THE ML MODEL RELEASE GATE ISSUES A SIGNED PASS/FAIL RELEASE RECORD.

Evaluate locally.
Enforce thresholds.
Seal evidence.
ISSUE the release record.
PASS or FAIL.

Download the Evaluation Runner

The evaluation runner is the local Benchmark lane.
‍
Run it to produce a deterministic receipt bundle for a signed PASS/FAIL release record.

Enter your evaluation code to download:
‍

lokogate_runner-0.1.2-py3-none-any.whl
‍
‍‍lokogate_runner-0.1.2-py3-none-any.whl.sha256
‍‍

Offline verification: no network required during the run.
‍
‍Integrity checks: SHA256 published for every download.
‍
‍Binary trust: code-signing status stated clearly (SBOM available on request).
‍
‍Safe to share: the receipt bundle contains only verification outputs when using the sample package.
‍
‍Return options: secure upload.

Enter your evaluation code to mint time-limited downloads.

RUN THE EVALUATION

Run the evaluation runner locally to generate a deterministic receipt bundle.
‍

Time to run: 10-15 minutes.

Accepted model: YOLO object detection (.pt weights).
‍
‍Labels: YOLO bounding box format (class_id 0 only).
‍
‍Inference size: Fixed at 640 (Benchmark scope).

Evidence directory must follow standard YOLO val structure (images + labels).

Linux (Ubuntu 22.04+, x86_64)

1. Create a workspace and enter it.
‍

mkdir -p ~/lokogate
cd ~/lokogate

2. Place the downloaded wheel and `.sha256` file in this folder and verify integrity.
‍

sha256sum -c *.sha256

3. Create a virtual environment and install the runner.
‍

python3 -m venv .venv
source .venv/bin/activate

pip install -U pip
pip install ./*.whl

lokogate --version

4. Initialize the workspace and run the Benchmark evaluation and point to your model and validation/evidence set.
‍

lokogate init --workspace "$(pwd)"

lokogate run \
  --workspace "$(pwd)" \
  --gate "$(pwd)/policies/gate_benchmark_v1.yaml" \
  --weights /ABS/PATH/TO/weights.pt \
  --evidence-dir /ABS/PATH/TO/evidence_root \
  --evidence-id customer_val_01 \
  --device 0 \
  --model-id model_v1

5. The upload artifact will be created under:
‍

$(pwd)/records/<gate_id>/<model_id>/<RUN_ID>/bundle/<RUN_ID>_upload_payload.zip

UPLOAD THE RECEIPT BUNDLE

Upload exists for one reason: to verify the signed release record.

Local outputs

Full HTML report (stays local):

report/report.html

Receipt bundle (upload this):

bundle/<RUN_ID>_upload_payload.zip

Upload

Enter evaluation code + RUN_ID.

RUN_ID format: `YYYYMMDDThhmmssZ`

Enter your evaluation code to mint time-limited upload & download links.

Upload: bundle/<RUN_ID>_upload_payload.zip

Download

When ready, download:`<RUN_ID>_release.loko.zip`
`<RUN_ID>_release.loko.zip.sha256`

ML MODEL Release Gate Lanes

Benchmark is a free, deterministic evaluation lane.

Need more than fixed-scope evaluation?
‍
Pilot and Production expand governance authority.

Capability	Benchmark (Free)	Pilot	Production
Class scope	1 class	Multi-class	Full class map
Resolution	Fixed 640	Declared	Enforced
Stress	Diagnostic	Expanded	Gating optional
Delta	Informational	Optional enforce	Enforced
Policy	Fixed	Custom	Organizational standard
Certification	Yes	Yes	Yes (governed)
Governance posture	Demo	Controlled	Release authority

Benchmark

Free evaluation lane. Fixed scope. Deterministic certification.
‍

Pilot

Expanded validation with custom policy and regression control.
‍

Production

Formal release authorization infrastructure.

Request a control plane briefing

If it doesn’t fail deterministically,
it’s not evidence.

Ml MODEL RELEASE GATE

The evaluation runner is the local Benchmark lane.‍Run it to produce a deterministic receipt bundle for a signed PASS/FAIL release record.

Enter your evaluation code to download:‍

Linux (Ubuntu 22.04+, x86_64)

1. Create a workspace and enter it.‍

2. Place the downloaded wheel and .sha256 file in this folder and verify integrity.‍

3. Create a virtual environment and install the runner.‍

4. Initialize the workspace and run the Benchmark evaluation and point to your model and validation/evidence set. ‍

5. The upload artifact will be created under: ‍