Agent drafts, owner submits: Superteam payments could not be verified, and Algora's bounties stopped in July

1. The answer

In one sentenceOn Superteam Earn's agent-allowed listings, winners are announced, but no payment can be seen on any public route (30 of 30 sampled winners); with about one eligible listing every 11 days, an owner can expect about $0.24 a day at any daily cap and a median of about 301 days to first cash, and Algora's paying boards have made no new award since 2 July 2026.

• 30 of 30sampled Superteam winners whose payment could not be verified: no wallet, paid flag or transaction on any of 7 public routes (measured)
• about $0.24 a dayexpected at a $0.75, $1.50 or $5 daily cap: supply binds, not budget, at about one agent-allowed listing every 11 days (estimated; announced prizes)
• about 301 daysmedian wait to first cash at that pace: about 25 entries to a first win, plus award and payout delays (estimated)
• about $16 an hourof owner time at 10 minutes of review per entry; $8 to $32 for 20 to 5 minutes (minutes assumed; announced prizes)
• 10 wins, 0 claimedagent accounts that won (US$790 face), still unclaimed 97 to 120 days after the announcement, so none is shown paid (inference)
• 0 since 2 Julynew Algora awards on the 6 boards that paid in 180 days; about 2 open paying bounties remain (measured)
• 8.3% → about 21.7%TaskMarket: run 7's provisional unpaid share, now confirmed at the higher branch (dated correction, Section 9)

• Superteam Earn: winners are announced, but payment could not be verified on public routes. We sampled 30 winners (20 on agent-allowed listings, 10 on human-only listings for comparison) and tried 7 public routes. The payment fields (isPaid, paymentDetails, the winner's wallet) exist in Superteam's open-source code but are returned only after sign-in. Superteam stays at evidence level E3.
• Agent accounts have won, but none of those wins has been claimed, so none is shown paid. 10 winners carry Superteam's agent marker (US$790 face value). 4 of the 105 winners on agent-allowed listings are agent accounts; the other 6 won on two listings that are human-only today. All 10 were still unclaimed 97 to 120 days after winners were announced. "Not paid" is an inference: an agent has no wallet until a human claims it.
• Per day, supply is the limit. About 0.088 agent-allowed listings appear a day (8 in 90 days), and one account may enter each listing once. Raising the daily cap from $0.75 to $5 changes nothing: about $0.24 a day expected, on announced prizes.
• Algora: 6 boards paid 27 awards (US$3,335) in 180 days; the latest new award was on 2 July 2026, no new bounty has appeared on those boards since 9 July, and about 2 open paying bounties remain. We did not write an Algora walkthrough: there is no live paying supply to point an owner at.
• TaskMarket in daily terms: at a $1.50 daily cap, about 3.1 attempts a day (the supply limit) cost about $0.99 at mid-tier prices and bring about $0.08 expected, with a median of about 17 days to first cash. Only budget-tier prices break even, and only just (+$0.001 per attempt).
• What to do with this. If you point an agent at Superteam, enter only listings marked agent-allowed or agent-only, expect about one a fortnight, have a human claim the agent, and record any win as unpaid until the money arrives. Read the untested walkthrough and the plain answers in Is it real?

2. Superteam Earn: can a payment be seen?

No. We looked for the winner's wallet, a paid flag or a transaction for every sampled winner on every public route we could find, and read the open-source code to see what each route returns.

Superteam does track payment per winner; it just does not publish it. Its winner interface moves Foundation-paid winners through KYC, "KYC done" and "paid" stages, and sponsors record a transaction ID that is checked against Solana. Neither is visible without signing in. So the days from announcement to payment, and who paid, cannot be measured.

Secondary evidence only, none of it a payment record:

• The Earn FAQ says winners of Foundation-paid listings are paid "within 7 days of submitting the form"; external sponsors pay the winner's Earn wallet.
• The Terms of Use say Superteam Earn has no liability for a sponsor's non-payment.
• The homepage total (US$15,954,140) is the listed value of announced rewards plus approved grants, not money paid.
• Two web searches found no first-hand 2026 report of payment or non-payment; one undated personal blog claims a small prize arrived (not verified, not used).

3. Agent accounts have won, and none has been claimed

Superteam creates agent users with the last name "agent", and each of the 10 marked winners also shows isAgent: true on its public talent page. They are 10 different accounts with one win each. Claiming an agent moves its wins to the human's account, so a claimed agent win becomes invisible: 10 is a lower bound.

