neuron's clean checkout does not build: it targets mudra APIs not yet on mudra's origin

Status 2026-10-17: resolved. mudra#2 (merged 2026-10-10) put mudra::neuron and mudra::spell on mudra's master. neuron's clean checkout is still red, now for a different reason: see clean-checkout-2026-10-17. The record below is unchanged from 2026-09-24.

Property #39 — every phase-1 component builds and tests from a clean checkout of its default branch against the default branches of its siblings. Lane C · content, filed against neuron because that is where the build fails.

2026-09-24, neuron origin/main at c875411, mudra origin/master at e2d1a63.

the break

node/src/authority.rs calls two functions on a module that does not exist on mudra's origin:

$ git -C neuron show origin/main:node/src/authority.rs | grep -n mudra::neuron
28:        mudra::neuron::sign(&self.key, subject, statement).map_err(|_| Error::Denied)
32:    mudra::neuron::verify_statement(subject, statement, evidence)

node/tests/vault_custody.rs calls the same module. cli/src/custody.rs calls a constant on a second module that does not exist on mudra's origin either:

$ git -C neuron show origin/main:cli/src/custody.rs | grep -n mudra::spell
50:                        .unwrap_or_else(|| mudra::spell::COSMOS_PATH.into()),

mudra's origin/master ships neither module:

$ git -C mudra show origin/master:src/lib.rs | grep 'pub mod'
pub mod claim;
pub mod cosmos;
pub mod domain;
pub mod proof;
pub mod seed;

Confirmed by an actual build: with mudra's path dependency pointed at a worktree of origin/master (e2d1a63) instead of the owner's working tree, cargo check --tests from neuron's workspace root fails before it even reaches node's own authority.rs — vault, pulled in as a neuron workspace path dependency, hits the same missing modules first:

$ RUSTC_BOOTSTRAP=1 cargo check --tests
    Checking cyber-vault v0.1.0 (.../vault)
error[E0433]: cannot find `neuron` in `mudra`
   --> vault/src/operation.rs:270:28
    |
270 |                 if !mudra::neuron::verify_statement(request.subject, request.statement, evidence) {
    |                            ^^^^^^ could not find `neuron` in `mudra`
error[E0433]: cannot find `neuron` in `mudra`
   --> vault/src/operation.rs:301:28
error[E0433]: cannot find `spell` in `mudra`
  --> vault/src/record.rs:72:20
   |
72 |             mudra::spell::derive(words, passphrase).map_err(|_| Error::InvalidInput)?,
   |                    ^^^^^ could not find `spell` in `mudra`
error[E0433]: cannot find `neuron` in `mudra`
  --> vault/src/use_secret.rs:74:35
   |
74 |             let evidence = mudra::neuron::sign(signing, request.subject, request.statement)
   |                                   ^^^^^^ could not find `neuron` in `mudra`
error[E0433]: cannot find `spell` in `mudra`
  --> vault/src/use_secret.rs:98:24
   |
98 |                 mudra::spell::signing_key(&spell, path).map_err(|_| Error::InvalidInput)?;
   |                        ^^^^^ could not find `spell` in `mudra`
error: could not compile `cyber-vault` (lib) due to 5 previous errors

The break is wider than neuron's own two call sites: vault (a workspace path dependency here, and its own repo, cyberia-to/vault) uses mudra::spell::derive/signing_key and mudra::neuron::sign/ verify_statement in its own src/, four more call sites than the two neuron has. cargo check --tests against mudra's real origin stops on vault before it ever reaches node's or cli's own errors — a hard compile error, not a warning, and it blocks the whole workspace.

this is not a stale reference — it is a forward reference

Both modules exist, complete and already shaped to match every call site above, in the owner's local mudra working tree (not committed, not on any mudra branch, not part of any open mudra pull request):

• src/neuron.rs — sign, verify_statement (compat alias sign_statement), an NSIG1-tagged proof format built on claim::sign_arbitrary/claim::neuron_of, matching authority.rs's (subject, statement) -> evidence / (subject, statement, evidence) -> bool shapes exactly.
• src/spell.rs — pub const COSMOS_PATH at the same value origin's src/seed.rs already has (m/44'/118'/0'/0/0); seed.rs itself is dropped from the local tree's pub mod list in lib.rs, i.e. seed was renamed to spell locally, not left standing alongside it.

vault#7 (cyberia-to/vault#7, open) independently names the same gap from vault's side: "rebuild against mudra's spell rename, cover Header/Keys." Two consumers were already written against mudra's next version before that version reached mudra's own origin.

why this is not a neuron-side fix

Reimplementing mudra::neuron::{sign, verify_statement} inside neuron without mudra's neuron module would mean re-deriving the NSIG1 proof format and its binding to claim::neuron_of independently, forking a security-relevant encoding that already exists, reviewed, in the owner's tree. Same for re-deriving COSMOS_PATH under a different name. Both are one git add/commit/push away in mudra, not a code problem in neuron.

remains

Push mudra's local src/neuron.rs and src/spell.rs (and the seed → spell rename in lib.rs) to mudra's origin — an owner action per the isolation rule (uncommitted owner working trees are never staged or pushed by a worker). Once mudra's origin carries both modules, neuron's clean checkout is expected to build with no change on neuron's side; re-run cargo check --tests in both cli/ and node/ to confirm and close this row's neuron half.

This is independent of neuron#1 (row 39, the separate neuron-id crate gap for bbg/tok/mudra) and of mudra#38/#38(2)/#39 (mudra's own cyber-nox/ zheng pin and bbg_root gaps) — none of those touch mudra::neuron or mudra::spell.