I built five SQLite engines from one spec in a week

Refresh of engine-source tarballs to include Pin α29-pc-rebase splice-site extension applied across leap-zig (16 sites; previously the source of 119 random/expr timeouts) plus a defensive splice-site rebase in leap-python insert_compile.py (CASE-in-VALUES via Goto/IfNot, latent — not corpus-visible).

Audit confirmed leap-c, leap-rust, leap-go architecturally exempt: their splice paths already rebase PC-relative ops on intake (emitter_append_list, rebase_pcs/append_with_rebase/splice_compiled, AppendWithRebase) or never side-build PC-relative ops at all. No material changes to those three trees vs. v0.1.0.

Corpus on this build (Linux x86_64 native, 622 files, 90 s/file, 240 s python):

| Target | excl-SKIP | exec/mainline | crashes | timeouts |
|--------|-----------|---------------|---------|----------|
| rust | 99.98% | 99.97% | 0 | 2 |
| c | 99.97% | 99.96% | 0 | 2 |
| zig | 98.88% | 99.96% | 0 | 2 |
| go | 99.92% | 99.93% | 0 | 2 |
| python | 99.98% | 98.94% | 3 | 11 |
| sqlite | 100.00% | 100.00% | 0 | 0 |

Source: tests/sqllogictest/results/corpus_2026_04_30_post_zigrebase/summary.json. The 2 timeouts on every target are the same 2 files (select4.test/select5.test, deeply nested 5-way unindexed equijoin planner stress).

Verify with shasum -a 256 -c SHA256SUMS. See bench/PUBLISHED.md §A.1 and README.md §What's load-bearing today for full per-target framing.