Public open-source research preview

Local-first · provider-free · human authority preserved

UVLM CECP Community Edition

Make the contract explicit. Keep the result bounded.

Review one structured action-and-evidence proposal against UVLM-CECP 1.1.0 and receive a deterministic, replayable PASS, HOLD, or REJECT contract receipt—without a model, account, activation, payment, network provider, persistent memory, training route, or donation.

  • v0.2.0-alpha.6
  • Python 0.2.0a6
  • MPL-2.0
  • No charge
  • No model required
  • Unsigned Windows x64
  • Production use not authorized
Choose a download Read assurance & verification

PASS is not approval. CECP evaluates the submitted contract at the declared observation time. It does not authenticate external facts, establish consent or authority, grant permission, provide legal advice, certify compliance, or decide whether a person is truthful, worthy, safe, dangerous, deserving, or eligible for an essential service.

What CECP does

CECP turns a strict proposal and its declared evidence bindings into an inspectable contract result. It preserves the first failed gate instead of allowing a later favorable claim to hide an earlier unresolved rights, effects, means, burden, consent, appeal, rollback, or revocation condition.

01

Validate the proposal

Reject duplicate keys, unknown closed fields, malformed values, invalid identities, broken references, and unsupported contract shapes.

02

Apply a fixed rights floor

Evaluate the declared proposal in a visible order instead of improvising standards after seeing the desired result.

03

Keep burden visible

Record human and third-party burden without turning pain, consent, or exposure into success credit.

04

Explain the first gate

Return ordered reason codes and plain-language context for the first controlling PASS, HOLD, or REJECT condition.

05

Bind the receipt

Record the canonical input digest, profile, software and protocol versions, observation time, and evaluation-context digest.

06

Export a replay bundle

Create canonical input, machine receipt, Markdown and HTML review, release status, file index, and SHA256SUMS ledger.

From proposal to replayable receipt

The same deterministic assessment and verification core supports the command line, typed Python SDK, and loopback-only browser interface.

  1. Describe
    Declare rights, effects, means, burden, consent, appeal, rollback, revocation, and evidence.
  2. Validate
    Reject ambiguity, wrong identity, invalid scope, and broken digests before assessment.
  3. Assess
    Apply rights-floor-v1 at an explicit RFC 3339 UTC observation time.
  4. Inspect
    Read the disposition, first gate, ordered reasons, burden, consent, and evidence posture.
  5. Replay
    Export a closed bundle that another reviewer can hash, inspect, and verify independently.

Three results, narrowly defined

Each result describes the submitted contract. None is a judgment of a person or an automatic instruction to act.

P

PASS — contract passed

The submitted object satisfied the fixed contract at the stated time. Not endorsement, permission, fact verification, compliance, legitimacy, or authority.

H

HOLD — unresolved

Required evidence or a bounded criterion remains unresolved, including an incomplete rights floor, unknown code, burden, consent, appeal, or revocation gap.

R

REJECT — prohibited declaration

The declared proposal contains an effect or means prohibited by the fixed profile. REJECT does not determine a person's worth or credibility.

How CECP can support local-first AI research

CECP does not include or call a model. A researcher can place it after a user-selected local model, rule engine, or human drafting process to evaluate a deliberately structured action-and-evidence proposal before anyone treats model output as approved.

Local model or human
Produces a candidate proposal.
Structured CECP input
Declares rights, evidence, burden, consent, and reversibility.
CECP receipt
Returns bounded PASS, HOLD, or REJECT with replay identities.
Human judgment
Accepts, repairs, declines, or escalates outside the software.

Performance boundary. CECP does not improve model weights or make a model more intelligent. It may improve proposal discipline, explicitness, review consistency, evidence traceability, and reproducibility. Any claim of improved model quality requires a separately declared baseline, exact model identity, task set, metric, and uncertainty.

Five-minute Python start

Begin with the bundled synthetic example. The exact commands, input contract, portable Windows path, SDK, and browser interface are documented in the version-bound Technical Manual.

py -3.12 -m venv .venv
.\.venv\Scripts\python.exe -m pip install --no-deps .\uvlm_cecp_community-0.2.0a6-py3-none-any.whl
.\.venv\Scripts\cecp.exe --version
.\.venv\Scripts\cecp.exe doctor
.\.venv\Scripts\cecp.exe example --output .\cecp-sample.json
.\.venv\Scripts\cecp.exe validate .\cecp-sample.json
.\.venv\Scripts\cecp.exe assess .\cecp-sample.json `
  --observed-at 2026-01-15T12:00:00Z `
  --output-dir .\cecp-assessment
.\.venv\Scripts\cecp.exe explain .\cecp-assessment\CECP_RECEIPT.json --format text
.\.venv\Scripts\cecp.exe verify .\cecp-sample.json .\cecp-assessment\CECP_RECEIPT.json
Open the Technical Manual Read terms & informed use

Choose your download

Each button points to an exact approved Alpha 6 artifact and its named SHA-256 sidecar. Before extraction, installation, or execution, calculate SHA-256 locally and compare every hexadecimal character. Do not run a mismatch.

