Skip to content
AzielEliabPublic

About

AZ-CLCE: Cross-Layer Consistency Engine. Detects inconsistency across representation (R), description (D), and reality (P). Not intent, not a scanner, not a lie detector.

Topics

Resources

Contributing

Stars

2 stars

Watchers

0 watching

Forks

Repository files navigation

AZ-CLCE

Compare what something looks like, what was written, and what it actually does.

Author: Aziel Eliab Version: 0.3.0 License: Apache-2.0

Start

  1. Install

    python -m venv .venv && source .venv/bin/activate && pip install -e .
  2. Open the local page

    clce ui
  3. Open the address it prints (http://127.0.0.1:8845/) and press Score.

Check this install with clce doctor. List commands with clce --help.

See RUN.txt. Spec: docs/whitepaper.md. How to contribute: CONTRIBUTING.md.

Forks are welcome and always allowed.

One-click install

curl -fsSL https://azclce-download-tracker.vibelock.workers.dev/install.sh | bash

The script downloads the counted tarball, extracts it, makes a venv, and runs pip install -e .. Then:

source .venv/bin/activate
clce ui

Counted download (Cloudflare Worker)

This is the counted download. GitHub releases exist as a mirror. The Worker serves the gzip itself (HTTP 200, no 302 to GitHub).

Direct tarball (also counted): az-clce-0.3.0.tar.gz

Isolated counter: Worker azclce-download-tracker, KV AZCLCE_DOWNLOADS. Not mixed with any other product. /v1 does not increment downloads.

Notes

  • CLCE detects inconsistency, not intent. Type D is a label, not a finding of malice.
  • SPRE and CLCE are two of three Aziel triad verifiers. PhysLing lives in aziel-corpus. Merge fields are 0–1 (score_100 is 0–100).
  • SPRE scores structural similarity to historically confirmed failures. Never guilt or conspiracy.
  • Official narrative is not evidence. Official-only stories lower E and raise poison-suspicion flags.
  • Human validation required.
  • Not a cybersecurity exploit, not a scanner of other people's systems, not a lie detector.
  • Advisory scores only. Threshold 0.7 is the paper's "acceptable" line, not a pass/fail of truth.
  • Loopback UI, no CDN, no telemetry. Size limits on inputs. Empty fields are OK.
  • Standalone from ForgeReceipts, ZionPattern, DecisionGATE, AZ-OS, Glossa Filter, MirageGrid.
  • Not a VPN. AzielTether is a software queue, not MirageGrid.
  • AZCoherence is a peer Softwares product, not this engine. Separate product. Do not merge. CLCE is not Coherence. AKM-TRIAD-1.0 stays fabric.

What it computes

Layers are normalized token sets (lowercase, split on non-alnum).

Quantity Formula
Triple |R ∩ D ∩ P| / |R ∪ D ∪ P| (empty-all → 1.0)
Pairwise R↔D, D↔P, R↔P as Jaccard; §5 final = average of three
CLCE+ |R ∩ D ∩ P| / (|R ∪ D ∪ P| + |N|)

Band: 1.0 perfect; ≥0.7 acceptable; <0.7 structural inconsistency.

Mismatch types (deterministic)

Type When Meaning
A Surface Error R↔D low (<0.7) while D↔P and R↔P are higher Docs/UI disagree; function closer to one layer
B Functional Error R↔D high (≥0.7) while D↔P or R↔P is low Pretty alignment, function diverges
C Structural Gap High |N| relative to union, or all pairwise mediocre and triple <0.7 Hidden or missing structure
D Intentional Obfuscation LABEL ONLY. High N AND D↔P very low AND R↔D high Representation matches description while reality and missing-elements diverge. Never a finding of malice.

The report lists every matching type and prefers the most severe (D > C > B > A) as primary.

The page shows that sentence after Score. Jaccard numbers and types A–D are under Advanced.

Install

Python 3.10+. Stdlib only at runtime.

python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

CLI

People get short text. Programs add --json and receive the same report as before.

clce                                             # welcome and next step
clce --help
clce version
clce doctor                                      # self-check, no network
clce ui                                          # prints Open http://127.0.0.1:8845/
clce score --r "..." --d "..." --p "..." [--n "..."]
clce score --json --import layers.json --export report.json
clce classify --r ... --d ... --p ... [--n ...]
clce gate --min 0.7 --r ... --d ... --p ...      # exit 0 if triple ≥ min else 1
clce verify-transfer PATH                        # short summary
clce verify-transfer --json PATH                 # full report
clce verify-transfer older_payloads/ --backfill --ndjson
spre score --official "..." --physics "..." --json
spre score older_payloads/ --ndjson              # batch/backfill directory
spre verify-transfer PATH

--import reads JSON or labeled .txt {r,d,p,n}. --export writes the report JSON and a sibling .txt receipt with input_sha256. --json prints the full report. --help lists ui, doctor, and version. CLCE_DEBUG=1 logs tokens and scores to stderr.

