Update README.md
Browse files
README.md
CHANGED
|
@@ -1,10 +1,25 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Loko Release Gate Records
|
| 2 |
+
|
| 3 |
+
This repository contains public Release Gate decision artifacts for pinned vision model revisions.
|
| 4 |
+
|
| 5 |
+
Each run documents:
|
| 6 |
+
Model repository and exact revision
|
| 7 |
+
Dataset repository and exact revision
|
| 8 |
+
Declared release predicate
|
| 9 |
+
Deterministic PASS/FAIL decision
|
| 10 |
+
Receipt bundle and offline report
|
| 11 |
+
|
| 12 |
+
The purpose is simple:
|
| 13 |
+
To demonstrate that model releases can be governed by an explicit, machine-verifiable decision process.
|
| 14 |
+
|
| 15 |
+
Run A — Baseline Repro (COCO)
|
| 16 |
+
Pinned YOLO model evaluated on COCO val2017 to reproduce benchmark surface.
|
| 17 |
+
This run demonstrates deterministic reproducibility and release decision recording.
|
| 18 |
+
|
| 19 |
+
Run B — Production Gate (Low-Light, Class-Scoped)
|
| 20 |
+
The same pinned YOLO revision evaluated under operational criteria using a low-light detection dataset.
|
| 21 |
+
This run demonstrates deployment-grade gating for a specific class (e.g., person) under stricter production predicates.
|
| 22 |
+
|
| 23 |
+
What This Is Not
|
| 24 |
+
This repository does not claim model superiority or benchmark competition.
|
| 25 |
+
It demonstrates auditable release governance.
|