Delete the six duplicated src/abi/*.idr modules and repoint every reference

[hyperpolymath]

Closes #815. Follow-up to #120, as agreed: #120 repointed the Elixir drift guard onto the live module; this PR deletes the dead copies and corrects every reference.

What was duplicated

src/abi/hypatia-abi.ipkg sets sourcedir = "..", which resolves to src/, so the modules it compiles are src/Hypatia/ABI/*.idr. A byte-identical copy of all six sat in src/abi/, compiled by nothing — divergence between them was undetectable. That is DEBT-REGISTER C-4, marked RESOLVED here.

Both .ipkg files and README.adoc stay, so src/abi/ still exists and idris2 --build src/abi/hypatia-abi.ipkg is unaffected. Verified directly — it builds all 6 modules from ../Hypatia/ABI/, which is the proof the deleted copies were never compiled:

2/6: Building Hypatia.ABI.RuleEngine (../Hypatia/ABI/RuleEngine.idr)
...
6/6: Building Hypatia.ABI.GraphQL (../Hypatia/ABI/GraphQL.idr)

Proven lossless before deleting

Five of the six blobs were byte-identical across the two directories. Types.idr differed, so I checked rather than assumed: the dead blob 40d2ed6c is a strict ancestor revision of the live 788493e6 — the live copy carries #120's authority-direction correction on top. Nothing is lost that the live module's own history does not already contain.

The one functional consumer — and the vacuous gate behind it

test/zig_ffi_smoke_test.exs asserted File.exists? on five src/abi/*.idr modules, so deleting them turned mix test red.

Its sibling test was worse. It listed the directory, filtered for .idr, and asserted SPDX headers — after deletion that list is empty, so Enum.each over nothing passes. It would have gone vacuous, not red: a deletion silently disarming a gate while the gate next to it screamed.

Both are fixed. @abi_dir now points at src/Hypatia/ABI; the required-module list gains RuleEngine.idr and Gen.idr (it named only 5 of 7 — a pre-existing coverage gap); and the SPDX test gets a non-empty denominator assertion.

Mutant-tested. Pointing @abi_dir at an empty directory now gives 2 failures including no .idr files found in …, where before it would have given 1:

2) test Idris2 ABI source integrity Idris2 ABI modules have SPDX headers
   no .idr files found in /…/emptyabi
   code: assert idr_files != [], "no .idr files found in #{@abi_dir}"
19 tests, 2 failures

Worth flagging for future sweeps: this consumer was missed by three earlier searches because it holds the directory ("../src/abi") and the basenames (~w(Types.idr …)) as separate tokens, so grepping the joined path never matches it — and because I had been searching for code that reads those files rather than code that asserts they exist.

Other corrections found on the way

• .hypatia-exemptions.adoc cited an inline marker -- hypatia:ignore … that does not exist (git grep hypatia:ignore → nothing). The real directive is -- hypatia: allow …. The prose above it claimed the scanner has no directive recognition — also false; scanner_suppression.ex file_allowed?/4 implements it. Repoint only, no new row: 6 insertions, 6 deletions, so the Exemption ratchet sees no growth.
• docs/proofs/HANDOVER-neural-convergence.adoc asserted that src/abi/*.idr "are the real files" and src/Hypatia/ABI/*.idr "are symlinks into them" — backwards, and they were never symlinks.
• .github/CODEOWNERS owned /src/abi/ but nothing owned /src/Hypatia/ABI/, leaving the normative ABI unowned. Added.
• stapeln.toml [layers.abi-verify] had cache-key = "src/abi/*.idr", which after deletion would match nothing and cache forever.

Deliberately untouched

• verification/PROOF-STATUS.adoc — verification/PROOF-STATUS.adoc is stale: wrong path, wrong LOC, wrong proofs #816 ("stale: wrong path, wrong LOC, wrong proofs") owns the path column; two PRs editing the same six rows would collide. Commented on verification/PROOF-STATUS.adoc is stale: wrong path, wrong LOC, wrong proofs #816 instead.
• test/unified-api-adapter-contract_test.exs:11 — a past-tense note explaining why chore(deps): bump thiserror from 1.0.69 to 2.0.18 #120 repointed the guard. Still historically true.
• verify-proofs.yml paths:, CODEOWNERS /src/abi/, Justfile compile-abi/abi-gen, pack.toml, src/README.adoc — all reference the directory or the ipkgs, both of which survive.

