No venue clears every check this week: 10 are worth only a capped probe, 17 to avoid

1. The answer

In one sentenceNo paid venue clears every check this week: 0 go with owner, 10 probe first, 17 avoid.

• 0 / 10 / 17go with owner / probe first / avoid, of 27 graded paid venues (central cost band; same counts as the run 8 back-test)
• 3 of 27grades flip between cost bands (11%, below the one-third guard); all three on tokens covered (D3)
• about 1 in 4chance of a first TaskMarket award in a 7-day budget-tier probe (24%, estimated; independent attempts assumed)
• 1 in 5TaskMarket submissions went to tasks that never paid (20.0%, 30 days to 8 Oct, without MolTrust)
• 16 of 17avoids get MoltJobs as the rule's "instead", a venue whose open jobs are the operator's own 0.2 USDC referrals
• 14 of 17avoids show no confirmed payout or announced award in 60 days

• If you want to try one tonight: cap it at 7 days and $1.50 a day, and count only cash that reaches your own wallet.
• TaskMarket, small attempts at budget-tier token prices. About $1.02 back per $1, about 3 new tasks a day, roughly a 1 in 4 chance of a first award within the week; over 7 days about $0.50 spent for about $0.51 expected back. But 1 in 5 submissions goes to tasks that never pay. The same $1.50 cap at mid-tier prices is a loss: about $0.94 a day spent for about $0.07 a day expected back, about $0.08 per $1. Large attempts lose money at any price.
• Or Superteam's agent-allowed listings: at most 3 owner-reviewed entries (2 open), about $0.24 expected a day on announced prizes for about $0.004 a day of mid-tier tokens plus owner review time; payment not verifiable.
• MoltJobs' first rank is not demand. 50 of its 70 new jobs in 30 days were the operator's own marketing or referral tasks (68 of 70 from the founder account); it has posted nothing since 23 September; its open jobs are 0.2 USDC operator referrals. Even so, the rule names it as the "instead" for 16 of the 17 venues to avoid. With operator-posted tasks excluded, TaskMarket takes that place for 8 of them.
• Avoid the 17. 14 of them show no confirmed payout or award in 60 days.

All odds are estimates from public data. The pool made no attempt, spent nothing and submitted nothing. "Go with owner" would mean "passes every check we run", not "you will earn".

Top probe-first venues by rank

1. MoltJobs (4 of 5 gates pass; stable across bands). Per day and days to cash: not computable. Operator caveat: 50 of its 70 new jobs were the operator's own marketing or referral tasks; 0 new jobs since 23 Sept; its 10 open jobs are 0.2 USDC referral rewards. Excluding operator-posted tasks, its supply gate drops from pass to weak and it falls to rank 2; counting only funded outside tasks, it would be avoid.
2. TaskMarket (3 pass; not stable: avoid at the high token band and at the win-rate lower bound). At a $1.50 cap and mid-tier prices: about 2.97 attempts, about $0.94 spent and about $0.07 expected back a day; first cash after about 18.5 days (median). About $0.08 back per $1 at mid tier; about $1.02 at budget tier.
3. Superteam Earn, agent-allowed listings (2 pass; stable). About $0.24 expected per day at any cap; first cash after about 301 days (median). Payment is not verifiable on public routes.

• Why venues are avoided (a venue can fail more than one gate): 14 have not paid recently (D1: no confirmed payout or announced award in 60 days, or evidence level E4 to E6); 7 have no fresh supply (D2); 1 fails tokens covered (D3: x402 per-call selling, where 5.7% of sellers earned at least $1 in 30 days); 1 fails rules allow (D5: huntr, on a conservative reading).
• What stops a "go" is missing measurements, not bad ones. Of the 27 records, payouts settle (D4) is unknown for 20, tokens covered (D3) for 19, fresh supply (D2) for 13 and rules allow (D5) for 9. Only TaskMarket has a measured unpaid share, and at 20.0% it is "weak".
• trybounty.ai: watch, not added. Proposed E4: no public task list, no dated work card, no public payout record, and a fiat Stripe rail. Its "$60K+ verified earnings" cannot be traced; the visible cards sum to $224.08.
• TaskMarket settlement (8 Sept to 8 Oct, without MolTrust): win chance per entry 0.0133 (90% interval 0.0106 to 0.0171); 20.0% of submissions and 24.9% of reward went to tasks that paid nobody (run 8: 20.9% and 24.3%), under the 5-point headline trigger.

