Summary

L-TRI-3 Week 2: wires the V2 GF16 matmul module (landed Week 1, cf951b2c) into the live entry points.

• POST /prove (cli/tri): accepts V1 (4-byte) and V2 (32-byte SHA256) responses; dispatches on new version: u8 field (default 1).
• Solana program (contrib/solana/programs/tri-mining): new submit_proof_v2 instruction performs the full V2 verification fully on-chain (derive 16×16 challenge → GF(2⁴) matmul against CHAMPION_WEIGHTS → SHA256 packed product → constant-time compare).

Both changes are purely additive. V1 clients keep working without code change; existing IDL methods untouched.

V2 spec (recap)

• GF(2⁴) reduction polynomial: x⁴ + x + 1 (0x3)
• Challenge: SHA256("TRI_PHI_CHALLENGE_V2" || epoch_le8 || node_id || row_byte) × 16 rows; high nibble of bytes 0..32 → 16×16 matrix
• Response: SHA256(pack_gf16_matrix(CHAMPION_WEIGHTS × challenge)) — 32 bytes (2^256 security; random-guess prob ≤ 2^−64 after birthday)
• Verification: constant-time XOR-fold compare

Test results

Cross-impl parity verified between cli/tri and tri-mining (Solana program) via:

• test_onchain_v2_champion_first_row_matches_cli_tri — first row of CHAMPION_WEIGHTS
• test_onchain_v2_gf16_mul_matches_cli_tri — gf16_mul(0x3,0x7)=0x9, gf16_mul(0xF,0xF)=0xA, gf16_mul(0x0,0xA)=0x0, gf16_mul(0x1,0xB)=0xB

cargo check for the Solana program: ✅ exit 0 (only cfg(feature="anchor-debug") warnings from the macro, no errors). cargo build-sbf not run — toolchain not present in sandbox. anchor test not run for the same reason; the on-chain GF(2⁴) matmul + SHA256 path is covered by host-side Rust unit tests.

Files changed

cli/tri/src/depin/prove.rs                    +168  -81   (V2 dispatch + 5 tests)
cli/tri/src/depin/types.rs                    +6    -0    (version field)
contrib/solana/programs/tri-mining/src/lib.rs +215  -0    (submit_proof_v2 + 6 tests)
contrib/solana/tests/tri-mining.ts            +174  -0    (V2 TS test helpers + 2 tests)

Compliance checklist

• L1 Traceability — both commits reference Closes #607
• L7 Unity — Rust + TS only, no new .sh or .py
• No unsafe outside rusb (none added)
• No production unwrap() in handler — all conversions via match/? patterns, helper err() for early returns
• No commit-msg --no-verify
• Out of scope unchanged: cli/dlc10/, fpga/, .github/workflows/vivado-build.yml — untouched
• NOW.md — not touched (no real reason to bump)

Test plan

• CI runs cargo test -p tri → expect 30/30
• CI runs cargo test -p tri-mining --lib → expect 7/7
• Optional follow-up: run anchor test on a machine with solana-test-validator + Anchor CLI installed
• Optional follow-up: cargo build-sbf (or anchor build) to produce the deployable program and confirm CU budget is sufficient for full 16×16 GF(2⁴) matmul (~4096 gf16_mul calls per instruction)

Notes for the user

• Sandbox has no Solana toolchain (cargo-build-sbf not installed), so the on-chain BPF build wasn't validated. Native cargo check and host-side Rust unit tests pass; the only thing left to verify is CU budget when deployed.
• All three commits cited in the prior session (643892c6, 0fa1cb7f, cf951b2c) are on origin/main; this PR targets main per the user's Option-A decision.

🤖 Generated with Claude Code