DMSD Horizon Edition v0.8 — Online Calibration Memory / Learning Router
Append evidence · decay staleness · detect drift candidates · actively benchmark high-value gaps.
Ledger entries
114
114
Cache points
105
105
Historical crossover
94.99
94.99
Refined local crossover
80.91
80.91
Drift candidate
Certificate scalar @100 shifted +30.8%; vectorized shifted +11.6%. Fingerprint unchanged; cause unresolved.
Environment fingerprint
52715b12b3d096388d272ab95e6f963a6b0a620df71023e5ff121926dc292074
| Rank | Candidate | Reason | Priority | Status |
|---|
| Probe | Winner | Key result |
|---|---|---|
| cert @85 | numpy_vectorized | local crossover → ~80.91 |
| GEMM n=96 | numpy_openblas | ~1469× vs Python naive |
| sieve N=30M | odd_only_bytearray | π(30M)=1,857,859 |
Half-life = 4 calibration epochs; routing heuristic, not probability.