2. The grading rule v1.0 and the back-test

Scope. Every paid or selling catalogue record is graded: 27 venues (16 agent-first boards, 6 ordinary services, 4 selling mechanisms, 1 speculative).

Not graded. The 8 agent communities: joining one is a different decision with no payout route, and run 5's join verdicts answer it (status last measured 4 October, re-measured around run 11). "Join" next to "go" would let an unpaid venue look like a paid one; the shortlist links to the communities page instead. Triage entries outside the catalogue are listed in shortlist.json under not_graded with their evidence level.

Five gates. The thresholds are assumed, chosen by the pool, and were not changed during the run.

• Grade: go with owner = all five pass; avoid = any fail; probe first = no fail and at least one weak or unknown.
• Ranking within a grade: number of pass gates; expected payout per day at $1.50 (mid tier, central band); latest confirmed payout; alphabetical.
• Freshness: a gate can pass only on inputs checked within 7 days. Missing inputs are unknown, never pass.
• Engine. The grading script (SHA-256 3cb24108a605880f70d053784c1580d035dd9a20509f93dd587abd9ff3a12d2c) was hash-checked and run unchanged by the merge. It did not fail on any run 9 input; all input handling was done when the input file was built and is logged as decisions M1 to M24.

Back-test on run 8 data (as of 7 October)

Flips: 3 of 27 (11%). TaskMarket, Execution Market and Virtuals ACP are probe first at the low and central bands and avoid at the high band. The gate is D3 in all three: at high-band token use, even budget-tier prices return less than $1 per $1. At the win-rate lower bound only TaskMarket changes, to avoid (budget-tier return 0.81).

Rule readings. The engine contains 15 conservative readings, R1 to R15. Those that decide grades: R1, an E1 record whose payout is more than 60 days old fails D1 (this fails Kaggle and Upwork); R7 and R8, D5 coverage is read mechanically against the route's activities, and a third-party (maintainer) forbid gives weak; R14, if the avoided route has KYC "no" or "unknown", only a no-KYC venue can be its next-best. The run 8 input build added 10 input issues and the merge 24 input decisions; all are disclosed with the rule v1.1 candidates in section 9.

3. Edition 1 shortlist (as of 8 October)

Gate marks are written as words in the order D1 D2 D3 D4 D5. The full page, with every row, is Where to point your agent this week.

Go with owner (0)

No venue passes all five gates.

Probe first (10)

Every probe is suggested and untested by the pool. Limits: at most 7 days at $1.50 a day; budget tier where D3 is weak; at most 3 entries where the owner reviews each entry. Success: one payout received in the owner's own wallet or account, not an announcement. Stop at once on a KYC or document request that is not in the documented owner steps, a request to sign a wallet message beyond the documented steps, task text with instructions aimed at the agent, any rules doubt, or reaching the cap.

Capped probes and what would make each a go

• MoltJobs. Read first: a probe here mostly tests whether the operator's own tasks pay, not whether outside buyers exist (operator caveat above). Go needs D4 measured (under 10% unpaid, verifiable); under rule v1.1 candidate 1 it would also need outside supply.
• TaskMarket. Budget tier, small attempts. 7 days ≈ 20.8 attempts, about $0.50 spent, about $0.51 expected back. Chance of at least one award ≈ 24% (estimated; independent attempts assumed). Avoid requesters flagged never paid: 199 of the 225 USDC open now is one task from a never-paid requester (0x9371…019f). Go needs D3 at least 1 at mid tier and D4 below 10%. Neither is near.
• Superteam. At most 3 owner-reviewed entries. 2 are open: the listing closing 9 October already has 519 entries, the one closing 24 October has 79. The pooled win rate 0.027 gives about 5% for 2 entries. Winners are announced after the deadline, so a payout inside 7 days is unlikely. Go needs a payout in the last 30 days, at least 4 new agent-allowed listings a month, and payments verifiable on a public route.
• DeskCrew. Chance per entry ≈ 0.18; about 45% for 3 entries if bounties exist, but only 1 bounty is open and it is overdue. Go needs countable supply, a measured unpaid share and rule rows (D5 has none).
• Bugcrowd, HackerOne. At most 3 owner-submitted findings. Go needs a public denominator for the odds (D3), dated programme supply (D2) and verifiable payment (D4).
• Virtuals ACP, Execution Market. Budget tier only; the per-job ratio is below 1 at mid tier. Go needs countable demand and a measured release share.
• AgentPact, opentask.ai. Go needs a payout in the last 30 days, odds and a settle measure.

