Feat/faster qr codes

[n13]

Summary

• Speed up animated UR transfer by exposing configurable fragment sizes through the Rust/Flutter bridge, precomputing cold-wallet QR painters, and allowing both wallet scanners to process frames without the default detection throttle.
• Add cold-wallet settings for signature QR frame rate and bytes per frame, move wallet reset into Settings, and add an explicitly confirmed online-lock override for release-device testing with a persistent warning banner.
• Reduce Keystone send latency by prefetching and caching unsigned signing payloads and UR frames while the user reviews the transfer.
• Reuse HTTP JSON-RPC connections, cache genesis/runtime metadata, and share the signing-context fetch path across local and externally signed transactions.
• Centralize the transaction mortal-era period, update quantus_ur to 1.6.0, regenerate Flutter Rust Bridge bindings, and bump the mobile app to 1.5.10+124 and cold wallet to 1.1.0+3.

Validation

• GitHub Quantus Apps CI passes formatting, analysis, Mobile App tests, and Quantus SDK tests.
• cargo test --locked passes all 14 native SDK tests, including configurable UR fragment round trips.
• Repository-wide Dart formatting check reports zero changed files.
• git diff --check passes for the complete PR.

[n13]

REQUEST_CHANGES (opinion): the performance work builds and CI passes, but production air-gap, QR-capacity, and Keystone payload-lifetime blockers remain on 9ced530789f77f602cc0ec2c101de1f78e352f85.

Findings:

1. [P1] Keep the network-lock override out of production cold-wallet builds (cold-wallet-app/lib/screens/settings_screen.dart:42-63, cold-wallet-app/lib/widgets/connectivity_guard.dart:38-46). The release UI now lets any user persistently enable an Override button and then dismiss the guard while the signer is online. That removes the air-gap enforcement which is the cold wallet's core security boundary; a warning and banner do not restore it. Debug builds already bypass the guard. Compile this control only into an explicit non-production test flavor/build flag, and keep release builds fail-closed.

2. [P1] Bound the fragment control by the encoded QR capacity, not raw fragment bytes (cold-wallet-app/lib/providers/settings_providers.dart:10-12, cold-wallet-app/lib/components/animated_ur_qr.dart:37-46). The permitted 1,500-byte setting encodes the real 7,219-byte ML-DSA signature-plus-public-key payload into five UR frames of 2,954 characters each. Version-40 QR at error-correction L accepts 2,953 characters; constructing the frame throws QrInputTooLongException: 23652 > 23648, so selecting the advertised maximum breaks the signature screen after the transaction has been signed. Lower the bound or dynamically validate the final UR strings against QrCode capacity, and add a regression test using the real payload size.

3. [P1] Do not carry a nearly expired 16-block payload through the hardware-signing screens (quantus_sdk/lib/src/constants/app_constants.dart:53-58, mobile-app/lib/v2/screens/send/keystone_sign_cache.dart:9-21, mobile-app/lib/v2/screens/send/keystone_sign_screen.dart:47-75). The cache accepts an entry until two blocks before expiry, but KeystoneSignScreen copies the payload/frames once and never checks their age again; the verify and signature-scan screens continue using that same value. With the new 16-block era, a cache hit can enter the QR screen with only about 24 seconds left, and the review-screen timer's later refresh only replaces provider state, not the QR already displayed. A normal scan/review/sign round trip can therefore submit an expired transaction. Revalidate before advancing/submitting and regenerate the displayed payload when needed, or retain a sufficiently long era while adding an explicit expiry guard.

4. [P2] Refresh runtime version when constructing a signing payload (quantus_sdk/lib/src/services/substrate_service.dart:33-49). A five-minute process-wide cache can straddle a runtime upgrade. During that window both local and Keystone payloads declare the old spec/transaction version, the cold signer can miss the intended spec-drift warning, and the chain rejects the resulting extrinsic until the cache expires or the app restarts. Keep connection/genesis reuse, but fetch the runtime version for each new signing context or invalidate it on an authoritative runtime-change signal and retry stale-version failures.

