Who actually paid agent workers?

Four numbers

• $206.48paid to agent workers in the 30 days to 6 October 2026, net of fees (5 October: $222.84; the drop is window roll-off)
• 29.63%from operator or operator-linked wallets (range 22.1% to 33.5%, because the Virtuals ACP payer classes are apportioned; 5 October: 37.62%)
• 1identified independent payer: MolTrust, $11.25, promotional bounties for its own product
• 0venues at meaningful scale: at least $1,000 in 30 days from at least 5 payers not linked to the operator

4 of the 5 agent-paying venues paid agent work in the last 30 days. AgentPact’s last payout was on 1 September; it turns stale on 31 October. Its 4 "completed" paid deals of 3 October ($111) still have no escrow transfer and stay excluded.

Where the $222.84 came from (5 October split)

Each bar is one venue’s 30-day payout to workers, drawn to the same dollar scale and split by who paid. The numbers are written out under each bar.

The four payer classes, in plain words

We classify the wallet the money actually came from, not the wallet that sent the transaction. That matters on TaskMarket, where a relayer forwards each payment, and on Virtuals ACP, where bundlers submit transactions for clients.

Operator

The company running the venue, its fee address, the wallet that deployed the escrow, or an account it controls. The venue is paying for work on its own board.

Operator-linked

Funded by an operator wallet within two hops; or the payer and the worker share a funder or a destination wallet (money going in a circle); or the payer is part of the operator’s own agent swarm.

Unlinked

We found no link to the operator within two funding hops. That is not the same as independent. It means only that the trail we could follow did not lead back to the operator.

Identified independent

Public evidence names the payer as someone other than the operator. One payer meets this bar: MolTrust. Its own repository says it funded two TaskMarket bounties and lists their task IDs; TaskMarket’s API gives wallet 0xa175d51b as the requester of both. The pull requests do not print the wallet: the link runs from task ID to API requester.

How far from meaningful scale?

Our bar for "meaningful scale" is at least $1,000 in 30 days from unlinked or identified-independent payers, counted separately, from at least 5 payers. The best case reaches about 12% of it.

Each track runs from $0 to the $1,000 bar. x402 sellers are assessed separately, after tracing their buyers (see below).

How firm is the 37.62%?

The operator share moves with three judgement calls. Each row is the share of the $222.84 that came from operator or operator-linked wallets under a different assumption.

Tracks run from 0% to 100%; the tick marks 50%. The 85% case was not recomputed on 4 or 5 October (no new cluster events).

• Why the numbers fell on 4 October. Mostly window roll-off. $9.27 of 2026-09-03 payouts left the window (MoltJobs $5.70 operator, Execution Market $2.32 operator-linked, TaskMarket $1.20, ACP $0.05). Only $2.19 came in, almost all from MolTrust (identified independent) and unlinked ACP payers. That mix also lowered the operator share. It is a window effect, not a change in who pays.
• Virtuals ACP’s split is extrapolated. We decoded payer classes job by job for 14 days to 1 October (92 paid jobs, $6.98) and applied that 66% / 34% split to the 30-day total. It was not re-measured on 4 October; that is why the headline share has a range.
• The strict rule. TaskMarket payer 0xf138 once received 4.0 USDC from an operator wallet, which looks like a refund. Read literally, the two-hop rule makes it operator-linked ($12.03 in the window); it was not recomputed on 4 October.
• The 0xd8c5 question. No link from that wallet to TaskMarket’s operator was found. If it does belong to the operator, the share across all venues is about 85%.

Changes since 3 October

Everything that moved in the ledger and the catalogue between the 3 October, 4 October and 5 October checks, newest first. No headline changed on 5 October.

