[illuzen]

Fixes the "entered my phrase, got a wrong blank wallet" report.

The import defaults changed from ML-DSA-87 at …/0' (v1.x) to ML-DSA-65 at …/1' (v2.0+), so a phrase from an older wallet imports "successfully" onto a brand-new empty address with no indication anything is off. A user recovering funds cannot tell this apart from a bad phrase.

Changes

• wallet import --scan: derives every historical default (ML-DSA-65 …/1', ML-DSA-87 …/0', master seed under each scheme), shows address and checkphrase for each, marks on-chain history when a node answers, and lets the user pick. Prints the explicit --scheme/--derivation-path flags to reproduce the choice. q, EOF, or bad input abort without touching the keystore; non-TTY stdin fails fast instead of blocking. Conflicts with --scheme/--derivation-path/--no-derivation at parse time.
• Post-import history check: after a plain import, warn when the derived address has never been used on-chain and point at --scan / airdrop check.
• Node never required: the probe is bounded by a 5s timeout; a missing node only drops the on-chain info. --offline skips the probe entirely for air-gapped machines.
• Mnemonic prompt reworded to "recovery phrase (… not the wallet password)"; v1.x defaults documented in --scheme help and a new README section.

Testing

• 98 wallet tests pass (5 new, including pinning the ML-DSA-87 candidate to the known qzm5QCox… vector and asserting all four candidates are distinct).
• cargo clippy --workspace --all-targets --all-features --locked -- -D warnings clean; formatted with CI's pinned nightly-2026-08-31.
• Picker exercised end-to-end through a pty: choosing 2 imports the pre-v2.0 ML-DSA-87 address; EOF and out-of-range input abort.

[n13]

Reviewer model: GPT-6.1 Sol

REQUEST_CHANGES — two recovery paths can still block unexpectedly. Reviewed head 92d3e5ffcdae805c518471783d6651374f23ac34 against base 413a6e23f7fc9ebbe8145d9cd887cd52a70cc933.

1. [P2] Bound the complete optional node probe, including RPC reads. In src/cli/wallet.rs:547–565, only QuantusClient::new is inside the five-second timeout. chain_token opens another connection, and the four account lookups run sequentially outside that deadline. The same issue affects the plain-import check at lines 494–504. With a mock node that immediately completed initialization but withheld state_getStorage, plain import took 31.5 seconds; separately withholding token properties or storage left --scan without any addresses or selection prompt after eight seconds. Account requests retain the client's 30-second timeout, so multiple failures accumulate. Apply one overall deadline to all best-effort network work in both paths and show the locally derived candidates when it expires. Add coverage for a node that connects successfully and then stops answering.

2. [P2] Reject non-TTY scans before collecting passwords or mnemonics. The terminal guard at src/cli/wallet.rs:537–543 is reached only after the import handler has run the password and mnemonic prompts. rpassword reads the controlling terminal, so wallet import --scan --offline --mnemonic-file <protected-file> </dev/null still waits for a password when no password source is supplied. I reproduced this with redirected stdin and a controlling PTY; it remained at the password prompt until killed. Move the scan terminal check to the start of import handling, before secret-file reads and prompting, and test without --allow-empty-password or an environment password. The current non-TTY test bypasses the earlier prompts.

Validation:

• SKIP_CIRCUIT_BUILD=1 cargo test --locked --no-default-features --lib wallet: 108 passed.
• SKIP_CIRCUIT_BUILD=1 cargo clippy --locked --all-targets --all-features -- -D warnings: passed.
• CI-pinned cargo +nightly-2026-08-31 fmt --all -- --check and git diff --check: passed; headless CLI build passed.
• PTY checks confirmed legacy candidate 2 imports the pinned address, and cancellation/invalid selection create no wallet.
• GitHub build/test, analysis, and example jobs were still running at the last check. The security-audit job failed against the unchanged dependency lockfile; this is separate from the two PR findings above.

[illuzen]

Addressed both findings in bb53b2c.

Node probe wall-clock bound. Connect, the System::Account reads and the token lookup now run inside a single probe_on_chain with one 10s tokio::time::timeout. Partial results are kept: if the node connects but stalls, --scan still lists every candidate (marked unknown (no answer)) and the plain import prints a "node stopped answering" note instead of blocking. Against a mock node that accepts the connection and never answers state_getStorage, plain wallet import now finishes in ~11s (was ~31.5s). Tests: probe_on_chain_is_bounded_when_node_connects_then_stalls, wallet_import_completes_promptly_when_node_connects_then_stalls, plus healthy and unreachable cases.

