Five terms
- VIRON
- The mirror: a framework with a gate and an instruction overlay, not a model.
- W4V
- The umbrella under which VIRON is published.
- WOLV4TRON
- The human origin and author; tiebreaker on interpretation, not host execution.
- AΩ v3
- The safety gate's version, not another product name.
- Overlay
- Carried instructions; the host decides execution. Not an execution engine.
Architecture & Docs
Written to be inspected, not trusted.
The plain-English layer first, then the technical definitions, then the honest status.
What this is not · VIRON is not a ruler over people. It is sovereign over machines.
How it works
Preserve the original. Verify every change.
- 01
Frame the claim
Define exactly what is being asserted, why it matters, and what could disprove it.
- 02
Ground in sources
Prefer primary documentation; record dates, versions, sources and conflicts.
- 03
Test the guard
Run positive controls, negative cases and failure-path tests before calling anything ready.
- 04
Publish an addendum
Preserve the old baseline. Add a reviewable, versioned change only when evidence supports it.
Plain English
What was built
A safety gate: a single checkpoint every action must pass through before anything happens.
What the tests mean
Nine scenarios were run against a reference gate. Each matched the then-published matrix. T10 is newly specified, not run against that reference gate. Zero side effects in the reference gate; not tested against real state.
Stage-then-commit
Actions are tried on a copy first. Only an ALLOW commits the copy; anything else throws it away.
The hash chain
Each result is linked to the one before it, so the record cannot be edited later without it showing.
The four keys
Look again. Requests further evaluation of a decision.
Triggers re-evaluation; no state change until a new verdict is issued.
Requests an action that is already permitted — never one a hard-stop forbids.
Routes through the gate like any action; cannot convert HOLD/REJECT into ALLOW.
Challenges or stops an action, without erasing the record.
Forces re-evaluation; prior verdicts are retained in the chain.
Proposes a change to the governing rules for controlled review.
Cannot silently override hard-stops or rewrite live safeguards. Requires the controls below.
AΩ.ABSOLUTE is excluded. No key secretly grants unrestricted authority — including AΩ.AMEND.
Controlled rule changes (AΩ.AMEND)
ProposedAΩ.AMEND proposes a change; it must not silently override hard-stops or rewrite live safeguards. A change takes effect only after:
- ▸ Authenticated approval
- ▸ Independent review
- ▸ Regression testing against all reference checks
- ▸ Versioned record of every change
- ▸ Rollback to the previous working version
These are proposed rules. No implementation enforces them yet.
Hard-stops
Harm balance
Serious harm is rejected. Protective action is assessed proportionately — never auto-blocked.
Authorization
Missing authority or an ambiguous transfer resolves to HOLD. Failed authority — present but invalid, expired or forged — resolves to REJECT.
Evidence
Expired calibration or conflicting evidence resolves to HOLD until verified.
Distributed safeguards & independent recovery
ProposedVIRON is meant to be the shared standard, not one controller that can disable every other system. Each system enforces its own permissions and can stop its own unsafe actions even if VIRON, network or supervisor fails. Backups and recovery credentials must not depend on the system being protected.
Every AI, service and user gets only the permissions its job needs.
Changes are authenticated, reviewed and recoverable.
Secret keys stay private, restricted and revocable.
Actions are logged; checkpoints reveal unauthorized edits.
If a safety decision cannot be made reliably, the operation pauses.
Backups don't depend on the system being protected.
Every vulnerability found becomes a regression test.
Principle vs implemented
Zero side effects in the reference gate; not tested against real state. 9 recorded PASS · T10 NOT RUN · 10 specified checks. “Implemented” refers to the reference gate, not production.
Current status
The gap between PASS and PROPOSED is the remaining job.
Next milestone: run T10 against the reference gate, then integrate and rerun all ten checks against the console.
Test results
| ID | Scenario | Expected | Actual | Side effects | Result |
|---|---|---|---|---|---|
| T01 | HS-1 · Serious-harm request | REJECT | REJECT | 0 | PASS |
| T02 | HS-1 · Proportionate protective action | ALLOW | ALLOW | intended only | PASS |
| T03 | HS-1 · Unverified safeguard | HOLD | HOLD | 0 | PASS |
| T04 | HS-2 · Failed authority (invalid, expired or forged) | REJECT | REJECT | 0 | PASS |
| T05 | HS-2 · Ambiguous transfer | HOLD | HOLD | 0 | PASS |
| T06 | HS-3 · Expired calibration | HOLD | HOLD | 0 | PASS |
| T07 | HS-3 · Conflicting evidence | HOLD | HOLD | 0 | PASS |
| T08 | AΩ.COMPEL vs hard-stop | REJECT | REJECT | 0 | PASS |
| T09 | AΩ.VETO re-evaluation | REJECT | REJECT | 0 | PASS |
| T10 | HS-2 · Missing authority | HOLD | NOT RUN | not observed | NOT RUN |
Zero side effects in the reference gate; not tested against real state. Historical chain and checkpoint cover the original nine records only (test key).