Model card: property suite scores 4/4 on the external corpus
Browse files
README.md
CHANGED
|
@@ -101,9 +101,11 @@ things that run*, not argued ([full results](web/TEST_RESULTS.md)):
|
|
| 101 |
standard in exact BigInt arithmetic proves the JS epilogue mirror is
|
| 102 |
spec-correct — and rejects the old round-once mirror on 34% of inputs.
|
| 103 |
- **Metamorphic properties + oracle mutation scoring** (`test_metamorphic.js`,
|
| 104 |
-
`test_corpus.js`): properties needing no reference implementation,
|
| 105 |
-
|
| 106 |
-
|
|
|
|
|
|
|
| 107 |
- **Exact kernel gates on every live kernel**, a continuous random-cell audit
|
| 108 |
at live shapes, and a cross-device kernel probe (same seeded int8 GEMM,
|
| 109 |
same hash on every honest device, any backend).
|
|
|
|
| 101 |
standard in exact BigInt arithmetic proves the JS epilogue mirror is
|
| 102 |
spec-correct — and rejects the old round-once mirror on 34% of inputs.
|
| 103 |
- **Metamorphic properties + oracle mutation scoring** (`test_metamorphic.js`,
|
| 104 |
+
`test_corpus.js`): properties needing no reference implementation, scored
|
| 105 |
+
against an externally-authored bug taxonomy — **4/4**, matching the exact
|
| 106 |
+
differential gate's 4/4. Relations own the loop bugs; two definitional
|
| 107 |
+
absolutes (ReLU output range, a unit-scale integer anchor) own the value
|
| 108 |
+
bugs no relation can see.
|
| 109 |
- **Exact kernel gates on every live kernel**, a continuous random-cell audit
|
| 110 |
at live shapes, and a cross-device kernel probe (same seeded int8 GEMM,
|
| 111 |
same hash on every honest device, any backend).
|