feat(circuits): release 0.15.0 with memo-bound transfer/unshield v2 beside v1, and multi-version build tooling

[nol4lej]

Description

Releases @orbinum/circuits 0.15.0. It ships memo-bound v2 of transfer and unshield alongside the published v1, with v2 active, and removes the unused value_proof circuit. The chain activates v2 and retires v1 in the runtime spec 16 upgrade, using keys embedded in the runtime.

Ceremony (v2):

• Single contributor on pot16_final.ptau, with fresh random entropy for each circuit.
• Final beacon: hash of Orbinum testnet finalized block #1045242, 0x39c0d3042e4de951d72540b487175894ae4e3092237c4d71833e2884acb1f6ec, with 2^10 iterations.

Circuit	v1 vk_hash (unchanged)	v2 vk_hash
transfer	0x4f93eaab…4f21	0x8eb9dbb295ce10498fbe7088a4b0f327a675094d657170725d02770711634525
unshield	0x9928482f…945e	0x729c4cced3306e89d0dd977448768cc9c39cc806a393dea8cd8fe2946eb19bab

v1 artifacts and their manifest entries are byte-identical to 0.14.0.

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that causes existing functionality to not work as expected)
• Documentation update
• Performance improvement
• Code refactoring

Related Issue

None. This is part of the runtime spec 16 release (memo binding and public relay fee claims).

Changes Made

Circuits

• memo_hash = blake2_256(SCALE(memos)) mod r is the new last public input of transfer and unshield (v2). Until now a copier could swap the encrypted memos of a pending spend and make its notes unrecoverable. Transfer hashes both output memos, in order; unshield hashes its change memo, which is empty for a total unshield.
• value_proof removed. Relay fees are claimed publicly since spec 16 (no note, no proof), so circuit id 6 has no caller; no node or client ever proved it. Its source, tests, fixture, manifest entry and docs are gone. Packages up to 0.14.0 keep its artifacts.

Multi-version tooling

The build assumed one version per circuit, so a rotation broke the tree:

• Compile / setup / pack / fixtures now build the manifest's active version, or ROTATE_VERSION for circuits listed in ROTATE_CIRCUIT, under _v{n} names. Before, compiling overwrote the v1 wasm and r1cs and deleted <circuit>_js/.
• pnpm run manifest without rotation keeps every version. It used to rebuild each circuit as a single v1 entry, so the CI regeneration check would have dropped v2.
• ROTATE_CIRCUIT accepts a comma-separated list. Circuits not being rotated keep their published entry.
• transfer / unshield tests compile into build/test-circuits/, so they no longer corrupt the published artifacts.
• Tests read the source version's artifacts, and rotation.test rotates to the version after the highest one in the manifest.

npm package

• getCircuitPaths(circuit, version?) returns the active version by default, together with its version. Before, it always returned the v1 paths.
• The accepted circuits are now "transfer" | "unshield".

Docs

• docs/RELEASE.md: how to build and publish a rotation.
• README, guides, architecture, and circuit docs: updated for v2, with value_proof removed.
• CHANGELOG.md: new 0.15.0 section.

Testing

• All existing tests pass: test:strict 243 passing, 0 failing, 0 pending
• New tests added for new functionality: rotation with several circuits, and regeneration keeping every version
• Manual testing performed:
  • release:restore → v1 canonical; release:verify passes before and after the test suite
  • pnpm run check: lint, typecheck, format, circom lint
  • CI flow simulated (build-all --compile-only → release:verify): v2 compiles deterministically
  • v2 keys embedded in node spec 16 (embed-v2.sh): hashes match the manifest
  • Node e2e against these artifacts: spec 15 → setCode → spec 16 upgrade 35/35, relay suite 114/114, memo clients 32/32

Circuit Changes

Affected Circuits:

• Transfer
• Unshield
• Core components

Constraint Count Impact:

• Transfer: 33,687 → 33,688 (+1, +0.003%)
• Unshield: 16,903 → 16,904 (+1, +0.006%)
• The new constraint is memo_hash_sq <== memo_hash * memo_hash, which keeps the input from being optimized away.

Performance Impact

Benchmarks: not measured. The change is one constraint per circuit, so no measurable change in proving or verification time is expected.

Breaking Changes

• v2 public signals: 8 instead of 7.
  • Provers need @orbinum/proof-generator ^7.2.0.
  • @orbinum/wallet-sdk must bump CIRCUITS_PACKAGE_VERSION to 0.15.0.
• value_proof is no longer in the package. getCircuitPaths("value_proof") throws.
• getCircuitPaths defaults to the active version (v2). Callers that need v1 pass 1.

Checklist

• Code follows the project's style guidelines
• Self-review of code performed
• Comments added for complex logic
• Documentation updated
• No new warnings generated
• Tests added/updated and passing
• Commit messages follow Conventional Commits
• CHANGELOG.md updated

Additional Notes

• Expected CI failures on this PR: Build & Test and Canonical vk_hash fail with a 404 for the v2 keys. CI restores keys from npm latest, and v2 only exists on the release machine until it is published.
• Plan (see docs/RELEASE.md): publish 0.15.0 from this branch, re-run CI, then merge without squashing so the tag v0.15.0 stays on main.
• Clients (proof-generator, protocol, wallet-sdk, app) must be published before the spec 16 setCode.