4. What it is worth per day, and per hour of the owner's time

Owners budget per day, so we translated every expected value we could into a day and an hour. The daily caps ($0.75, $1.50, $5.00) and the owner's review minutes (5, 10, 20 per entry) are assumed; supply and delays are measured; everything else is estimated from them.

Why the cap does not matter on Superteam: there is little to enter

Expected payout per day, at a $1.50 daily cap

Median days to first cash

Expected payout per hour of the owner's time

All translations

The return per $1 of tokens comes last on purpose. On Superteam and Algora it is large because an attempt is cheap next to a $100 prize, but it says nothing about how often you can enter, when you are paid, or whether you are paid at all.

Assignment venues have no win rate; payout per job ÷ cost per job at mid prices: Execution Market 0.178, Virtuals ACP 0.378, MoltJobs 2.102 (no job since 2026-09-23). The TaskMarket single-file-tasks-only row has return per $1 only (budget 3.71, mid 0.28; run 7 values); 12 rows are not computable. Algora's April–July supply of 0.194 bounties a day would have given about $1.76 a day (historical only). Try your own numbers in the calculator, which now has a daily cap and review minutes.

5. Superteam census and overdue stock

• 337listings in the census: 263 completed, 51 in review, 23 open; 27 agent-allowed, 310 human-only
• 26,534submissions; 1,505 winners read; 166 listings paid by the Foundation
• 51listings past deadline with no winners, 3,078 submissions waiting (6 October: 48 and 3,020)
• 3of them agent-allowed, with 224 submissions waiting (6, 48 and 136 days overdue)

Of the 51 overdue listings, 30 are more than 14 days overdue (1,473 submissions); the median is 33.6 days and 10 are more than a year overdue. The review feed is not capped: the code limits results only for the home and region views. Agent-only listings appear only on a verified-sponsor route and none falls in the window. On listings that did announce, the median wait from deadline to winners on agent-allowed listings was 6.5 days. All counts measured; source: human_route.json, census_summary and overdue_stock.

6. Sponsor records, and a check that did not work

We kept a record for each of the 133 sponsors in the census: listings, announcements, median days from deadline to winners, and listings overdue now. Sponsors are named only when the listing names an organisation. Flag counts, with thresholds we chose (counts, not judgements): too new 84, prompt 28, has an overdue listing 25, slow 17. Of the 23 sponsors with agent-allowed listings, 4 have an overdue listing now. Days from announcement to payment cannot be measured for any sponsor.

The sponsor check is a negative result. We registered the rule before looking: flag a listing if, when it was published, its sponsor already had an overdue listing; call the check useful only if it flags at least 10% of listings and the flagged listings announce winners within 30 days at a rate at least 15 points lower, with intervals that do not overlap.

The flagged listings did slightly better (−3.3 points, the wrong way round) and the intervals overlap. Even with look-ahead the check is uninformative (it would flag 56% of listings, −1.0 points). Under our own rule the check is not useful, so we do not recommend it.

7. Rules for "agent drafts, owner submits", and the owner's steps

21 new rule rows were quoted on 7 October (10 for Superteam, 11 for Algora and its maintainers). Each links to its row on Can my agent do this? Not legal advice.

Superteam Earn

• Agents may submit only to agent-allowed or agent-only listings. The agent API answers a human-only listing with 403 "Agents are not eligible for this listing" (superteam-earn-automation_ai-3).
• AI-drafted work submitted by a human: no rule found in the skill file, the FAQ, the Terms or the Notion FAQ (automation_ai-2, none found). Treat it as unclear, not as allowed.
• Some sponsors disqualify submissions with over 15% plagiarism (automation_ai-5).
• One submission per user, and per agent, per listing, enforced in code (account_identity-2). We found no clause on several accounts or several agents per claimant.
• Claiming an agent needs a completed talent profile (account_identity-3); users must be of age (18 or the age of majority) and may not open accounts in another person's name (account_identity-4).
• KYC (Sumsub) applies to Foundation-paid winners announced after 6 August 2025 (kyc_payout_eligibility-2).
• Payment is due "within 7 days of submitting the form", and the Terms deny liability for a sponsor's non-payment (payment_timing-1).
• Region rules are enforced at submission (region_eligibility-1); project listings need a Telegram contact from the human (automation_ai-4).

14 steps. Step counts and the 6 required profile fields are measured from code; minutes are assumed (low / central / high). An Earn wallet is created when the profile is completed. See the untested walkthrough for every step, its rule row and its stop conditions.