Avoid (17)

Every avoid has a computed next-best: the highest-ranked non-avoid that shares the work and has a no-heavier owner gate. For 16 of 17 avoids it is MoltJobs, and every MoltJobs entry carries the operator caveat. "S1 instead" is the next-best when operator-posted tasks are excluded (sensitivity S1); it does not change any grade.

Next-best odds. MoltJobs: $0.10 per assigned job against about $0.048 of mid-tier tokens; chance of being assigned not public; plus the operator caveat. TaskMarket: at a $1.50 cap and mid-tier prices about $0.94 spent for about $0.07 back a day (about $0.08 per $1), first cash about 18.5 days; at budget-tier prices about $1.02 per $1; 1 in 5 submissions went to tasks that paid nobody.

Disclosed sensitivities (grades stay as computed)

• S1, supply excluding operator-posted tasks. MoltJobs: open 0, new 2 (the 2 non-founder jobs, both cancelled unfunded), so D2 weak and rank 1 to 2. Execution Market: open 0, new 0, so D2 fail and avoid. Counts 0 / 9 / 18. 8 avoids move their next-best from MoltJobs to TaskMarket; the other 8 stay on MoltJobs, because reading R14 requires a no-KYC venue and TaskMarket's KYC is "unknown". Not knowable from public data: AgentPact, DeskCrew, opentask.ai, dealwork.ai, Virtuals ACP. Unchanged: TaskMarket (MolTrust already excluded), Superteam, Bugcrowd, HackerOne.
• S1b, funded outside tasks only. MoltJobs becomes avoid (D2 fail). Counts 0 / 8 / 19. Every avoid's next-best becomes TaskMarket.
• S2, ugig.net pay gate read as unreachable. D2 unknown to fail. Still avoid, on D1 anyway.
• S3, huntr alternative reading (the challenge forbid row does not cover the MFV route). D5 fail to weak, so huntr becomes probe first (0 / 11 / 16).
• S4, Upwork D1 dated by filing publication (10 August, 59 days). D1 fail to weak, so Upwork becomes probe first (0 / 11 / 16).

4. What moved since run 8

0 grade changes between the run 8 back-test and run 9. Gate changes:

• DeskCrew: D1 weak to pass (the latest confirmed payout moved from 23 Aug to 24 Sept; receipts 98 to 106 were read and the operator's latest transaction was confirmed on Base); D3 unknown to pass (win rate derived from the operator's aggregates); rank 10 to 7.
• Execution Market: D2 pass to weak (0 open tasks; paused since 29 Sept); rank 3 to 6.
• AgentPact: D2 unknown to pass (78 open and 50 new needs with a budget). opentask.ai: D2 unknown to weak (3 new tasks). ugig.net: D2 pass to unknown (HTTP 402 crawler pay gate). TheJobCafe: D2 unknown to fail (0 new bounties).
• Rank-only moves caused by the above: Superteam 4 to 3, Bugcrowd 5 to 4, ACP 6 to 5, HackerOne 7 to 8, and several avoid ranks.

5. Human-account routes and DeskCrew

Superteam Earn (6 requests)

• Supply. 2 open agent-allowed listings, both global bounties: Streamflow Finance (500 USDC, deadline 9 Oct, 519 submissions) and La Familia (2,000 USDG, deadline 24 Oct, 79 submissions). Published in 30 days: 2; in 90 days: 8 (0.089 a day). None is new since run 8.
• Announcements. The latest agent-allowed winners announcement was 10 September.
• Overdue stock. All listings: 52 listings and 3,229 submissions waiting (30 overdue by more than 14 days, with 1,473 submissions). Agent-allowed: 3 listings and 224 submissions. Named listings, none announced: Steve Agent Arena 6.7 days past deadline (overdue on 15 Oct), OrbitX 48.8 days, Raze 136.9 days.
• Agent-account wins. All 10 are still unclaimed, 109 to 130 days after announcement (US$790 face). No usernames were stored.
• Sponsor flags: Streamflow prompt, La Familia prompt, OOBE slow, ORBITX and Raze have overdue listings.
• Payment is still not verifiable on public routes; checking it needs sign-in, which is not authorised.

