Who actually paid? Payout ledger and economics

1. Summary answer

The headline survived, and it is now a measured number rather than an absence of evidence. No agent-first venue shows an independent buyer paying agents at meaningful scale. We traced every payout on the five venues that pay agents for work:

• $230.37 reached agent workers in the 30 days to 1 October. This is the net amount after platform fees; the gross is $251.47. It is spread across TaskMarket, Virtuals ACP, Execution Market, MoltJobs and AgentPact.
• About 45% of it ($102.64) came from the operator or from wallets linked to the operator. The range is 36% to 49%, depending on how Virtuals ACP's payer split is extrapolated (see section 5).
• The other 55% ($127.73) came from "unlinked" wallets. These are wallets with no link to the operator within two funding hops. That is not the same as proven independence, and 92% of the unlinked money is on TaskMarket.
• 0 payers were identified as independent, and $0 came from identified independent payers.
• 0 venues reach meaningful scale. The pool's bar is US$1,000 in 30 days from at least 5 unlinked or identified-independent payers. The best case is TaskMarket: $117.80 net from 10 unlinked wallets, which is 12% of the bar. Those 10 wallets are really about 8 owners, because the top 3 payers (78.5% of gross) share one funder.

Venue by venue:

• TaskMarket is the only venue where outside money might be real. All 10 payers are unlinked, and no operator wallet funds them. But one funder cluster, around wallet 0xd8c5763b…, stands behind 78.5% of the money. We could not identify its owner.
• MoltJobs and AgentPact are 100% operator-funded.
• Execution Market is 100% paid by the operator's own agent swarm. It moves from "unknown" to "no" on independent payers. It has also been idle since 2026-09-29 20:33 UTC.
• Virtuals ACP is tiny: $29.19 paid to providers in 30 days, against $62.50 refunded. In the 14 days we decoded job by job, 66% of the paid USD went between client and provider wallets that share a funder or a sink. Under the pool's rule that counts as operator-linked (self-dealing).
• x402 is different. Real buyers exist for agent-facing paid APIs: dTelecom took $35,381 from 26 buyers in 30 days, and Cluster Protocol $55,226 from 722. But these are per-call API sales, and the buyers were not traced. The x402 sample is kept in the ledger and excluded from the agent-work totals.

Is it worth the tokens?

• On TaskMarket, an equal-chance attempt at the median task is worth about $0.02.
• On Superteam Earn's agent-allowed listings, a submission is worth about $4–5. These rewards are advertised, not traced, and a human must claim them.
• On Algora, recent paid bounties had a median of 3 attempters for $100. But only 2 awards dated July to September 2026 were found.

2. The homepage numbers, recomputed

independent_payer_evidence was decided under the written rule for all six agent-paid or selling E1 records:

Two new headline numbers:

• $230.37 paid to agent workers in the last 30 days, net to the worker, across the 5 venues.
• 45% of it came from operator or operator-linked wallets (range 36–49%). The share from identified independent payers is 0%.

3. Ledger, 30 days to 2026-10-01

USD is net to the worker unless marked as gross.

What each payer actually is

• TaskMarket. The bootstrap called the funding "opaque" because escrow deposits arrive from relayer 0x3C0820e2…. That is now corrected.
  • The relayer is the deployer of the escrow and of the TaskMarketForwarder. It is also a pass-through: each requester pays it by x402 (EIP-3009 transferWithAuthorization), and it forwards the same amount into escrow.
  • All 90 of the 90 escrow deposits in the window matched one-to-one to a payment from the task's listed requester, so the listed requester is the real payer.
  • The fee address 0xD822… has never sent USDC to anyone.
  • Top payer 0x436326b6… (agent 63859, creative 2-USDC bounties) paid $72.00 gross.
  • Requester 0xa175d51b… paid $10.00 for agents to register with MolTrust, which the task names. MolTrust has not confirmed this, so it is not counted as identified independent.
• Execution Market. Every one of the 3,128 completed tasks since February was published either by a kk-* KarmaCadabra swarm agent (2,856 tasks) or by an operator test or demo wallet (272 tasks).
  • kk-* wallets are funded through the Disperse multisend. The funding call, tx 0x674238d4…, was sent by 0xD3868E1e…, which also publishes the platform's own "[GOLDEN FLOW HEDERA]" demo tasks.
  • Non-swarm executors earned $3.74 (8.4%) in the window and $5.76 (5.1%) over the venue's lifetime. All of it was paid by swarm or operator wallets.
  • The largest payment ever made to a non-swarm executor is $0.25 ($0.2175 net), made on 2026-04-22 by an operator test wallet.
• MoltJobs. Every escrow deposit since at least 2026-08-25 came from founder wallet 0xc66c3109….
  • The escrow deployer has sent 98.12 USDC to that founder wallet (founder inflows).
  • The 4 jobs ever posted by anyone other than the founder are all cancelled and were never funded (jobs API).
• AgentPact. The only payer, 0x4DDcf20a…, is the escrow's creator and also takes the 10% fee. There has been no deposit or release since 2026-09-01.
• Virtuals ACP. Three client–provider pairs carry 66% of the paid USD in the 14-day sample:
  • 0xeb5d → 0xb37f, $3.35: both sweep to the same account, 0xcfdde4d2…;
  • 0x3ffc → 0xfcf0, $0.90: the provider pays the client's funder;
  • 0x60ff → 0x436f, $0.36: both share a funder.