Algora and the maintainers who paid

• Algora's robot clause applies to the Algora service itself, so the owner operates the account (algora-github-bounties-automation_ai-1).
• Payouts arrive "typically 1-3 business days after a payment is completed", and the Algora bot comments when the contributor is paid (payout_timing-1, payout_timing-2).
• In India and the UAE, individuals cannot receive international payouts (kyc_payout_eligibility-1).
• Maintainers: Turso welcomes AI coding agents if you understand the code (turso); Coolify and Qdrant require disclosure, and Qdrant asks that PR descriptions and replies not be AI-written (coolify, qdrant); Activepieces auto-closes all unsolicited outside PRs (activepieces 1, 2); Archestra, Cap and PG-AGI have no contributor policy (none found).
• Owner steps for Algora: 10, all minutes assumed. One-off 13 / 35 / 75 minutes; 41 / 97 / 248 minutes per entry, because the owner must understand and defend a code change. On 8 awarded PRs the median was 2.5 reviews and 3 commits (measured proxy).

8. Algora: the paying boards went quiet in July

• Boards that paid in 180 days: 6. archestra-ai 11 awards, activepieces 7, CapSoftware 4, tursodatabase 2, PG-AGI 2, coollabsio 1. Two archestra awards show "onboarding pending" with no later paid comment.
• Supply is zero. A GitHub search for Algora bot comments on items created since 9 July returns 0; the paying boards opened 35 bounty issues in 180 days, all before 3 July. About 2 open bounties remain, both from individual sponsors. UnsafeLabs is kept apart: 188 open, 0 ever completed, newest bounty 15 May.
• Delays: from merge (or issue close) to award, median 0.09 days, 90th percentile 15.0 (23 awards). The payout delay is Algora's own statement.
• Evidence level stays E1 on a payout confirmation dated 26 August 2026 (the payout of a 2024 award). It passes 60 days on 25 October 2026; if nothing newer appears by then, the level will be reviewed.
• Whether any paid PR was agent-drafted is not visible on GitHub.

9. TaskMarket: the 8.3% becomes about 21.7%

• Grace recomputed: 16 days. The 95th percentile of positive award delays is 15.40 days over 284 awarded tasks (it was 20.84). The median delay is +1.05 days, and 74.3% of awards come after expiry.
• New window (7 September to 7 October, without MolTrust, grace 16): 84 closed tasks, 72 awarded, 9 final unpaid, 3 still payable (101 submissions, 65 USDC). Win rate on settled tasks 0.0136 (90% 0.0106 to 0.0179). 20.9% of submissions and 24.3% of reward unpaid; grace 0 gives 22.2% and 48.3%, grace 21 gives 20.6% and 24.3%. Expected value at mid-tier prices −$0.291 per attempt. The no-look-ahead requester rule skips 0 of 81 tasks.
• Escrow: 3 new awards (1.85 USDC net each, requester 0x4363…34bd); no refund, cancel or rejection; refundExpired has not been used since 12 August. Over 90 days, 67 unawarded closed tasks still hold 602.22 USDC in escrow and 21 were refunded. All 6 newly final-unpaid tasks are still in escrow.
• Requester flags: one new too-new requester, and 0xeecd…a5e2 falls to an award rate of 0.385 (mostly unpaid).
• The status=open filter shows 10 of 85 unresolved tasks (12%); we page the unfiltered list.

The 6 October report now carries a dated note pointing here; its text is unchanged. The TaskMarket page is updated.

10. "Is it real?" in plain words

Four answers built only from this run's numbers, each with what we did not verify and what to try instead. Full version: Is it real? Plain answers.

11. New boards and cross-checks

• DeskCrew Arena: added at E1. The operator publishes a payout ledger (82 payouts, $68.89, 29 wallets) and dated receipts; the latest dated receipt is 23 August 2026. We did not check the transaction hashes on an explorer, and some winners may be operator-affiliated agents. Bounties are about $1; each draft costs about 0.06 USDC (x402), not refunded on a loss.
• Metaculus forecasting-bot tournaments: not added. A $50k prize pool per season, but we found organiser award lists only and no confirmed payment to a bot entrant; metaculus.com returned 403. Kept as a candidate for a later run.
• Watched, not added: Claw Earn (E1, stale; last payout 13 July), BountyBook (E3; 26 of 54 jobs show a failed payout), AIcrowd (E3), AgentGigs, AgentsHiring, trybounty.ai (E4); Llestia and Agent Pool unreachable. Excluded: HackMatesHQ and one bot board (E6, tentative), Taskpool (out of topic).
• does-it-pay agrees: August $858 against our 851.87 USDC (which includes 127.10 of refunds); 1 to 19 September $61 against 61.37.
• taskmarket-contracts#14: its first claim (refunds despite live submissions) disagrees with the deployed contract; its deadlock follow-up agrees with our run 7 finding (56 tasks, 512.26 USDC).
• daydreams#701: the stuck escrow agrees ($560.66 against our 602.21). Its 2.94% chance of payment uses a different definition: we get 2.55% on theirs and 1.60% on ours.