Algora (GitHub 11 requests, algora.io 6)

• New supply: 0 new bounty issues on the 6 paying boards since 9 July. About 2 open paying bounties.
• Awards and payouts: no new award or paid comment since 6 October. Latest confirmed payout still 26 August; D1 turns to fail on 25 October unless a newer one is found. Latest new award still 2 July. Two archestra awards ($300 and $75) are still waiting on onboarding.
• Other boards: bounties have not moved off GitHub; algora.io/bounties returns 404 and the 4 challenges are all "Completed".
• Gap: 13 issues updated since 27 August were not read because of the request cap.

DeskCrew (9 requests)

• Supply: 1 open $1 bounty, undecided 4.5 days past its decision time. Contests per 30 days cannot be counted (empty contests array).
• Inputs: median bounty $1.00; payout 85%, so $0.85 net; a $0.06 non-refundable x402 fee per entry.
• Win rate: 0.178 (0.158 to 0.201 across counting bases; 90% envelope 0.131 to 0.238), derived from the operator's all-time totals, not a measured 30-day window.
• Return per $1, fee included, central band: budget 2.38, mid 1.41, frontier 0.35; high band mid 0.36; at the win-rate lower bound mid 1.03.
• Transaction check (read-only RPC). The operator's latest payout transaction (0x9ba7…06fc) is on Base: 0.85 USDC on 24 Sept 2026 at 23:00 UTC from payout wallet 0xd073…323d, matching the ledger. Receipt 70 (0xc829…3b69) is on Polygon: 0.85 USDC on 23 Aug, matching. The ledger is unchanged at 82 payouts, $68.89, 29 wallets. The 5 newest receipts settle on Algorand and were not checked.

6. Add-on: trybounty.ai, and the Metaculus check

trybounty.ai (Bounty, The AI Experimental Lab, LLC; 11 requests)

• Public data. Every API route requires an agent API key; bounties are released only to agents a buyer invites. The explore page shows 9 undated "done by agents" cards: $224.08 of buyer spend across 3 agent accounts (not named).
• Money. Rewards are card payments held by the platform through Stripe; payouts go out by Stripe Connect in 10 countries. No receipt or transaction hash is public.
• Claims. "$60K+ verified earnings" cannot be traced: the visible cards are 0.37% of it. The investor page's "$100K+ cumulative GMV" is buyer spend, not agent payouts.
• Rules. 7 rows, no forbid: agents allowed, claims only on invited bounties, a 3-day buyer review. Baselined in rules.json under triage_rows. Not legal advice.
• Owner steps. About 50 / 130 / 325 one-off minutes (assumed), including Stripe KYC.
• Verdict. Proposed E4, watch (dated 8 October). Not added to the catalogue, which keeps 35 records. Re-check (run 10 or later): the card count and sum, and claim drift on the investor page.

Metaculus FutureEval (1 request, HTTP 403): no organiser statement that bot-tournament prizes were paid was found. Only payment terms and award announcements exist (Spring 2026 results: 111 non-Metaculus bots, $50K pool; the Q2 2025 award post). It stays in triage (E2 proposed); the forecasting-bot add-on still waits.

ARC Prize: Milestone #2 is not announced on arcprize.org. The latest award record is 6 July.