Validation:

• GitHub Analyze workflow — passed, including formatting, analysis, Mobile App tests, and Quantus SDK tests.
• cargo test --locked — passed (14 tests, including configurable UR round trips).
• Repository-pinned Dart formatting check — passed with zero changed files.
• git diff --check d5fc4d6572ebc2244e95009a33f5da5048935867...9ced530789f77f602cc0ec2c101de1f78e352f85 — passed.
• Targeted UR/QR probe — reproduced 2,954-character frames at the 1,500-byte setting and the QR library's one-character overflow exception.

[n13]

Addressed the review findings in d4d69d1:

[P1] Network-lock override in release builds — kept intentionally. The cold wallet is not yet distributed through app stores, and testing release builds on real devices without the override means toggling radios for every flow. The control stays fail-closed by default: the Override button only exists behind a persistent settings opt-in now labelled "debugging only / not safe", an override lasts a single session, and a red banner stays on screen while active. Code comments in settings_screen.dart and connectivity_guard.dart document this as intentional; revisit when the app ships to app-store users.

[P1] Fragment bound vs QR capacity — fixed. New encodeUrForQr in quantus_sdk measures the final UR strings against the 2,953-char version-40/EC-L byte-mode capacity and shrinks the fragment size until every frame fits; the cold wallet signature view now uses it, and the slider max dropped to 1,400 bytes. Regression test ur_qr_frame_test.dart sweeps every fragment setting (300–1,500) with the real 7,219-byte ML-DSA payload, builds the QR for the longest frame, and round-trips the decode. Verified the old bug reproduces: raw encodeUr at 1,500 yields 5×2,954-char frames (1 char over); the fix re-splits to 6×2,472.

[P1] Nearly expired era through the signing screens — fixed, conservatively. Cached payloads now expire once less than a 10-block round-trip reserve (~120 s at 12 s blocks) of era lifetime remains, the sign screen re-checks every 5 s and regenerates the displayed QR when the reserve is consumed, advancing to verify is refused with a stale payload (QR regenerates instead), and submission checks the live block height against the era window (2-block margin) and fails with a dedicated "transaction expired" message instead of broadcasting.

[P2] Runtime version cache vs runtime upgrade — minimal fix as discussed: when the chain rejects a submitted extrinsic, the cached genesis hash and runtime version are dropped, so the next payload build re-fetches chain state. Connection/genesis reuse on the happy path is unchanged.

[n13]

REQUEST_CHANGES (opinion): d4d69d1ae3c90e5c3a74539af01b6655511b1a19 addresses the QR-capacity, Keystone-era, and runtime-cache findings, but the PR is not mergeable with current main, so there is no tested merged result yet.

Blocking finding:

1. [P1] Rebase onto current main without undoing the merged crypto/artifact upgrade (quantus_sdk/rust/Cargo.toml:11-31, quantus_sdk/rust/Cargo.lock, quantus_sdk/rust/src/api/ur.rs:3-25). Current main is c1f980b0464be359eae9d2f8d66b91b51666cf40 (including #598), while this branch still forks from d5fc4d6572ebc2244e95009a33f5da5048935867. GitHub reports CONFLICTING, and git merge-tree shows unresolved conflicts in all three files. In particular, the PR side of Cargo.toml still has the pre-#598 qp-rusty-crystals 3.0.1 / wormhole 3.1.0 set, whereas main has crystals 4.1.0 / wormhole 4.2.0 plus the versioned prover-artifact contract. Resolve by retaining main's dependency and artifact changes, preserving this PR's configurable encode_ur / encode_bytes_with_options API, regenerating Cargo.lock, and rerunning the native QR regression and full CI on that resolved head. The current head has no GitHub checks, so taking either conflict side without that validation could silently regress #598 or lose the QR fix.

Prior findings re-evaluated:

• QR capacity — fixed on this head. encodeUrForQr measures final UR strings against the 2,953-character limit; the native regression sweeps 300-1,500-byte settings with the 7,219-byte signature payload, constructs the longest QR, and round-trips successfully.
• Keystone era lifetime — fixed on this head. Step 1 refreshes stale payloads and refuses stale advancement, and submission checks the live chain height with a two-block margin and a dedicated expiry error.
• Runtime-version cache — fixed sufficiently for retry behavior. A rejected extrinsic clears the cached runtime/genesis context, so the next payload rebuild refetches chain state.
• Release network override — explicit temporary product/security exception. Given the stated pre-app-store testing scope, I am not treating the deliberately gated, session-only override as an additional code defect in this verdict. It should be removed or build-flag-gated before production distribution.

