# Car: the funded oracle and the trading book Capture: 16 September 2026. Oracle state is pinned to Polygon block 93,917,716 (18:00:07 UTC), hash `0x35c130cd6b4e6b787a701baa490a9bef663a160d73af58fd6c1ae5de58cbde8b`. ## Addresses and identity Car's trading Safe: `0x7c3db723f1d4d8cb9c550095203b686cb11e5c6b`. The captured official public-profile response maps the Safe to Car and X handle CarOnPolymarket. It returns verifiedBadge true, communityMod true, creator true and mod false. These are separate API fields. Funded oracle address: `0x3e8379fa4e49648d88dc6c885abd776adbdc2132`. The financial link consists of exact ERC20 Transfer logs between this address and Car's Safe. This establishes funding in both directions. It does not identify the oracle private-key holder. Address prefixes or suffixes are not used for attribution. ## Proposal census and request results Polygon Blockscout account transactions were paginated to the requested January 1, 2024 boundary: 382 transactions, complete. Select transactions signed by the oracle address, status ok, decoded method proposePrice or disputePrice, targeting either: - OptimisticOracleV2: `0xee3afe347d5c74317041e2618c49534daf887c24` - ManagedOptimisticOracleV2: `0x2c0367a9db231ddebd88a94b4f6461a6e47c58b1` 121 successful proposals and 20 successful dispute submissions pass raw receipt checks. Six failed proposal attempts and one failed dispute attempt are excluded. The checks match transaction signer, destination, block, ABI input parameters, emitted oracle contract, requester, identifier, request timestamp, exact ancillary bytes, submitting address and proposed price. ABI padding outside the stated bytes length is ignored, as Solidity ignores it. Proposed values are signed int256, not unsigned integers. For each request, eth_call getRequest and getState at the pinned block provide the proposing address, disputer and settled answer. All 121 proposals are settled. 43 were disputed: 31 settled at the proposed value and 12 at a different value. Another 78 settled without a dispute. Thus 109 settled as proposed. The 20 dispute submissions are a separate action category; a dispute signed by this address does not make it the proposer of that request. Exactly `-2**255` is the Too Early response. Eleven rejected requests returned this value. One returned zero (No). Do not relabel Too Early as a final No market resolution. The full oracle proof bundle contains the raw contract calls and results alongside the transaction inputs and oracle event logs. Primary contract sources: - https://github.com/UMAprotocol/protocol/blob/master/packages/core/contracts/optimistic-oracle-v2/implementation/OptimisticOracleV2.sol - https://github.com/UMAprotocol/protocol/blob/master/packages/core/contracts/optimistic-oracle-v2/interfaces/OptimisticOracleV2Interface.sol - https://github.com/UMAprotocol/managed-oracle/blob/master/src/optimistic-oracle-v2/implementation/ManagedOptimisticOracleV2.sol ## Historical proposer permission For each of the 68 managed-oracle proposals, call getProposerWhitelistWithEnabledStatus(requester, identifier, ancillaryData) on the managed oracle at proposal block minus one. All 68 return enabled true and a membership array containing the funded oracle address. The first is February 6, 2026; the latest September 3. The public whitelist response excerpt includes the ABI header, member count, cited proposer entry and array index, plus a SHA256 hash of the full raw response bytes. Unrelated member entries are omitted. The exact call is included for replay. Legacy-oracle proposals are labeled separately. Enabled-list membership establishes permission to submit a proposal; disputed outcomes are still subject to the oracle's resolution process. ## Market joins and holdings The position cohort is all twelve rejected proposals plus all six proposals since August 17: 18 submissions across 16 markets. It is deliberately selected, not a sample of all proposals for estimating the prevalence of trading overlap. Use market_id from the exact ancillary question where present. For the two older questions lacking IDs, use the exact market slug. Require the returned Gamma market's canonical title to equal the ancillary question title. Gamma's clobTokenIds and outcomes map each Yes/No token to that question. The selected market metadata is captured with the research artifacts; market_coverage.csv records IDs, conditions and source activity URLs. Read balanceOf(Car Safe, token ID) on ConditionalTokens `0x4d97dcd97ec945f40cf65f87097ace5ea0476045` at proposal block minus one, for both outcomes. Six