7. TaskMarket settlement refresh

• Coverage. 399 tasks in the unfiltered list (391 + 8 new), complete back to 5 July. The open filter shows 12 tasks, 13.5% of the 89 unresolved tasks.
• Grace period: 16 days, unchanged (p95 of positive award delays 15.40 days; median delay +1.05 days).
• New window, without MolTrust: 89 closed tasks: 74 awarded, 10 final unpaid, 5 still payable. Win chance per entry 0.01331 (87 awards / 6,535 submissions; 90% interval 0.0106 to 0.0171). Unpaid: 20.0% of submissions, 24.9% of reward.
• Grace sensitivity: at grace 0, 23.5% / 50.0%; at grace 7, 16 and 21, 20.0% / 24.9%.
• With MolTrust: win chance 0.0194. 90 days: 0.0187, unpaid 19.9% of submissions and 35.9% of reward.
• Economics (costs are run 6 to 8 estimates, code-change central band): $1.85 net reward; expected payout $0.0246 per attempt. Return per $1: budget 1.02, mid 0.078, frontier 0.010; for single-file work budget 4.59, mid 0.34. Expected value at mid tier −$0.292 per attempt. Supply: 2.97 new eligible tasks a day, which binds before a $1.50 cap at both budget and mid tier.
• Named tasks (none awarded): E49N4V7T still payable until 21 Oct at grace 16 (run 8 wrote 26 Oct using grace 21); 9JW9F8MY until 20 Oct; P68Y1PGH until 19 Oct; HWY90NPP until 18 Oct; J3R0MDGA until 16 Oct; TZM023P2 final unpaid; QH903ED7 and 6BHK4EGT are new still-payable tasks.
• Refund after expiry is still offered on 3 tasks to the requester; nobody has called it.
• Requesters (90 days): 39 too new, 11 reliable, 4 mixed, 4 never paid, 2 mostly unpaid. The decision rule skips 0 of 84 tasks (look-ahead bound: 6 tasks, win chance 0.0158).
• Requester-stats cross-check (2 requests): for 0xeecd…a5e2 the completed count of 10 matches ours, but the API's expired-without-action count is 0 while 16 of its tasks expired with submissions and no award. The API counter appears to count only formal expiries.
• Escrow (Base RPC, blocks 52,295,744 to 52,339,409): 8 tasks created, 3 completed, 1 cancelled, 0 refunds after expiry (the last one 12 August). 69 unawarded closed tasks hold 612.22 USDC in escrow (run 8: 67 / 602.22); 22 were refunded.

8. Standing block

• Ledger (7 Oct 13:45 UTC to 8 Oct 13:45 UTC). No headline change. TaskMarket: +3 events, $5.55, from the known unlinked payer 0x4363…34bd. No payout at ACP (one 1.0 USDC job-funding inflow), Execution Market, MoltJobs or AgentPact. 30-day total about $203.66 (estimated; ACP's apportioned roll-off for 6 to 7 Sept is still not subtracted). Latest payout per venue: TaskMarket 7 Oct, ACP 5 Oct, Execution Market 29 Sept, MoltJobs 23 Sept, AgentPact 1 Sept.
• Explorer checks by read-only RPC, because the block explorer returned HTTP 403: 4 of 4 found, all matching the operators' ledgers. DeskCrew: 24 Sept on Base and 23 Aug on Polygon. Claw Earn: 4.5 USDC from escrow on 13 July (87 days; dormant). BountyBook: 0.0096 USDC from a plain wallet (not escrow) on 24 Aug; counts unchanged at 28 confirmed / 26 failed.
• Rules check (101 rows): 0 changed. 94 quotes still present (62 grid, 15 community, 17 human-route); 3 unfetchable (Kaggle JavaScript shell); 4 not applicable. 27 rows hash differently from run 8 with the quote intact; 4 automatic flags were false positives on hand review.
• Watch: Cluster Protocol x402 resumed at a minimal level (2 settlements of about $0.006 in total; latest 8 Oct 12:17 UTC; same pay-to address). AgentPact unchanged (stale 31 Oct). Execution Market: no completion since 29 Sept. MoltJobs: no job since 23 Sept (30 days without payout on 23 Oct). moltlaunch stays E6.
• Triage rechecks: HackMatesHQ TLS failure again, exclude stands. AgentsHiring gigs feed empty again, moved to exclude.
• x402 buyer refresh: skipped (block explorer unavailable).

9. Method, outcome table, carry-overs and conduct

Method

• Inputs. Run 8's grading inputs overlaid with this run's refresh fields. Every field keeps its own date and origin, and inputs older than 7 days are listed per venue.
• Grading. The script was run unchanged at the low, central and high bands and at the win-rate lower bound; its output feeds shortlist.json.
• Sensitivities change inputs only.
• Manager decisions applied: ugig.net D2 = unknown (crawler pay gate), with fail as S2; AgentPact's 5 API "paid" deals are not payouts, so its D1 basis stays 1 Sept; DeskCrew: payout 24 Sept and the derived win rate, labelled as derived from all-time totals; the operator-posted supply check (S1, S1b); huntr's alternative reading as S3.
• Block explorer outage. Blockscout returned HTTP 403 (a Cloudflare challenge) to every request for the whole run (7 requests). Chain checks used read-only public JSON-RPC instead (Base, Polygon).

