CI: main has no build.zig matching its own source tree (2217 files vs a build script from March)

[gHashTag]

Summary

CI / Build & Test had failed 12 of the last 12 runs at the very first step, hiding every other gate behind it. Two causes, one fixed, one needing your artefact.

What was fixed (#615)

1. build.zig was absent from main while build.zig.zon and ~2800 .zig sources were present — a Zig project root missing only its build script. Restored from cleanup/git-history-reduce-6gb-to-220mb.
2. ci.yml invokes zig build author-guard, but no build.zig on any branch defines that step (checked cleanup/git-history-reduce-6gb-to-220mb, feat/tri-net-skill-radio-ops, and the working tree — zero matches). Since it ran first and could never pass, it masked the build, tests, format check and schema-drift gates. Marked continue-on-error — deliberately not deleted, since it encodes a policy that should be retired by decision, not as a side effect.

Result: CI now runs 1m18s instead of 23s and reaches the real Build step for the first time.

What still needs you

The restored build script is from 2026-03-18; main is at 2026-08-05. It references sources that the tree has since moved on from:

src/bigint.zig:      error: unable to load 'bigint.zig': FileNotFound
src/hybrid.zig:      FileNotFound
src/packed_trit.zig: FileNotFound
src/vsa.zig:         FileNotFound
src/vsa/agent.zig:   FileNotFound
src/vsa/core.zig:    FileNotFound

Measured:

	src/**.zig
main (2026-08-05)	2217
pre-cleanup branch (2026-03-18)	1548
in March but not on main	98

So main has grown, not been gutted — those 98 look like five months of ordinary refactoring. The conclusion is therefore not "files were lost", but:

main has never contained a build.zig that matches its own source tree.

The one that does exist is in a working tree / on feat/tri-net-skill-radio-ops, and it does not define author-guard either.

Suggested resolution

1. Commit the build.zig that actually builds current main (yours, if it is the working one).
2. Either implement an author-guard step in it, or remove that step from ci.yml — a deliberate choice either way.
3. Then the format check, leak detection and schema-drift gates in ci.yml will run for the first time in months, and are likely to have real findings.

Until step 1, Build & Test cannot go green — but it now fails honestly, which is why this issue exists rather than a silently red badge.

🤖 Generated with Claude Code