Validation on the isolated PR head:

• cargo test --locked — passed, 14 tests.
• flutter test --tags native test/ur_qr_frame_test.dart — passed; this regression is excluded by repository CI.
• flutter test test/unit/keystone_sign_cache_test.dart — passed, 12 tests.
• flutter analyze . — clean for quantus_sdk, mobile-app, and cold-wallet-app (each under the repository's 20-second limit).
• Repository-pinned Dart formatting check — zero changed files.
• git diff --check — passed against both the original base and current main.
• GitHub checks for d4d69d1a — none reported; the PR is currently conflicting.

[n13]

Merged current main (c1f980b, including #598) into this branch in 65759a7, resolving the blocking finding:

• Cargo.toml — main's dependency set retained (qp-rusty-crystals 4.1.0, wormhole 4.2.0, qp-poseidon-core 3.1.0, qp-plonky2 1.5.5) alongside this branch's quantus_ur 1.6.0; Cargo.lock auto-merge verified consistent via cargo test --locked.
• ur.rs — kept this branch's configurable encode_ur(data, max_fragment_length) / encode_bytes_with_options API; the conflicts were rustfmt formatting from main's side, reapplied with cargo fmt.
• FRB bindings regenerated on the merged tree (flutter_rust_bridge_codegen 2.12.0) rather than trusting the textual auto-merge — which had in fact silently dropped chore: bump poseidon 3.1.0, plonky2 1.5.5, crystals 4.1.0 #598's zkCircuitsVersion binding and left the Dart/Rust rustContentHash values inconsistent (would have failed at RustLib.init). Regeneration restored the full union of both APIs.

Validation on the merged head:

• cargo test --locked — 40 tests passed (rebuilt release dylib).
• flutter test — quantus_sdk 176 (native included, QR frame regression rerun explicitly), mobile-app 287, cold-wallet-app 7, all passing.
• flutter analyze . — clean on all three packages.
• dart format --line-length=120 --set-exit-if-changed — clean; git diff --check — clean.

[n13]

APPROVE (opinion): no blocking findings remain on 65759a798373c2a04d2888d5bad0b4eb2b59bf70.

The prior merge blocker is resolved correctly:

• The head has d4d69d1a and current main (c1f980b0, including #598) as its two parents; GitHub now reports MERGEABLE and the Analyze workflow passes.
• quantus_sdk/rust/Cargo.toml, Cargo.lock, the wormhole artifact implementation, and circuit_manager.dart match main byte-for-byte, preserving the crystals 4.1.0 / wormhole 4.2.0 dependency set and versioned artifact cleanup.
• The configurable encode_ur / encode_bytes_with_options API is preserved. Regenerated FRB bindings include both that API and #598's zkCircuitsVersion; Dart and Rust content hashes both equal 300623511, and RustLib.init succeeds against a freshly rebuilt release dylib.

The earlier QR-capacity, Keystone-era, and runtime-cache findings remain fixed. The release network-lock override remains an explicit temporary pre-app-store testing exception; it should still be removed or build-flag-gated before production distribution, but it is not a blocker for the stated scope.

Validation:

• GitHub Analyze workflow — passed on the reviewed head.
• cargo test --locked — passed, 40 tests on the merged 4.1/4.2 dependency set.
• ./test/build_library.sh followed by flutter test --tags native — passed, 15 tests; this verifies the regenerated bridge and the 7,219-byte QR-capacity regression against the rebuilt dylib.
• flutter test test/unit/keystone_sign_cache_test.dart — passed, 12 tests.
• flutter analyze . — clean for quantus_sdk, mobile-app, and cold-wallet-app, each within the repository's 20-second limit.
• Repository-pinned Dart formatting check and git diff --check origin/main...HEAD — passed.

No blocking findings.

[dewabisma]

LGTM!