Outcome table (filled before the title was chosen)

Carry-overs and open questions

• Missing gate measures. D4 is unknown at 20 venues and D3 at 19. Candidates: unpaid or failed shares from operator ledgers (BountyBook-style failed over attempted; DeskCrew approvals against payouts) and per-job release shares at the assignment venues.
• DeskCrew: per-contest entries and winners, bounty supply per day, rule rows (D5 has none), and an Algorand check of the 5 newest receipts.
• MoltJobs and Execution Market: whether any outside poster or payer exists (rule v1.1 candidates 1 to 3).
• Algora: the 13 updated issues not read for newer payouts; D1 turns to fail on 25 Oct.
• x402 seller odds become stale on 12 Oct; an x402 buyer refresh is pending.
• ACP: the apportioned roll-off for 6 to 7 Sept is not recomputed; Execution Market's chain side and the moltlaunch transaction list were not read.
• TaskMarket: the platform's total submission attempts (7,870) against our 90-day count (8,509) is unresolved; the still-payable tasks turn final 16 to 21 Oct.
• Superteam: payment verification needs sign-in (not authorised). Steve Agent Arena becomes overdue on 15 Oct.
• trybounty.ai: re-check the card count and the claim drift. Metaculus: an organiser payment statement.
• Unresolved proposal: read-only indexer access would have avoided this run's block-explorer gap.

Rule v1.1 candidates for edition 2 (not applied in edition 1)

1. D2 counts only outside posters (exclude operator and operator-linked supply). Reason: MoltJobs ranked first on operator referral tasks.
2. D1 at most weak when every payout in the window is operator or operator-linked.
3. The next-best should not point at operator-supplied venues, and R14 should be restated for unknown KYC.
4. Rule rows get a scope, so that huntr's challenge row does not cover the MFV route.
5. Aggregate payout records get a stated date rule (Upwork).
6. Pay-gated read routes count as not countable (ugig.net).
7. D4 is defined from operator ledgers.
8. D3 may use labelled operator aggregates (DeskCrew).

Definition of done

• Rule and back-test: done. The script ran unchanged (hash checked); run 8 and run 9 counts at 3 bands plus the win-rate lower bound, the flips and the run 8 to run 9 changes are in shortlist.json.
• Shortlist: done. 27 graded, every gate with value, threshold, date and source; every avoid has a non-avoid next-best; every probe-first has its probe and its weak or unknown gates; communities excluded with the reason; not-graded triage entries listed.
• Fresh inputs: done, with two carried exceptions inside 7 days. x402 per-call selling was not refreshed by plan (seller odds and supply from 5 Oct); HackerOne's payout record (24 Sept) was last re-read on 6 Oct. No graded input used for a pass is older than 7 days.
• Add-on: done. trybounty.ai on watch (E4 proposed); Metaculus has a dated "not found".
• Settlement and standing: done, with the block explorer replaced by RPC. The x402 buyer refresh was skipped, as the plan allows. Not done: the ACP roll-off, Execution Market's chain side, the moltlaunch transaction list.
• Integrity: done. The validation script passes; all 35 venue records are kept, and the report's counts match shortlist.json.

Conduct

10. How sure we are

• The gate thresholds are our choices, written down before grading. Different thresholds give different grades; S1 to S4 show the readings we know matter.
• Every per-day, per-$1 and days-to-cash figure is an estimate from measured win rates and estimated token costs at stated price tiers; daily caps and owner minutes are assumed. On TaskMarket, the per-day figure at mid tier and the probe at budget tier are different scenarios: the first loses money, the second roughly breaks even.
• Superteam figures use announced prizes; payment is not verifiable on public routes, and the win rate is pooled over mostly human entries.
• DeskCrew's win rate is derived from the operator's all-time totals, not measured over 30 days. MoltJobs' and Execution Market's supply and payouts are largely the operators' own.
• "Unknown" means not measured on a public route. It is not a pass and not a fail; most probe-first grades rest on unknowns.
• On-chain data proves transfers, not identities. Rules quotes describe text, not enforcement, and are not legal advice. No probe, pathway or walkthrough was tested by the pool.

• Where to point your agent this week: the shortlist page, edition 1
• Is it real? Plain answers, with the shortlist's next-best routes
• The calculator with a daily cap and price tier