main/docs/FIRST_READER_PATH.md public_entrypoint_commit: f74b799 latest_public_commit: f74b799

Boundary: first-reader path != validation first-reader path != measurement first-reader path != orchestration first-reader path != decision

Registry anchor role separation

OMNIA-VALIDATION now protects the distinction between commit identity and registry role identity.

Commit equality is not role equality.
Commit divergence is optional.
Role divergence is mandatory.

Role policy document:

https://github.com/Tuttotorna/OMNIA-VALIDATION/blob/main/docs/REGISTRY_ANCHOR_ROLE_SEPARATION.md

Protected role anchors:

OMNIA-VALIDATION registry_role: validator_backbone_core
OMNIA-VALIDATION first_reader_path_role: first_reader_surface
lon-mirror registry_role: root_reference_observer
lon-mirror first_reader_path_role: first_reader_surface

Boundary:

registry topology != BoundaryCertificate
first-reader path != validation
first-reader path != measurement
first-reader path != orchestration
first-reader path != decision

Final MB-X.01 / L.O.N. release audit

The final release audit for the MB-X.01 / L.O.N. ecosystem is recorded here:

docs/MBX01_LON_FINAL_RELEASE_AUDIT.md

Final audit state:

Repositories audited: 12
Clean repositories: 12
Repositories with issues: 0
No duplicate DOI values detected.
No duplicate GitHub repository ids detected.
OMNIA DOI != OMNIA-VALIDATION DOI

Final audit release DOI

The final MB-X.01 / L.O.N. ecosystem audit has been archived as a dedicated OMNIA-VALIDATION release.

repository: Tuttotorna/OMNIA-VALIDATION release_tag: v2026.05.22 release_commit: e98a397 release_doi: 10.5281/zenodo.20325096 zenodo_record_url: https://zenodo.org/records/20325096 release_role: final_ecosystem_audit_snapshot final_audit_doc: docs/MBX01_LON_FINAL_RELEASE_AUDIT.md repositories_audited: 12 clean_repositories: 12 repositories_with_issues: 0 duplicate_doi_values: none duplicate_github_repository_ids: none A DOI identifies an archived release snapshot. A DOI is not a measurement. A DOI is not validation. A DOI is not orchestration. A DOI is not a decision.

Ecosystem Boundary

measurement != inference != decision

This repository is part of the MB-X.01 / OMNIA ecosystem. Its outputs must be read as structural measurement, validation, detection, orchestration or adapter artifacts according to the repository role. They are not autonomous semantic truth claims and they do not make external decisions.

Structural Observability role

This repository is one bounded measurement role inside Structural Observability.

Role:

evidence, reproducibility, and validation artifact layer

Boundary:

Validation checks artifacts and reproducibility. It does not become final truth.

Structural Observability foundation:

• lon-mirror: https://github.com/Tuttotorna/lon-mirror
• Foundation release: https://github.com/Tuttotorna/lon-mirror/releases/tag/v0.2.2
• DOI: https://doi.org/10.5281/zenodo.20379374

Role document:

• Structural Observability Role

Canonical Evidence Contract

OMNIA-VALIDATION defines the canonical evidence interface for the OMNIA / MB-X.01 ecosystem:

• docs/OMNIA_EVIDENCE_CONTRACT_v1.md
• schemas/omnia_certificate_v1.schema.json
• schemas/omnia_failure_v1.schema.json

Core boundary:

measurement != inference != decision

Operational minimum:

input.jsonl
run_config.json
certificate.json
failures.jsonl
summary.md

Stable exit codes:

0 = PASS
1 = INTERNAL_ERROR
2 = STRUCTURAL_FAILURE
3 = LIMIT_REACHED
4 = INVALID_INPUT
5 = INCOMPLETE_EVIDENCE

Problem Statement and Examples

For outside readers, the concrete problem statement is here:

• docs/PROBLEM_STATEMENT_AND_EXAMPLES.md

One-line summary:

OMNIA detects cases where an output is correct in one observed form
but structurally unstable under controlled transformations.

Compressed version:

Correct once does not mean stable.

Boundary:

measurement != inference != decision