docs: remove AI attribution from validation report
Browse filesReplaces KOS-MOS/Gemini validator references with neutral 'independent'/'builder' language per the no-AI-attribution rule for public artifacts.
VALIDATION_COMPRESSOR_V2.md
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
# Independent Validation — Compressor Reflex v2 Re-delivery
|
| 2 |
|
| 3 |
-
**Validator:**
|
| 4 |
**Date:** 2026-10-01
|
| 5 |
**Delivery:** Drive folder `compressor-reflex` (`1E1HBQQdzab-fqx_6MKHaOGi0LLzlIjt8`), 19 files
|
| 6 |
**Local copy:** `~/workspace/compressor-reflex/delivery-v2/`
|
|
@@ -57,7 +57,7 @@ live A/B still open if productizing).
|
|
| 57 |
|
| 58 |
## A. Held-out integrity (independent re-verification)
|
| 59 |
|
| 60 |
-
| Check |
|
| 61 |
|---|---|---|
|
| 62 |
| verbatim item overlap eval↔train | 0.00% | **0/217 = 0.00%** |
|
| 63 |
| verbatim item overlap eval↔val | — | **0/217 = 0.00%** |
|
|
@@ -237,7 +237,7 @@ to 0.50, `eval_report.json` + `REPORT.md` regenerated. Independently verified:
|
|
| 237 |
- **Sweep confirms calibration:** 100% at τ ≤ 0.50; 610/611 at 0.55 and 0.60
|
| 238 |
(the 0.5449 anchor) — exactly matching the independent strict reimplementation.
|
| 239 |
- **Cross-check:** every refreshed number matches the independent
|
| 240 |
-
|
| 241 |
|
| 242 |
## Artifacts
|
| 243 |
|
|
|
|
| 1 |
# Independent Validation — Compressor Reflex v2 Re-delivery
|
| 2 |
|
| 3 |
+
**Validator:** Independent (separate from the build loop)
|
| 4 |
**Date:** 2026-10-01
|
| 5 |
**Delivery:** Drive folder `compressor-reflex` (`1E1HBQQdzab-fqx_6MKHaOGi0LLzlIjt8`), 19 files
|
| 6 |
**Local copy:** `~/workspace/compressor-reflex/delivery-v2/`
|
|
|
|
| 57 |
|
| 58 |
## A. Held-out integrity (independent re-verification)
|
| 59 |
|
| 60 |
+
| Check | Builder claim | Independently measured |
|
| 61 |
|---|---|---|
|
| 62 |
| verbatim item overlap eval↔train | 0.00% | **0/217 = 0.00%** |
|
| 63 |
| verbatim item overlap eval↔val | — | **0/217 = 0.00%** |
|
|
|
|
| 237 |
- **Sweep confirms calibration:** 100% at τ ≤ 0.50; 610/611 at 0.55 and 0.60
|
| 238 |
(the 0.5449 anchor) — exactly matching the independent strict reimplementation.
|
| 239 |
- **Cross-check:** every refreshed number matches the independent
|
| 240 |
+
strict measurements (§F.4): 611/611, 86.73%, 42.90%, 50/50.
|
| 241 |
|
| 242 |
## Artifacts
|
| 243 |
|