decimals. The 36 exact calls and raw results are downloadable. These are end-of-previous-block balances, not API trade-net estimates. They concern this Safe, not an unobserved person's entire portfolio. Balances measure shares; they are not realized profit. For all eleven Too Early proposals, shares of the proposed outcome exceed the opposite outcome at that block. These are eleven submissions across nine distinct markets, including repeated requests. The twelfth rejection, Oman, proposed Yes while Car held zero Yes and 1,047.17 No. Final market status is taken from captured Gamma closed plus outcomePrices fields; an open market remains labeled Open. The oracle response and the market-at-capture result occupy separate columns. The December uranium market was still open at capture. ## Trading timestamps Collect complete v2/activity feeds for Car, filtered by each condition, with start=1, end fixed to the capture time, ascending timestamps and all supported activity types. All 16 market feeds completed. For each rejected proposal, find the latest prior BUY timestamp on the proposed outcome and retain each distinct transaction tied at that timestamp. An absent prior matching buy produces no example. The twelve resulting proposal/trade pairs include the same purchase preceding two repeat proposals. Verify successful receipts and block headers. Decode only OrderFilled events emitted by the allowlisted official V1/V2 exchange addresses. Require the order maker to be Car, the side BUY and the token ID to match the proposed outcome. Sum unique log-indexed maker-side fills per transaction. Check the share amount against the distinct corresponding API rows. V2 quote fees are recorded separately from filled quote amount; the activity API quote includes those fees. The Minecraft No buy has 3,329.657 PUSD of filled quote and 0.16698 PUSD fee. Preserve denomination: older USDC.e fills and newer PUSD fills are separate. Timing is proposal block timestamp minus trade block timestamp, in seconds. One case is 176 seconds, another 324 seconds. These intervals describe the verified examples; no population-level statistical significance is asserted. Primary event definitions and contract configuration: - https://github.com/Polymarket/ctf-exchange/blob/main/src/exchange/interfaces/ITrading.sol - https://github.com/Polymarket/ctf-exchange-v2/blob/main/src/exchange/interfaces/ITrading.sol - https://raw.githubusercontent.com/Polymarket/py-clob-client/main/py_clob_client/config.py - https://docs.polymarket.com/api-reference/feeds/list-account-activity ## Direct funding The complete Polygon oracle token-transfer history contains 624 explorer records. Normalize exact allowlisted token contracts, retaining zero thresholds during collection, then select direct transfers between the two addresses. Deduplicate by transaction hash and log index. Verify receipt success, token contract, from, to, amount, log index, block hash and block timestamp. All 56 selected transfers use Polygon USDC.e `0x2791bca1f2de4661ed88a30c99a7a9449aa84174`: 37,966 from Car to the oracle in 37 transfers and 20,798 back in 19 transfers. Cross-chain routes are outside these totals. The 650 transfer at 00:48:18 on April 3, 2025 precedes the next sanctions proposal at 00:49:52 by 94 seconds. ## Corrections to the archived original chapter The old reversal table mixed request dates, disputes and eventual market results. The new record uses actual successful proposal transactions and the request's signed resolved price. The headline counts are twelve rejections, eleven Too Early and one No. One self-dispute is directly verified for this oracle address; this dossier makes no aggregate claim involving a separate inferred hub. Historical balance corrections: April 23 ceasefire, 44,785.66421 Yes (old trade-net estimate 39,343); July 3 diplomatic meeting, 2,577.67725 Yes (old estimate 9,699). The verified April 2 last-time sanctions buy is 176 seconds before the proposal. The old chapter is retained in an explicitly marked archive with a prominent link to these corrections. ## Reproduction Research modules: `python3 -m research.car oracle` then `python3 -m research.car_evidence`. Cached raw API/RPC captures preserve the dated run. `research/publish_car.py` renders the evidence into the site; `python3 build_site.py` rebuilds all pages. Unit tests exercise signed-price decoding, ABI padding, proposal/dispute separation, exchange-event filtering, transfer directions, holdings alignment, timing and aggregate reconciliation. Evidence-file hashes are published in manifest.json.