Redeem miner rewards

[n13]

Redeem miner rewards

Mobile-app flow for redeeming wormhole rewards (claims aggregated leaf proofs and submits an unsigned verify_aggregated_proof extrinsic to the chain).

Memory optimization

The original implementation peaked at 4.4 GB resident on iOS, which OOM-killed the app on devices without the com.apple.developer.kernel.increased-memory-limit entitlement (≈1.4 GB jetsam ceiling). After matching the wormhole-memprof configuration exactly, peak is now 2.3 GB resident (~50% reduction) and phys_footprint stays under 300 MB throughout the run.

Root causes fixed

1. Wrong task_vm_info struct layout. Earlier instrumentation read min_address as phys_footprint on ARM64 iOS, reporting a phantom constant ~4 GB and sending us chasing the wrong things. Fixed by matching Apple's kernel layout (integer_t = i32 for region_count / page_size) with a compile-time size_of assert at 152 bytes.
2. Mobile aggregator was loading aggregated_prover.bin (972 MB) from disk and then rebuilding the aggregation circuit anyway just to extract targets. Switched to the same fresh-build constructor wormhole-memprof uses (Layer0AggregationProver::new) via aggregate_proofs_fresh. This skips both the disk-load peak and the redundant target-rebuild.
3. Leaf circuit was bigger than the aggregator. wormhole_leaf_circuit_config() was returning standard_recursion_config() (143 wires / 80 routed), so the aggregator had to recursively verify a larger leaf than needed. Aligned the leaf with the aggregator's wire counts (135 / 60) — the same shape wormhole-memprof's leaf_config_matching uses to hit its 2.5 GB benchmark.
4. Rayon pool size. Single-threaded plonky2 (provingThreads: 1) allocated bigger contiguous FFT buffers than the chunked parallel path. Set to 8, matching the wormhole-memprof sweep's measured sweet spot.
5. Stale circuit binaries. Bumped the on-device cache directory circuits → circuits_v3 so the new 135/60 leaf bins regenerate on first run instead of failing common-data checks against old 143/80 bins.

Diagnostic infrastructure added

• Accurate per-phase memory checkpoints in wormhole.rs ([mem] <tag> phys=… resident=… compressed=… internal=… external=… virt=…) with a detailed dump variant.
• Phase markers inside the locally-patched Layer0Aggregator::aggregate() (build_prover / commit / prove) and Layer0AggregationProver::new_from_bytes so each substep is timed individually.
• Memory checkpoints around every claim-service and UTXO-service phase ([ClaimMem], [UtxoMem]).

ZK parameters (matches wormhole-memprof exactly)

Aggregator circuit (wormhole_aggregator_circuit_config())

Parameter	Value
zk_config	RowBlinding (leaf_hiding: true)
num_wires	135
num_routed_wires	60
num_constants	2
use_base_arithmetic_gate	true
security_bits	100
num_challenges	2
max_quotient_degree_factor	8
fri_config.rate_bits	3
fri_config.cap_height	4
fri_config.proof_of_work_bits	16
fri_config.num_query_rounds	28
fri_config.reduction_strategy	ConstantArityBits(4, 5)

Leaf circuit (wormhole_leaf_circuit_config(), updated this PR)

Parameter	Value
zk_config	Disabled (leaf_hiding: false)
num_wires	135 (was 143)
num_routed_wires	60 (was 80)
num_constants	2
use_base_arithmetic_gate	true
security_bits	100
num_challenges	2
max_quotient_degree_factor	8
fri_config.rate_bits	3
fri_config.cap_height	4
fri_config.proof_of_work_bits	16
fri_config.num_query_rounds	28
fri_config.reduction_strategy	ConstantArityBits(4, 5)

Runtime knobs

Knob	Value
maxProofsPerBatch	16
proofConcurrency	1
provingThreads (rayon global pool)	8
freshBuild	true (use aggregate_proofs_fresh)
Plonky2 hash	Poseidon (Goldilocks, D=2)
Plonky2 config type	PoseidonGoldilocksConfig
Plonky2 field	GoldilocksField

Memory before / after

Phase	Before (resident)	After (resident)
baseline / idle	~20 MB	~20 MB
16 leaf proofs	~1.7 GB	~700 MB
aggregation prove peak	4.4 GB (OOM on device w/o entitlement)	2.3 GB
post-claim	~3.9 GB (uncollected)	~400 MB

phys_footprint (the metric iOS Jetsam evaluates) stays under 300 MB throughout — the resident_size peak above includes allocator-reusable pages that don't count against Jetsam.

Files changed

• quantus_sdk/rust/src/api/wormhole.rs — fresh-build aggregator path, accurate task_vm_info reading, per-phase log_mem checkpoints.
• quantus_sdk/lib/src/services/wormhole_claim_service.dart — claim orchestration with memory checkpoints between every phase.
• quantus_sdk/lib/src/services/wormhole_utxo_service.dart — unspent-transfer discovery with memory checkpoints.
• quantus_sdk/lib/src/services/circuit_manager.dart — cache dir bumped to circuits_v3.
• mobile-app/lib/v2/screens/settings/redeem_progress_screen.dart — maxProofsPerBatch: 16, provingThreads: 8, freshBuild: true.
• mobile-app/ios/Runner/Runner.entitlements — extended-virtual-addressing kept; increased-memory-limit removed (no longer required at the current peak; can be re-added if a device still hits jetsam).