Direct-link public financial proof — 2026-10-31 05:19 pass

No new Car–drainer operational or causal connection is established. This bounded pass completes the actual-key comparison for 383 July-condition counterparty order appearances across 95 trade transactions (98 financial transactions including three merges). It resolves 332 ordinary signatures and 51 nested signatures, identifying 172 distinct counterparty authorizing keys; no exact match to the retained attacker-side or role-aware comparison inputs was found. These ordinary counterparties are not alleged Car associates or attackers. There are 479 executed appearances including 96 Car appearances, and 418 distinct signed-message/signature pairs across the entire condition corpus.

The actual Ethereum proceeds handoffs 3f701 → 74e9 → E5 independently recover their public keys. The stated incoming-native trace scope before 74e9's first proceeds credit is empty, with partition and positive controls. That first credit is attack proceeds, not an independent attacker seed. The indexer rows, native trace scope and historical state are bounded; they do not establish every asset, chain, private arrangement or human controller.

The 26-address role input is a selected observed subset, not the full verified financial network. The broad 117-address input adds five observed financial/controller endpoints omitted from an older mixed112 inventory. It retains public services, computed formats, financial associates and human-control leads with distinct roles; membership does not assert shared human control. Proven private financial association remains distinct from unproved Car ownership/control. The target list likewise separates actual flow, actual authorizing keys, computed formats and potentially independent payer roles.

The nested replay derives ERC7739 messages and public authorizing keys from actual calldata, binds positive filled-order logs and retained headers, and corroborates all51 with historical owner/ERC1271 responses. Those calls are execution-block end-of-block state, not signing-time or intrablock observations. The standard component's own51 unresolved long appearances are closed by the separate nested component in the combined report. A matching public-book trade, had one appeared, would not alone establish collusion.

Infrastructure recovery adds one historical runtime's source metadata: ten of16 referenced resources have retained bytes, six remain unrecovered. No new malicious artifact/control or vendor attribution source was recovered. Full JavaScript, social, article, profile, HTML and other human source bodies remain local. Curated metadata, hashes and source indexes do not authenticate those excluded bodies, constitute an archive-complete corpus or make local absolute references portable. The causal audit explains which exact session/deployment/control records would be needed to distinguish hypotheses; it authorizes no access to private systems.

Run from the extracted ZIP root:
  PYTHONDONTWRITEBYTECODE=1 python3 verify_package.py

The stdlib checker validates every selected byte, JSON/gzip/Python syntax, then runs the standard signature replay, nested replay, combined consistency join, standard altered-signature/known-counterparty controls, origin financial/trace controls and an independent ABI/RLP financial replay. Sockets and DNS are disabled in the checker and every Python child. The origin helper additionally rejects rehashed actual-signature and earlier-trace mutations. No signing, private keys or network is required. A source-manifest hash alone is not the adversarial check: the financial replays reconstruct actual message hashes and keys. Metadata/audit consistency is a separate, narrower claim.

The previous public July raw financial RPCs, headers, baseline order report and shared crypto helper are placed once under nested-signatures; root scripts use that directory. Lane selections remain exact bytes. Their local source-index entries can refer to deliberately excluded bodies. Independent component commands and boundaries are in each lane README. The selected read-only fetch script is provenance only and is never run by the offline checker.

Retained RPC/receipt/header consistency and key recovery are not a standalone chain-consensus proof, human identity, voluntary consent, intent, private-session authentication or exploit attribution. Different public keys do not prove different human controllers. Unrelated oracle/trading/deletion findings do not supply the missing Car–drainer edge.