Verification

• mix test → 1593 tests, 2 failures. Both pre-existing: proven by running test/research_extensions_test.exs at stock 86aa936 content, which gives the same 2. They are harden-runner rule tests, unrelated to ABI paths.
• just abi-gen → connectors emitted: 16 / files written: 3 of 3, zero diff against the tracked generated files.
• tests.yml dangerous-pattern denominator: 13 → 7, still non-zero.
• The scanner's -- hypatia: allow code_safety/believe_me structural_drift/SD008 marker survives at line 19 of the live RuleEngine.idr, matching what the exemptions ledger cites.

🤖 Generated with Claude Code

https://claude.ai/code/session_0113HQM9LVGkNCzU1WwkJZSV

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Advanced

Run ID: f4f3bc35-373d-404d-b9f2-95aae9972b8e

📥 Commits

Reviewing files that changed from the base of the PR and between 86aa936 and a27d78b.

📒 Files selected for processing (31)

• .claude/CLAUDE.md
• .github/CODEOWNERS
• .github/workflows/tests.yml
• .hypatia-exemptions.adoc
• Justfile
• clients/rust/hypatia-client/src/ffi.rs
• clients/rust/hypatia-client/src/lib.rs
• clients/rust/hypatia-client/src/types.rs
• docs/DEBT-REGISTER.adoc
• docs/governance/CRG-AUDIT-2026-04-18.adoc
• docs/operations/prover-wars-integration.adoc
• docs/proofs/HANDOVER-neural-convergence.adoc
• docs/proofs/needs.adoc
• ffi/README.adoc
• ffi/zig/README.adoc
• ffi/zig/src/main.zig
• lib/hypatia/scanner_suppression.ex
• schema/README.adoc
• src/abi/FFI.idr
• src/abi/GRPC.idr
• src/abi/GraphQL.idr
• src/abi/README.adoc
• src/abi/REST.idr
• src/abi/RuleEngine.idr
• src/abi/Types.idr
• stapeln.toml
• test/merge_orchestration/sensor_test.exs
• test/zig_ffi_smoke_test.exs
• verification/proofs/idris2/ConfidenceBounds.idr
• verification/proofs/idris2/DispatchStrategy.idr
• verification/proofs/idris2/SafetyTriangle.idr

 ___________________________________________
< GPU-powered code review. It's the future. >
 -------------------------------------------
  \
   \   (\__/)
       (•ㅅ•)
       / 　 づ

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

• 🔴 Error committing to branch - (🔄 Check to retry)
• Create a new PR

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[hyperpolymath]

Check results: 22 success, 4 failure, 4 skipped — none of the 4 caused by this PR

main at 86aa936 carries 9 failures; this PR carries 4, and each is accounted for below. I checked provenance rather than asserting it, because "the failures look unrelated" is not evidence.

Failing check	Provenance
governance / Code quality + docs	Already red on main. ##[error]Missing required documentation: CONTRIBUTING — a missing repo file, nothing to do with ABI paths.
governance / Validate Hypatia Baseline	Already red on main.
governance / Actions lockfile verify	Known, tracked as #818. This PR adds no uses: entry, so it cannot affect the lock.
governance / Workflow security linter	duplicate key(s): 'with' (line 143) → build-gossamer-gui.yml, a step with two with: blocks, introduced by #810 earlier today. Filed as #822.

The last two did not run on main's push event, so "not in main's failure list" would have been a false acquittal for them. They are cleared by mechanism instead: build-gossamer-gui.yml is byte-identical between this branch and main (git diff --stat origin/main HEAD -- .github/workflows/build-gossamer-gui.yml is empty), and this PR's only workflow change is a single comment line inside a run: block in tests.yml — which cannot create a duplicate YAML key or alter a lockfile.

Worth noting positively: governance / Exemption ratchet and governance / Debt ratchet both pass. Those were the two most likely to object, since this PR edits .hypatia-exemptions.adoc and docs/DEBT-REGISTER.adoc. The exemptions edit is a repoint with no new row (6 insertions, 6 deletions), so the ratchet sees no growth.

🤖 Generated with Claude Code

https://claude.ai/code/session_0113HQM9LVGkNCzU1WwkJZSV

[coderabbitai]

❌ Failed to create Coding Agent finishing-touch task. Please try again.

[coderabbitai]

❌ Failed to create Coding Agent finishing-touch task. Please try again.