docs(plonky3): solution-space research — NO-GO overturned (Probe Q), 9 paths

[TaprootFreak]

Summary

The Phase-0 gate (#212) recorded NO-GO ("no per-instance value channel across a batch-recursion layer"). This branch overturns that and exhaustively explores the solution space.

🟢 The NO-GO was scoped too narrowly. probe_q_custom_public_value empirically proves the channel EXISTS: a custom AIR with num_public_values()=1 proved with prove_batch and verified in-circuit via verify_batch_circuit surfaces its public value as a non-empty air_public_target (NOT [0,0,0]) and binds it soundly across the batch layer (correct value accepted, wrong rejected). This rides on upstream PR #407 "feat: support public values" (merged 2026-03-19, already in our pinned rev 524665d). My [0,0,0] finding held only for the PRIMITIVE tables / CircuitBuilder public inputs — not for a public-value-emitting AIR.

Deliverables

• MIGRATION_PLONKY3_SOLUTIONS_RESEARCH.md — all 9 paths enumerated + assessed with links/repo-pointers (real web + GitHub research via subagents).
• probe_q_custom_public_value (18th probe) — the empirical proof that overturns the gate.

The 9 paths (ranked)

#	Path	Verdict
1+5	Plonky3 + custom public-value-emitting tables	✅ channel proven (Probe Q); IVC chaining = public-API construction (~400-650 LOC, no fork)
3	Folding / Sonobe (Nova/CycleFold)	✅ native IVC (z_i→z_{i+1}); strong alternative; ≤5s latency unproven
2	Self-authored upstream PR (atop #407)	✅ plausible additive
7	RISC Zero / OpenVM (zkVM)	✅ shipped, heavyweight, GPU-leaning
4	Hybrid Plonky2/Plonky3	⚠️ low-value
6	Protocol redesign (off-circuit continuity via trusted node, §7.22 posture)	⚠️ cheapest, reduces soundness scope
8	Fork (excluded §16)	⚠️ surfaced
9	Stwo/Triton-VM/Halo2-accumulation	⚠️ paradigm rewrites

Recommendation

Pursue Path 1+5 (keeps the chosen stack; channel empirically proven) behind a carrier-table IVC-chain spike (Probe R, next), with Sonobe benchmarked in parallel (Path 3) as the cleanest native-IVC fallback, and Path 6 (off-circuit continuity) as the cheapest redesign. Operator chooses. Plonky2 stays in production until a path clears a latency gate.

Probes D/G/H/J remain valid — they correctly bound the high-level-API / stock-table behavior; Probe Q identifies the supported construction they didn't test.

Notes

• Builds on #212 (the spike). Spike still excluded from node/shared CI; fmt + clippy -D warnings clean.
• KEY correction to the gate — the gate memo will be updated to reflect GO-via-Probe-Q once the carrier-table chain (Probe R) confirms end-to-end IVC.