Add event-sourced many-market wind tunnel

[krandder]

What changed

• freezes v1 P0 and adds a v2 registrar/gateway/Space/arbitration/evaluator/relay/FLM event-to-state schema
• hydrates fixed contract getters and transition simulations at the indexed finalized block
• reconstructs treasury payloads only from sealed EconGateway events and site payload bytes only from the sealed Space event/release strategy
• exposes a committed-state next-action command; callers never provide startEvaluation payload bytes
• persists staticcall, external send, receipt, funding refusal, and linked race outcomes in SQLite evidence
• classifies a loser as a benign race only after one identical action landed and its postcondition holds
• adds an unlocked-account Anvil race and capped ten-instance pre-broadcast drill
• adds a fresh-state, loopback-only Anvil gate that stages ten real registrar receipts, deploys their full pinned economic core/evaluator and FLM, and reconciles one active evaluation per FAO
• binds every deployed module to the committed exact-target economic/FLM blobs, schema-v4 manifests, generated hashes, compiler evidence, and FLM submodule SHA before any send

Why

Lane 6 needs reproducible autonomous operation across independent FAOs without weakening the one-active-evaluation/one-conditional-market invariant. P1a closes the gap between log replay and safe keeper action construction. P1b proves that the same index/hydration path converges over ten full independent economic deployments while signing and any non-loopback broadcast remain outside the package.

Ancestry

• rebased onto final Lane 5 head 040ca17 (codex/lane5-agent-work)
• inherits Lane 5 schema-v4 economic manifests/runtime pins, buyback treasury path, and 17 passing reference-agent executions covering all 16 numbered drills
• P0: 358442e
• P1a: 0d7ad1d
• P1b: d878033
• P1a/P1b patches are unchanged from their previously green commits; the P0 README context alone follows the final Lane 5 prose

Safety boundary

• no production economic bytecode, daemon, runtime dependency, live-chain send, or stored key is added
• the P1b Solidity driver hard-reverts off chain 31337; its Python caller constructs only a 127.0.0.1 RPC, requires a fresh zero-nonce Anvil, and uses disposable unlocked accounts
• committed exact-target blobs—not broad-script type(...).creationCode metadata—are the deployment trust boundary; Python checks all blobs before send and receipts re-check core/FLM hashes on-chain
• failed partial local runs write explicit success: false artifacts; success is written only after stable replay and all ten postconditions
• each FAO has exactly one EVALUATING proposal, no queued second proposal, no evaluator market yet, and an independent spot-mode FLM
• this is local controlled A4 engineering evidence, not a live deployment, demand, subsidy viability, or information aggregation claim

Validation after rebase

• Python 3.9 full discovery: 93/93
• Python 3.14 full discovery: 93/93
• Forge: 207 passed, 0 failed, 0 skipped
• forge fmt --check, git diff --check, and forge build --sizes: green
• economic-core schema-v4 code-hash evidence: current
• operational FLM compiler/code-hash evidence: current and prepared on the clean commit
• AgentWorkIndex code-hash evidence: current
• Lane 5 reference-agent drill: 17 passing executions covering all 16 numbered drills; evidence SHA-256 0x45d0ee726f5ac83cb03cf71090864fd78452a1e6e734b3e838f5f9f5d39a3649
• actual Anvil keeper race: one same-action success, one benign-race revert, activeEvaluationProposalId=1
• deterministic ten-instance pre-broadcast drill: 10 actions, 20 successful simulations, 0 keeper broadcasts
• full economic gate on fresh Anvil: 10 receipts/evaluators/FLMs, 10 active evaluations, exactly 1 per FAO, 100 canonical logs, 0 public broadcasts
• canonical index report SHA-256: 0x5c7ea50b2f216b56bc152834f4fdde07e0c038de95e24f38d1f548283e8c1186
• deterministic full-gate SHA-256 (wall clock excluded): 0xb21d2dd3b2011f1108db513f6411d397a5bef36aabbf7caa15b0806514f5afe8
• resources: 102 transactions/blocks, 388,365,524 gas total, 18,203,139 max transaction gas, 101 chain seconds
• observed post-rebase wall time: 60.436 seconds

Remaining gap

Inject an external signer under the existing funding caps and exercise a controlled non-loopback deployment only when explicitly authorized. RPC fault injection, 100/1000 tiers, subsidy automation, evaluator market lifecycle/FLM conditional migration at scale, and UI remain later slices.