Library

from clce import score, classify, gate

report = score(
    r="login button blue",
    d="login form submits",
    p="login button submits",
)
print(report.triple, report.pairwise_avg, report.plus, report.band)
print(report.kid_plain, report.input_sha256)
print(classify(r="...", d="...", p="...", n="csrf session").types)
ok, report = gate(r="a", d="a", p="a", min_score=0.7)

UI

clce ui binds 127.0.0.1:8845 only and prints Open http://127.0.0.1:8845/. Four boxes:

  • What it looks like (R)
  • What they wrote (D)
  • What it actually does (P)
  • Missing pieces (N)

Press Score. Try a sample fills an example. Jaccard numbers, types A–D, import, and export sit under Advanced. Honest scope sits under About. Light and dark follow the system. Focus is a gold ring. Usable at 390px. Self-contained CSS, no CDN, no telemetry.

iPhone & Android

Flutter sources: mobile/. Application id com.azieeliab.azclce. Offline. No analytics. Light and dark follow the system. Gold focus.

Four fields with the same kid-plain labels, giant score, sample fill, paste import / copy export of JSON + receipt text.

cd mobile
flutter create --org com.azieeliab --project-name azclce .
flutter pub get
flutter run

The android/ and ios/ folders in this tree are skeleton READMEs until you run flutter create . (this machine has no Flutter SDK on PATH). Then open android/ in Android Studio or ios/Runner.xcworkspace in Xcode. Not a store listing.

Counted desktop download: https://azclce-download-tracker.vibelock.workers.dev/

Forks are welcome and always allowed.

Tests

pip install -e ".[dev]"
python -m pytest -q

Offline. No network. Stdlib runtime. pytest is the dev extra. Keeps A–D fixtures. Adds import/export roundtrip and clce doctor.

Worker

Isolated download counter for this project only. Worker azclce-download-tracker, project azclce, KV AZCLCE_DOWNLOADS bound as DOWNLOADS. GET /download serves az-clce-0.3.0.tar.gz (does not 302 to GitHub). See workers/download-tracker/README.md.

Counted downloads (number on the button, no user reporting): https://azclce-download-tracker.vibelock.workers.dev/

Layout

clce/                 library (engine, cli, ui, io, doctor, transfer, mesh)
spre/                 SPRE sibling package (SP(c), SSI, PC)
clce/web/             loopback UI
tests/                pytest
docs/whitepaper.md    spec
docs/spre.md          SPRE framework
docs/triad.md         SPRE + CLCE + PhysLing compositing (PhysLing in aziel-corpus)
docs/node-mesh.md     AzielTether hooks + hosted suite /v1/mesh/* PROXY (not a VPN)
docs/ingest-hooks.md  upload/download verify + Worker ingest
docs/source/          v1.0 TXT and v2.0 PDF
mobile/               Flutter iPhone + Android (`flutter create .`)
workers/download-tracker/   Cloudflare Worker

AI runtime

CLCE detects inconsistency, not intent. Type D is a label, not a finding of malice. Threshold 0.7 is advisory.

  • POST https://azclce-download-tracker.vibelock.workers.dev/v1/score {r,d,p,n}
  • POST https://azclce-download-tracker.vibelock.workers.dev/v1/classify {r,d,p,n}
  • POST https://azclce-download-tracker.vibelock.workers.dev/v1/gate {r,d,p,n,min}
  • POST https://azclce-download-tracker.vibelock.workers.dev/v1/spre SPRE case JSON
  • POST https://azclce-download-tracker.vibelock.workers.dev/v1/verify-transfer ingest hook (emits triad)
  • GET https://azclce-download-tracker.vibelock.workers.dev/v1/triad component score schema (0–1 / 0–100)
  • OpenAPI 3.1: https://azclce-download-tracker.vibelock.workers.dev/openapi.json
  • Help: https://azclce-download-tracker.vibelock.workers.dev/ai

/v1 does not increment the download counter.

One-URL catalog: https://aziel-runtime.vibelock.workers.dev/openapi.json MCP catalog (ships separately): https://aziel-runtime.vibelock.workers.dev/mcp. Suite mesh /v1/mesh/* PROXY via AZIEL_RUNTIME (default OFF; QNM-BUILD-1.0 live|locked|isolated; no Node Gate). Catalog MCP mesh_* + FragGate slug=mesh. QNS-CD-1.0 (photon QNS1 packet transfer) is a hub cite / Worker mesh cross-map only — local qnsd in qnm-node; runtime cites in aziel-runtime; pair custody in azinterface. Not a Softwares-tab product. No public qnsd proxy.

Use with Grok / ChatGPT / Venice

Catalog OpenAPI: https://aziel-runtime.vibelock.workers.dev/openapi.json Catalog MCP: POST https://aziel-runtime.vibelock.workers.dev/mcp This Worker skill: https://azclce-download-tracker.vibelock.workers.dev/v1/skill This Worker OpenAPI: https://azclce-download-tracker.vibelock.workers.dev/openapi.json

Grok: import the catalog or Worker OpenAPI as a custom tool. ChatGPT: GPT Actions (no auth). Venice: HTTP tools. Always send User-Agent: Mozilla/5.0.

Peers (do not merge)

AZ-CLCE and AZCoherence are peer Softwares products, not one engine. Separate product. Do not merge. CLCE is not Coherence.

Cite this

Aziel Eliab. AZ-CLCE + SPRE. https://github.com/AzielEliab/az-clce. https://azclce-download-tracker.vibelock.workers.dev.

License

Apache-2.0. See LICENSE.

Forks are welcome and always allowed.

About

AZ-CLCE: Cross-Layer Consistency Engine. Detects inconsistency across representation (R), description (D), and reality (P). Not intent, not a scanner, not a lie detector.

Topics

Resources

Contributing

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages