Multi-peer integration tests for keep-frost-net ECDH coordination (round-2 mutation kills)

[kwsantiago]

Follow-up to #417's round 2 (PR landing now). Mirrors the shape of #541 for the signing-module work.

What round 2 shipped

Unit tests on `keep-frost-net/src/ecdh.rs` killed 47 of 53 mutations that survived the baseline:

Function	Before	After
Pure helpers (derive_ecdh_session_id, compute_partial_ecdh, lagrange_coefficient, aggregate_ecdh_shares)	9 surviving	1 surviving
EcdhSession state machine + accessors	29 surviving	2 surviving
EcdhSessionManager lifecycle	15 surviving	0 surviving

Final: 63/110 caught (57%), up from 16/110 (15%).

What's still surviving in keep-frost-net/src/node/ecdh.rs (20 mutations)

All on the async coordination handlers, same shape as #541's signing-module survivors. They need multi-peer integration test infrastructure to kill.

Function	Surviving mutations	What that means
handle_ecdh_request	5	Responder gates: replay window check, peer-policy, participant-membership; multiple delete ! and ==/!= mutations on early-return predicates.
request_ecdh	10	Requester orchestration: match arms on EcdhComplete / EcdhFailed / Err, terminator predicates, the has_all_shares match guard.
handle_ecdh_complete	3	Final-secret delivery: subscriber notification + session-id equality check.
handle_ecdh_share	2	Share ingestion: full function "replace with Ok(())" and the share-index match.

Remaining 6 in keep-frost-net/src/ecdh.rs

Triaged but deliberately not chased:

• aggregate_ecdh_shares line 96 (+= ↔ -=): equivalent on the single-party case because IDENTITY + p = IDENTITY - p produces points with the same x-only output. Distinguishing requires multi-party, same scope as the node/ecdh.rs survivors.
• EcdhSession::state line 169 (> ↔ >=): off-by-one at the exact timeout boundary. Hitting it deterministically requires sub-nanosecond timing control.
• EcdhSession::try_complete line 222 (- ↔ /): byte-index arithmetic on the Identifier serialization buffer. Both len-2=30 and len/2=16 index into a 32-byte buffer; for small share indices the high bytes are 0 in both positions.
• shared_secret() accessor: 3 mutations on the return shape. The new `ecdh_session_shared_secret_returns_stored_value_only_after_completion` test kills the None and constant-leak mutations on the post-completion path; the pre-completion-None path is exercised by the same test.

Proposed approach

Same as #541 — an in-process multi-peer simulator that drives ECDH sessions without a real relay, covering:

• 2-of-3 happy path (kills the "replace Result<()> with Ok(())" mutations because the absence of a real send on any handler breaks the round).
• Replay window edge: a request just outside the window must be refused (kills the ==/!= mutations in handle_ecdh_request).
• Failed completion → subscriber notification (kills the handle_ecdh_complete mutations).
• Re-run `cargo mutants -p keep-frost-net --file keep-frost-net/src/node/ecdh.rs` and confirm survival rate drops materially.

Related

• Run cargo-mutants on security-critical modules #417 (parent).
• Multi-peer integration tests covering keep-frost-net signing/coordination decisions (round-2 mutation kills) #541 (signing-module sibling).
• Tracked under: Broader keep CLI testing tracker (beyond WDC) #434 / rotate-password, rotate-data-key, frost refresh security-critical mutations #438 broader security testing.