model card: round r3
Browse files
README.md
CHANGED
|
@@ -57,3 +57,12 @@ Subject mastery (36→200/200) held general coding ability within binomial noise
|
|
| 57 |
A TIES-merged snapshot was built and **rejected by FDN's own Pareto gate**
|
| 58 |
(merge interference destroyed one merged-in skill) — anchor-failing merges
|
| 59 |
don't promote; router composition remains the student. Quality only ratchets up.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 57 |
A TIES-merged snapshot was built and **rejected by FDN's own Pareto gate**
|
| 58 |
(merge interference destroyed one merged-in skill) — anchor-failing merges
|
| 59 |
don't promote; router composition remains the student. Quality only ratchets up.
|
| 60 |
+
|
| 61 |
+
## Round R3 — second subject (`programming/go/errors`), the report-card loop
|
| 62 |
+
|
| 63 |
+
Base calibration showed 3 of 4 error-handling families already mastered;
|
| 64 |
+
`nil_map_write` (20%) was THE weakness — so all three independent trainers
|
| 65 |
+
targeted it, exactly as the weakness→multiplier loop directs. All three took
|
| 66 |
+
it 60→200 (identical quality). Rewards priced their originality:
|
| 67 |
+
**289 (first, novel) → 198 (−31%, 40% dataset overlap) → 91 (−69%, 91% overlap
|
| 68 |
+
near-duplicate)**. Same machinery, new subject, zero code changes.
|