    One wallet, 0x12b5d81c…, ran 2,802 fund-and-refund loops of $0.01 each. No payer traced back to Virtuals' fee address or deployers.

Did workers withdraw? Yes, on both venues we checked.

• TaskMarket: 3 of the top 5 workers sent USDC to outside, untagged addresses. One example is 15.95 USDC on 2026-09-04.
• MoltJobs: the top worker wallet sent 21.85 USDC to an external contract wallet (tx).
• We don't know who owns the destination addresses.

Links across venues (found when merging the sub-passes):

• Wallet 0x732d8af9… was paid on both TaskMarket and Execution Market.
• Funder 0xf70da978… paid TaskMarket requester 0xb8746a06… ($1.39 net in the window) and also Execution Market's operator test wallet 0x857fe615…. The 0xb874 payer stays "unlinked" for TaskMarket, because the link is to a different operator.
• MoltJobs' top worker received 9.0 USDC from 0x26572fF2…, which is the x402 pay-to wallet of Frantic ("live bounties for AI agents"). Frantic is added to the backlog.

4. Net vs gross

"Paid to agent workers" means USD that reached the worker's wallet after platform fees. Here is how each venue was converted:

Execution Market events in payouts.json carry the gross bounty in usd and the net in net_to_worker_usd. For every other agent-work event, usd is already net.

5. How payers were classified, and the limits

Rule (the pool's written payer rule, applied to every event):

• operator: the operator, its fee address, the escrow deployer, or an account the operator controls.
• operator_linked: one of three cases:
  • funded by an operator wallet within two hops;
  • circular funding (payer and payee share a funder or a sink);
  • part of the operator's own agent swarm.
• unlinked: no link found within two hops.
• identified_independent: a payer that public evidence names as distinct from the operator.

Payers were classified by the token-transfer from, not by the transaction sender. This matters on TaskMarket (x402 relayer) and on ACP (ERC-4337 bundlers).

Limits, stated plainly:

1. On-chain data proves transfers, not identities. "Unlinked" means only that we found no link. It does not show the payer is independent. Nobody in the ledger is identified as independent.
2. One funder stands behind most of TaskMarket. The top 3 requesters account for 78.5% of 30-day gross: 0x4363 ($72.00), 0xeecd ($15.00) and 0xf138 ($13.00). All three trace within two hops to the EIP-7702 wallet 0xd8c5763b…:
   • it sent 1,100 USDC directly to 0xf138;
   • it sent 100 USDC to 0x2dc32ba2, which sent 85 USDC to 0x4363;
   • it sent 3.5 ETH to 0x279e08f7, which sent USDC to both 0x4363 and 0xeecd.

     No link from 0xd8c5 to TaskMarket's operator was found. One weak hint: 0xeecd's task is tagged lucid-agents, which is the operator's (Daydreams') SDK. If 0xd8c5 turns out to be operator-affiliated, the operator or operator-linked share across all five venues rises from 45% to 85%.

3. Under a strict reading of the rule, the share is higher. Payer 0xf138 once received 4.0 USDC from operator wallet 0x3C08. That is 0.36% of its inflow and looks like a refund. Read literally, the two-hop rule makes 0xf138 operator-linked: TaskMarket unlinked falls to $105.77, and the overall share rises to 49.8%.
4. The ACP class split comes from a 14-day sample. The Blockscout log pull hit HTTP 429 at about 2026-09-17, so payer classes were measured on 92 paid jobs from 2026-09-18 to 10-01 ($6.98).
   • We applied those shares (66.0% operator-linked, 34.0% unlinked) to the full 30-day $29.19: $19.26 operator-linked and $9.93 unlinked.
   • This is an extrapolation. Most 30-day releases fall before the sample (spikes on 09-06 and 09-10).
   • Bounds for the overall operator or operator-linked share: 36.2% if all ACP money is unlinked, 48.9% if all of it is operator-linked, and 41.4% if ACP is left out.
5. Hop depth.
   • TaskMarket: 2 hops on USDC, plus ETH where found. Hop 2 is incomplete for 4 small payers (under $3 together) because of HTTP 429.
   • ACP: only 1 hop, using each wallet's latest 50 transfers. Spam airdrops push older funders off that page. $0.405 from 11 small payers was not traced and is counted as unlinked (funding_traced: false).
   • Execution Market: swarm membership comes from kk-* display names plus funding traces on 9 of about 25 kk wallets.
   • Internal ETH transfers and other tokens were not traced anywhere.
6. Rate limits. Blockscout HTTP 429 limited:
   • TaskMarket hop-2 tracing;
   • the ACP log pull;
   • the x402 buyer-share sampling (only dTelecom has a measured top-buyer share: 10.3%);
   • one Execution Market Base transaction;
   • the Moltlaunch escrow re-check.

     basescan.org returns 403, so all Base data comes from Blockscout.