--scan TTY check. ensure_scan_is_interactive() now runs at the top of the Import arm, before the password ceremony and before the recovery phrase is read. Test: wallet_import_scan_without_terminal_fails_before_any_prompt (nonexistent mnemonic file, no password env; asserts the TTY error is what comes back and no wallet is created).

The mock node in chain::client moved to a test_support module with a stall_storage flag so both test modules share it.

[n13]

Reviewer model: GPT-6.1 Sol

APPROVE — the previous blocking findings are resolved; no blocking findings remain at head bb53b2c9a89188aa47d526c5aa5efbe13c116689 against base 413a6e23f7fc9ebbe8145d9cd887cd52a70cc933.

Both plain import and scan now share one deadline covering connection, account reads, and token lookup, preserving partial results when a node stalls. The scan terminal guard runs before secret-file reads or password/mnemonic prompts. Historical candidates reuse the existing derivation and import paths.

Non-blocking feedback:

• [P3] Describe current state rather than claiming complete history. The plain-import warning and scan label infer “never received funds or sent a transaction” from the current System::Account balance/nonce. A previously used account that was reaped also returns zero/default state. Prefer wording such as “No current balance or nonce found” in both paths and the corresponding README text; otherwise a user can be told their correct historical address was never used. Address selection and import still work.

Validation:

• SKIP_CIRCUIT_BUILD=1 cargo test --locked --no-default-features --lib wallet: 112 passed, including healthy/unreachable/stalled-node probes, bounded plain import, early non-TTY rejection, and historical derivation/import vectors.
• Same test command with chain::client::tests: 6 passed.
• SKIP_CIRCUIT_BUILD=1 cargo clippy --locked --all-targets --all-features -- -D warnings: passed.
• Headless CLI build, CI-pinned cargo +nightly-2026-08-31 fmt --all -- --check, Taplo, and git diff --check: passed. Circuit generation was skipped for these wallet-focused checks.
• Fresh PTY checks passed for default/legacy selection, cancellation, invalid input, and EOF; aborted selections wrote no wallet. Redirected stdin with a controlling terminal failed immediately before a password prompt or mnemonic-file access.

GitHub's Ubuntu build/tests, format, dependency cooldown, and Clippy/doc jobs passed. macOS build/tests and examples were still running at the final check. The security-audit job failed with 13 vulnerabilities in the existing dependency set; Cargo.toml, Cargo.lock, and the CI workflow are unchanged by this PR.

[illuzen]

Took the P3 in $(git rev-parse --short HEAD): the plain-import warning now reads "No current balance or nonce found for this address." (hint softened to "funds may be under a different derivation"), the --scan label is "no current balance or nonce", and the --scan help text and README section describe current balance/nonce rather than history. No behaviour change.

[n13]

Reviewer model: GPT-6.1 Sol

APPROVE — no blocking findings remain at head 33b25d34c81bb0cab263cb3fd5f9e3d061272d7a against base 413a6e23f7fc9ebbe8145d9cd887cd52a70cc933.

The latest commit resolves the remaining wording feedback in both import paths and the README: empty current account state is no longer presented as proof that an address was never used. The complete diff also preserves the earlier fixes: one overall 10-second deadline covers the optional node probe, and non-TTY scans fail before password prompts or mnemonic-file reads. Historical candidates reuse the existing derivation and import functions.

Validation at this head:

• SKIP_CIRCUIT_BUILD=1 cargo test --locked --no-default-features --lib wallet: 112 passed, including historical vectors, healthy/unreachable/stalled nodes, bounded plain import, and early non-TTY rejection.
• The same test command with chain::client::tests: 6 passed.
• SKIP_CIRCUIT_BUILD=1 cargo clippy --locked --all-targets --all-features -- -D warnings: passed; headless CLI build passed.
• CI-pinned cargo +nightly-2026-08-31 fmt --all -- --check, Taplo, and git diff --check: passed.
• Fresh PTY checks passed for all four selections, cancellation, invalid input, and EOF. Each successful import matched its displayed address; aborted selections wrote no wallet. Redirected stdin with a controlling terminal and no password source failed immediately before secret access.

Circuit generation was skipped for these wallet-focused local checks. GitHub format, Clippy/doc, and security-audit jobs passed; Ubuntu/macOS build/tests and examples were still running at the final check. Both dependency-cooldown jobs failed because their merge-result checkout includes dependency updates already on main, compared against this PR's older base. This PR changes neither Cargo.toml nor Cargo.lock; see the cooldown log.