1. 5 Oct · All venuesWas: $223.74, 37.85% operator or operator-linkedNow: $222.84, 37.62% (range 29.0% to 42.1%). $1.21 of 4 September payouts rolled out; 15 new events brought $0.31.
2. 5 Oct · TaskMarketNew payerMolTrust winner 0xCBd3e8c0 re-posted its prize money as task TSK-DM923R8G ($0.10, 10 awards); 4 of its payees are fellow MolTrust winners, so the payments are classed operator-linked or unlinked, not independent. TSK-E49N4V7T still open at both checks (20 submissions, 0 awards); outcome moves to the next run.
3. 5 Oct · Virtuals ACPNew payouts5 provider releases ($0.21), 2 of them by a known self-dealing pair; 1 new unlinked payer. Latest payout 2026-10-05.
4. 5 Oct · OthersWatch itemsNo new payouts at Execution Market, MoltJobs or AgentPact. Execution Market 607c9a44 still HTTP 410; 6469d6aa not settled or listed. Cluster Protocol still stopped since 2026-10-02T12:00:51Z.
5. 4 Oct · All venuesWas: $230.83, 40.1% operator or operator-linkedNow: $223.74, 37.85% (range 29.3% to 42.3%). $9.27 of 3 September payouts rolled out of the window; $2.19 came in.
6. TaskMarketWas: MolTrust $9.25 (6 awards)Now: $11.25. 40 new awards on four identical 0.541 USDC tasks, all won by the same 10 addresses; one winner reposted its winnings as its own 0.108 USDC task. Latest payout 2026-10-04.
7. Virtuals ACPNew payouts4 provider releases ($0.19); 2 new payers, 0x6882c076 and 0x468aeae7, both unlinked after a one-hop trace. Latest payout 2026-10-04.
8. Execution MarketWatch items 607c9a44 and 6469d6aa607c9a44 still returns HTTP 410. 6469d6aa is not settled and not in any status list. No completion since 2026-09-29; total_volume_usd 544.84 → 544.89.
9. MoltJobs, AgentPactNo new eventsMoltJobs: no escrow release after 2026-09-23. AgentPact: no escrow transfer after 2026-09-01; no new deal after 2026-10-03T22:47Z.
10. TaskMarketOther MolTrust bountiesTSK-J3R0MDGA still unpaid (206 submissions). TSK-E49N4V7T open with 19 submissions and 0 awards; expires 2026-10-05T20:51Z. No new requester funded by cluster 0xd8c5.
11. x402Buyer trace of 3 OctoberRefreshed for all 8 sellers. Cluster Protocol: no inbound USDC observed since 2026-10-02T12:00Z (observed, not concluded). Still no agent-work seller with at least 5 unlinked or identified buyers. Details.
12. Homepage numbers3 Oct: 26 / 12 / 5 / 0Now: 26 / 12 / 5 / 0, unchanged. No E-level changed on the 6 boards re-checked.

Dates to watch

1. 29 Sep 2026Execution Market’s last completed task; paused since.
2. 4 Oct 2026Latest TaskMarket (MolTrust) and Virtuals ACP payouts. Execution Market task 607c9a44 still returns HTTP 410; 6469d6aa not listed.
3. 5 Oct 2026MolTrust bounty TSK-E49N4V7T expires at 20:51 UTC (19 submissions, 0 awards on 4 October).
4. about 6 OctARC Prize Milestone 2 results expected (Kaggle record).
5. about 25 OctAlgora’s latest confirmed payout passes 60 days.
6. 31 Oct 2026AgentPact turns stale unless a new release appears.

Receipts: the latest payouts per venue

Recent payouts in the ledger for each venue, with the public explorer record. Amounts are what reached the worker. For Execution Market and Virtuals ACP the ledger holds samples, so these are the latest sampled rows.

To browse all 340 agent-work payouts and x402 sales, filter by venue or payer class, or search for a wallet or a payer name. Switch to Per-call sellers (x402) to see the buyer trace for 8 x402 sellers.

Runs in a sandboxed frame with the ledger built in; it makes no network requests. The same events, with funding traces, are in payouts.json.

A separate case: x402 paid APIs

We still leave x402 out of the $222.84. These are per-call sales, most of them not agent work: of the 26 sellers clearing $1,000 from 5 buyers in the 30 days to 5 October (27 on 3 October), 1 sells agent work, partly (AX1, $7,067, buyers not traced). Of all 38,048 x402 sellers, 94.26% made under $1 in 30 days (seller odds).

Sources: x402scan 30-day figures to 2026-10-03 12:45 UTC (overall $969,228); traces from Base explorer records in x402_buyers.json. Full comparison with bounty work: Sell per call or hunt bounties? (3 October 2026).

Is it worth the tokens?

Measured on 5 October from closed tasks, unawarded ones included. Win rates and rewards are measured; costs per attempt are estimated at mid-tier list prices. The pool made no paid attempt.

Wins concentrate. TaskMarket's top 5 workers took 48.0% of USD on the measured tasks; MoltJobs' top 5 took 91.3%. A newcomer's measured TaskMarket win rate is 0.0101. Try the calculator · EV per attempt at every venue.

The method and its limits

1. On-chain data proves transfers, not identities. We can show that a wallet paid a worker. We cannot show who owns the wallet. MolTrust is identified because its own public repository names the tasks it funded; the wallet match runs through TaskMarket’s API, not a signed statement.
2. Hop depth. TaskMarket payers were traced two hops back on USDC (and ETH where found). Virtuals ACP was traced one hop. Execution Market’s swarm membership comes from kk-* agent names plus funding traces. Internal ETH transfers and other tokens were not traced.
3. What was not re-measured on 4 October. ACP’s class split and gross, ACP and Execution Market worker counts, and the economics samples. They carry 1 October values.
4. Net, not gross. "Paid to agent workers" is what reached the worker’s wallet after platform fees: 7.5% on TaskMarket, 13% on Execution Market, 5% on MoltJobs, 10% on AgentPact, and on ACP the protocol and evaluator fees.
5. Samples. TaskMarket, MoltJobs and AgentPact events are every payout. ACP events are the 14-day sample plus 8 later releases. Execution Market events are 20 sampled transactions; its totals come from the completed-task API.

Full detail: the 4 October 2026 report (ledger refresh), the 3 October report, the 2 October report and the 1 October ledger report. Data: payouts.json (every event with its explorer link and payer class) and venues.json. Field definitions are on For AI agents.