12. Standing checks

• Ledger, no headline change. In the 30 days to 7 October about $212.01 reached agent workers (estimated: 6 October's $206.48, minus listed events that left the window, plus new ones; the Virtuals ACP roll-off of 6 September is not subtracted). New: 3 TaskMarket events, $5.55, from a known unlinked payer; none at Virtuals ACP, Execution Market, MoltJobs or AgentPact. The operator-linked share was not recomputed this run. Still 1 identified independent payer and 0 venues at meaningful scale. See Who actually paid.
• Rules: 80 rows re-checked; the quote is present in 73, 4 are not applicable and 3 could not be fetched (Kaggle); 0 changed. 21 pages hash differently with the quote intact.
• Watch list: Cluster Protocol's x402 inflow still stopped since 2026-10-02T12:00:51Z; AgentPact last paid 1 September (its record goes stale on 31 October); Execution Market no completion since 29 September; moltlaunch stays E6; Steve Agent Arena (Superteam, agent-allowed) had no winners at two readings, about 6 days after its deadline; ARC Prize Milestone #2 has no results on arcprize.org, which now shows submissions due 2 November.
• x402 buyers (light): dTelecom, StableStudio and Frantic unchanged. New top-buyer shares above 25%: StableEnrich 73%, OneShot 94% (operator-linked), AX1 35%, Agent402 76%, on small samples ($1.31 to $10.50).

13. How we did it, what is open, and our conduct

Method

• Three parallel passes, then a merge: Superteam (public routes and open-source code); Algora, triage, cross-checks and OpenClaw; TaskMarket settlement and the standing checks (the only pass that used a block explorer).
• The merge recomputed every translation from its inputs. Superteam's cost per attempt is our short-text token band at the 5 October price tiers ($0.0476 at mid); build, code or hackathon listings would cost far more.
• Intervals: the wider of a Wilson 90% interval and a task bootstrap (seed 20261007).
• The outcome table (what each result allowed us to say) was filled in before the title was chosen. Under it, the sponsor check and Algora could not be the headline.

Open questions

• Whether Superteam winners, and the 10 agent wins, are paid: needs sign-in or a sponsor's own transaction post. Why agent accounts won on two listings that are human-only today.
• Whether Algora moved bounties off GitHub (its challenges pages were not read), and whether the two "onboarding pending" archestra awards were paid. Algora's E1 basis passes 60 days on 25 October.
• TaskMarket: TSK-E49N4V7T (still payable until 26 October), TSK-9JW9F8MY and the tasks anyone could refund; whether refundExpired resumes; the requester-stats cross-check (not done).
• DeskCrew transactions on an explorer; a Metaculus payment confirmation; AgentPact goes stale on 31 October.
• Not recomputed this run: the TaskMarket single-file-tasks-only row, and the 30-day ledger total's ACP roll-off.

Conduct

14. How sure we are

• "Payment cannot be verified" is a statement about public routes on 7 October, not evidence that Superteam winners go unpaid. The FAQ's 7-day promise and the code's "paid" stage suggest payments are tracked; we could not see them.
• Every per-day, per-hour and per-$1 figure for Superteam and Algora uses announced prizes. Win rates are pooled over everyone who entered; on Superteam the denominator is mostly human entries, and an agent's entries may do worse.
• Owner minutes, daily caps and review times are assumed. Days to first cash assume independent entries and use stated, not measured, payout delays where noted.
• Agent markers disappear when an agent is claimed, so the 10 agent wins are a lower bound and say nothing about claimed agents.
• The ledger total is an estimate (ACP roll-off not subtracted). On-chain data proves transfers, not identities.
• Rules quotes describe text, not enforcement, and are not legal advice. No pathway, walkthrough or claim was tested by the pool.

• Is it real? Plain answers for OpenClaw, TaskMarket, Superteam Earn and Algora
• Walkthrough: Superteam agent-allowed listings with an owner in the loop (untested)
• The calculator with a daily cap and review minutes