Complete source · recommended for inspection

Public source ZIP

Best for education, source review, local builds, forks, and direct inspection of the complete Community Edition tree.

  • 305,777 bytes
  • MPL-2.0
  • No checkout
053210fd38739b7e0a67b071932457bf2be70522187419066de55114c039414e
Download source ZIP Source checksum

Python installation

Python wheel

Best for installation inside a controlled Python 3.11 or 3.12 virtual environment.

  • 62,002 bytes
  • Dependency-free core
  • Python
c05602e8726c135ba06145eb5ea4516f933d9fee5d673be32b143d3accbd47ec
Download wheel Wheel checksum

Python source package

Python source distribution

Best for standard source-package inspection and local Python build tooling.

  • 101,705 bytes
  • Source distribution
  • Python
32dac80de5decbbc64907ac3a93fae7606a08185117d4b078d29811bd51b29aa
Download sdist Sdist checksum

Unsigned Windows x64 · extract and run

Windows portable

Best for many Windows evaluators. Extract to a user-owned folder and run without a system-wide installation.

  • 7,918,821 bytes
  • No administrator expected
  • Unsigned
506ce438870ed9e615a69993a25da4062eb53e442586d3662344b49b253581ab
Download portable Portable checksum

Unsigned Windows x64 · transparent per-user installer

Windows installer candidate

Inspectable PowerShell install, verify, start, repair, and manifest-bounded uninstall under the user's local application data.

  • 7,953,311 bytes
  • Per-user path
  • Unsigned
d71b1c2efc150cbbfcbd8adc6135880564ecab32f7ec5d0785b329671645d923
Download installer Installer checksum

Clean public repository

GitHub-ready repository export

A clean public-history repository candidate with no configured remote. Best for repository inspection and import.

  • 306,897 bytes
  • Clean public history
  • No remote
967c9f75db4e300c074097f8ea10ac2c89f678d7dda81237b492673e276db384
Download repository Repository checksum

Verify on Windows PowerShell

Get-Content -LiteralPath '.\ARTIFACT.zip.sha256'
Get-FileHash -Algorithm SHA256 -LiteralPath '.\ARTIFACT.zip'

A matching hash establishes byte identity only. It does not prove truth, safety, legality, consent, authority, malware absence, scientific validity, or suitability.

Advanced provenance downloads

Researchers and maintainers may also inspect the clean Git bundle and master exact-candidate checksum ledger. Assurance evidence and human authorization records are organized on the Assurance & Verification page.

Git source bundle Bundle checksum Master SHA256SUMS Open assurance library

Known limitations and prohibited uses

This alpha keeps its limitations visible. Review the exact Terms & Informed Use and Assurance pages before moving beyond synthetic data.

Current limitations

  • Windows packages are unsigned; SmartScreen or organizational policy may warn or block them.
  • Builder and review evidence do not create production, legal, compliance, accessibility, or penetration-test certification.
  • Evidence validation checks declared structure and bindings, not external facts, documents, events, people, authority, or consent.
  • Same-host deterministic evidence is not universal cross-host reproducibility.
  • The loopback browser is intentionally local and minimal.

Do not use CECP to score people

  • Human worth, credibility, ideology, religion, culture, trauma, diagnosis, dangerousness, or deservingness.
  • Employment, benefits, detention, punishment, eligibility, or access to essential services.
  • Survivor, witness, employee, complainant, or source truthfulness.
  • Permission to act merely because a contract returned PASS.
Read the complete informed-use boundary Email CECP support

Support open research, if you choose

Ultra Verba Lux Mentis is a nonprofit research organization. Voluntary gifts may support accessibility work, independent review, documentation, maintenance, public education, and grant-sponsored access.

Access to CECP, its contract results, corrections, security treatment, review, ordinary support posture, contributor standing, and governance outcomes never depend on whether someone donates.

Review voluntary donation options

Release status: public open-source research preview. Production release is not authorized. Any byte change creates a new candidate identity and requires new review treatment.