1.0 readiness
Snapshot: 2026-11-12. Internal engineering is complete for the 1.0 scope in ROADMAP. 1.0 is not declared because both remaining gates need a person who is not the author.
Internal evidence
| Gate | Current result |
|---|---|
Full local CI, make ci N=0 | green; recorded after the 2026-11-12 sweep |
Fast per-change gate, make check-fast | green |
| Differential fixtures | 1,159 jobs, 0 failed |
| Corelib | 101 test trees, all green at the 0.9.2 sweep |
Deep fuzz, fuzz N=200 | 200/200, 0 findings |
Reject fuzz, fuzz-reject N=200 | 200/200 (19 accepted, 181 rejected) |
Leak fuzz, fuzz-leak N=150 | 150/150 |
TSan fuzz, fuzz-tsan N=200 | 63 ran, 137 skipped by design, 0 failed |
New-codec fuzz, make fuzz-newcomers | green in CI; 56,000 inputs in the deeper campaign |
| Benchmark suite gate | green on saved GCC and clang results |
| Release archives | release-check reproduces byte-identical linux-x86_64 and mingw64-x86_64 archives; release-content is in CI |
The tree has one maintained compiler. compiler/ is the self-hosted implementation; bootstrap/tycho_compiler.c is the generated cold-start artifact; the historical src/tychoc.c is deleted and its citations resolve through git history.
Performance snapshot (large rows)
| Row | Tycho | C |
|---|---|---|
dispatch default | 13 ms | 13 ms |
dispatch expert | 12 ms GCC / 26 ms clang | 13 / 30 |
json_tree time | 51 ms | 49 ms |
json_tree RSS | 81.7 MB | 103.6 MB |
lru time / RSS | 226 ms / 14.7 MB | 233 ms / 12.0 MB |
trie time | 89 ms / 90 ms clang | 79 / 74 |
binary/arena | 377 ms GCC / 516 ms clang | 292 / 504 |
The remaining losses are structural: the generic int map's duplicate key/value/index arrays (lru RSS) and the last few percent in trie and dijkstra. The expert binary/arena row now uses a 16-byte null-leaf node and is at clang parity; GCC spends the difference in recursive arena allocation.
What is still missing
1. A real program written by someone who is not the author
This is the blocking 1.0 gate. Four outside probes exist — tools/tycho-diff, tools/tycho-hash, tools/tycho-fold, and the 2026-08-19 FFI probe — each with its friction recorded. They do not cover generics, subscript, bounded[N], select, and error paths. The procedure is probe-procedure.md.
2. An external security review
No third-party audit has been performed. The reviewer packet is audit-brief.md. Three internal passes found 5, 9 and 10 issues with zero overlap, which is why this remains a requirement rather than a conclusion.
3. Owner actions outside the code
- Keep the macOS arm64 archive in the release set. 0.9.2 has a native Apple Silicon build; 0.9.1 shipped Linux and Windows only.
- Decide whether the ergonomics backlog is 1.0 work or post-1.0 work.
make status is the live health check. make release-check rebuilds and compares the archives. The commands in this document are the reproducible evidence for the table above.