7. Execution Market's numbers disagree with each other. Bounties on all 3,128 completed tasks sum to $113.83. The public metrics counter reads $504.84, and its basis is not documented. We use $113.83. The bootstrap had quoted $504.84 as lifetime volume.
8. x402 buyers were not traced. Sample events carry unlinked with funding_traced: false. The seller totals are left unclassified (usd_unclassified). Their sellers are not shown to be agents doing work. On raw volume, dTelecom and Cluster would pass the $1,000 / 5-payer bar if their buyers were counted as unlinked. That is why x402 is reported separately.

6. Economics and calculator presets

Per-venue economics. Data is from the economics field in venues.json; eight venues are filled.

Calculator presets (in payouts.json → calculator_presets). EV = reward × (1 − fee) ÷ competitors, which assumes an equal chance for every attempt.

Superteam vs TaskMarket. An agent-allowed Superteam submission is worth about $5 on equal chance, against about $0.02 for a TaskMarket attempt, roughly 200 to 250 times more. But the cases are not equal:

• Superteam rewards are advertised amounts, not traced payments.
• Most agent-allowed listings are X-post or content bounties.
• Superteam requires a human claimant, plus KYC on Superteam-sponsored listings.

Equal-chance EV is optimistic wherever wins concentrate:

• TaskMarket's top worker took 10.2% of worker USD.
• Execution Market's top worker took 18.2%.
• MoltJobs' top worker took 57.1%.
• Algora's top winner took 31% of the sample.

Algora caveats.

• Only 2 awards dated July to September 2026 were found across more than 30 org boards. The plan's target of 10 or more could not be met, so the sample uses the 12 most recent awards.
• 7 of those 12 are maintainer tips with no competition.
• 0 of 12 winning PRs disclose AI authorship.
• The most active Algora-bot org right now, UnsafeLabs, shows 188 open bounties and 0 completed.

7. What changed since 30 September

In total, 14 records changed and each one carries a change_log (64 entries). Twelve records were fully re-checked and have last_verified 2026-10-01. No record was promoted, demoted or dropped, and new_records is empty.

8. Backlog (probed, not promoted)

9. Open questions by venue

• TaskMarket
  • Who controls 0xd8c5763b…, the funder behind 78.5% of 30-day gross? Is it Daydreams-affiliated?
  • Does MolTrust confirm it funds requester 0xa175d51b…? If so, that $10 would be the first identified independent payment.
  • Hop 2 is still open for 4 small payers.
  • Is payer 0xb874 (shares funder 0xf70da978 with Execution Market's operator test wallet) an Ultravioleta agent?
• Virtuals ACP
  • Where are Revenue Network incentives paid from, in which token, and are they still running?
  • How much ACP v2 volume is on chains other than Base?
  • Who controls 0xcfdde4d2…, 0xcbf33618… and 0xcadb1953…?
  • Run a 30-day job-level class split with a two-hop trace.
• Execution Market
  • What does the $504.84 counter measure?
  • Has activity resumed since 2026-09-29?
  • Is 0xD3868E1e… the DAO treasury or a team member's wallet?
• MoltJobs
  • Is the top worker wallet 0x8A9508d2… (57% of the window) linked to the founder?
  • Was its 9.0 USDC from Frantic a bounty payout?
• AgentPact
  • Will any release happen before the 2026-10-31 stale line?
• x402
  • Two-hop tracing of the dTelecom and Cluster buyers.
  • What does Cluster actually sell?
  • Is KYC required on the CDP facilitator?
• Algora
  • Has bounty volume fallen in 2026?
  • Its latest confirmed payout (2026-08-26) passes 60 days on about 2026-10-25.
• Kaggle / ARC Prize
  • Milestone #2 results, expected around 2026-10-06.
• Moltlaunch
  • The escrow re-check failed with HTTP 429; retry next run.

For the planned walkthroughs: no venue has an identified independent payer. TaskMarket's unlinked volume is real but small ($117.80) and concentrated. Under the pool's research plan, the walkthroughs therefore default to the human route (a HackerOne or Bugcrowd hackbot with human sign-off) and an x402 seller walkthrough. TaskMarket is the only agent-first candidate, and only if the 0xd8c5 question resolves in favour of independence.

10. Conduct statement

This run opened no account, took no job, signed with no wallet and spent no money. All requests were read-only GETs, including two targeted GETs in the merge step to confirm the Frantic address match. Every payment described here was made between other parties and is shown by a public record linked above. None of the pathways has been tested by us, and pathway_tested stays "no" on every record.