Chapter 63
Validation report — full suite
github.com/openfluke/w2a✅ 228k cells
Why it exists
Claims are cheap; a stamped matrix is not. This is the actual output of one full w2a run so the book's ✅ marks are backed by numbers you can reproduce, not asserted.
What it is
Every timed layer sweeps its dtype × format × backend matrix; every suite runs its case checks. Latest full board: FAIL 0, cases all PASS; honesty stamps still record GAP cells (not silent skips).
Full suite: PASS228,054 matrix cells — OK 227,194 · GAP 860 · FAIL 0. 456 suite cases — PASS 456 · FAIL 0. Elapsed 1h21m32s. Good enough for Caber / GitHub release: no fails, gaps are declared.
Coverage by layer (highlights)
| Layer | Cells | OK | GAP | Cases |
|---|---|---|---|---|
| step | 43,903 | 43,903 | 0 | 13 |
| tween | 32,457 | 32,457 | 0 | 10 |
| dna | 16,159 | 16,159 | 0 | 6 |
| evolution | 16,152 | 16,152 | 0 | 6 |
| dense | 5,775 | 5,774 | 1 | 25 |
| mha · swiglu · cnn* · rnn/lstm · norms · … | ~5,763 ea | ~all OK | 0–few | 14–23 ea |
| parallel | 6,415 | 6,369 | 46 | 20 |
| convt1–3 | 5,763 ea | 5,649 ea | 114 ea | 14 ea |
| gdn | 555 | 93 | 462 | 17 |
| mamba | 5,763 | 5,755 | 8 | 14 |
| Total | 228,054 | 227,194 | 860 | 456 |
Cross-numeric train
W2A Step includes Cross-Numeric Train (kinds × weight dtype × act host; full census ~10.7k under the train op). Storage truth checked after StepMesh — no retained f32 master for non-float32 FormatNone. Public Dense demotion + W×A showcase: down-the-dem (chapter 65).
Go example
examples/63-validation/main.go
Run:
cd welvet/examples/63-validation && source ../env.sh && go run .package main
import "fmt"
func main() {
cells, ok, gap, cases := 228054, 227194, 860, 456
fmt.Printf("w2a suite: %d cells (OK %d GAP %d FAIL 0)\n", cells, ok, gap)
fmt.Printf("cases: %d/%d PASS result: PASS\n", cases, cases)
}
Output
w2a suite: 228054 cells (OK 227194 GAP 860 FAIL 0)
cases: 456/456 PASS result: PASS
Validate (harness)
cd w2a && go run . # full timed suite (~1h20m); writes logs/suite.txt