The “weather bot” proof wallets: every fill, reconciled.
The trend, and what it offered as proof
The pitch is everywhere on X right now: NOAA and ECMWF publish accurate short-range forecasts for free; Polymarket’s temperature-bucket markets are priced by people who haven’t read them; a bot that buys the gap prints money while you sleep. The guides come with install commands, config blocks — and proof wallets.
Five proof exhibits appear across the most-viewed posts. They resolve to four distinct on-chain wallets (one wallet is cited twice, under two different descriptions) plus one screenshot whose handle does not resolve to any Polymarket profile. For each real wallet we fetched the complete lifetime fill history — buys, sells, redemptions, open positions, maker fills included — and reconciled the viral number against the transaction record. The full machine-readable reconciliation is published at weather-bot-reconciliation.json.
The flagship: shown as +$80,749. The fills say −$2,418,918.
The trend’s most-viewed exhibit (188K+ views) is a profile screenshot: $80,749.71 all-time profit, captioned as a student’s AI-built weather bot. We fetched that wallet’s complete lifetime: 193,363 fills across 6,639 markets, November 2025 through July 2026.
| Leg | Detail | USDC |
|---|---|---|
| Cash deployed | 193,329 buy fills across 6,639 markets | −$5,683,307.38 |
| Sell proceeds | 34 sell fills — in the wallet’s entire life | +$20,878.16 |
| Redemptions | 5,035 resolution payouts | +$3,243,510.87 |
| Open positions | 1,669 holdings, all resolved worthless | $0.00 |
| Net result | complete lifetime | −$2,418,918.35 |
The buying was not small or brief: it escalated for months — roughly $1.74M deployed in April 2026 alone — then collapsed to $42K of buying in July. The wallet almost never sells; it buys temperature buckets and holds to resolution, market after market, collecting back 57 cents on each deployed dollar.
Where the +$80,749 comes from
The viral figure is the profile page’s displayed profit metric. That display is not a cash reconciliation: it has its own accounting for realized versus unrealized value, and it is not computed from the complete fill record shown above. We are not claiming the screenshot was fabricated — the display may genuinely have shown that number. We are claiming something simpler: the wallet’s own fills contradict it. Cash out minus cash in is negative $2.42M. A profile metric is not an audit trail.
We know this failure class from the inside: an earlier MSR Decode report built on this same API’s default settings produced a wrong accusation that we retracted in the trades-endpoint report. Every figure here uses the corrected fetch described in Section 06.
The double-cite: two exhibits, one wallet
The most-viewed how-to guide in the trend (399K+ views) lists three proof wallets. One is an anonymous address described as “+$39.3k for month” with an “80% win rate, pure weather strategy.” That address — 0x594edb…1c11 — resolves to the same wallet as the +$80,749 flagship in Section 02. The trend’s two most impressive exhibits are one wallet, and it is down $2.42M.
The “80% win rate” is worth a note of its own: a wallet that buys several temperature buckets per day redeems something almost every day. In this wallet’s record, 5,035 redemptions sit alongside a net loss of $2.42M. Frequent redemptions are not profit.
The supporting exhibits, reconciled
Actual: ≈ +$47,943 — and not “only weather”
Overstated ~1.5×
Genuinely profitable, and genuinely automated-looking: 13,937 fills across 1,169 markets. But weather is 41% of its activity (its single largest market family is esports match markets), and its monthly buy volume fell roughly 93% after April 2026. The figure is approximate because the redemption feed truncates at the API’s offset cap; we resolved the 91 uncovered held-to-resolution markets against each market’s on-chain winning outcome and credited winning shares at $1.00 — the reconstruction is itemized in the published JSON.
Actual: +$13,442.75 — and dormant
Overstated ~3×
12,691 fills, $472,248.32 deployed, 89% weather — a real weather bot, returning about 2.8 cents per dollar of turnover across its life. Its last substantive trading was July 2025. The article presenting it as a live example published in February 2026, about seven months later.
The one claim that checked out — and what it actually is
One author in the trend published his own wallet address alongside his claim: roughly $6.7K this month from Hong Kong temperature markets. Credit where due — the fills agree. Complete lifetime: +$7,885.54 on 8,115 fills across 536 markets since June 9, most of it in July. The claim is real.
What the fills show
6,676 buy fills averaging $7.10 each. 1,439 sell fills, 825 of them dumping positions below 15 cents — active loss-cutting. Maker rebates in the activity log — the wallet posts resting orders rather than crossing the spread. A median of 39 fills per market. At the market level: 69 of 135 closed markets net-positive (51%), with average wins about twice average losses. The profit is real and the structure is real — multi-bucket temperature ladders — but the execution is a high-frequency market-making operation with daily attention, not a set-and-forget bot.
Why the distinction matters
The accompanying guide sells a passive version: configure once, deposit, close your laptop. The wallet behind the claim does something almost none of its readers can replicate, in markets thin enough that its own July buying averaged about $1,255 per day — a few copycats of meaningful size would erase the spread being harvested. We are not presenting this wallet’s approach as a strategy to copy; we are recording that its published number reconciles and its published recipe does not describe its own fills.
Why checking this yourself gives the wrong answer twice
If you pull these wallets from Polymarket’s public trade feed the obvious way, you get materially wrong numbers, for two separate reasons we have now hit across multiple reports:
- The user-filtered feed truncates.
/trades?user=stops paginating at an offset cap; long histories silently end. The fix is per-market fetching: themarket=filter is honored server-side and uncapped, so we enumerate every market the wallet ever touched (from its trade slice, its open positions, and its redemption log) and fetch each one to the end. - The feed defaults to taker fills only.
takerOnly=trueis the silent default. For the ladder wallet in Section 05 that default hides 6,154 of its 8,115 fills — 75.8% — the entire maker side of the operation. Every fetch behind this page setstakerOnly=false.
Totals were then cross-checked against the wallet’s separate activity log in the window where both sources overlap — matching within boundary error (for the flagship: 5,104 events in both, $310.8K vs $310.4K) — and the buy/sell components were recomputed blind by an independent model working from the raw fill files with no access to our conclusions. The recompute matched to the cent.
What remains unverified
- Why the flagship’s displayed profit differs from its cash flows: we can state that the fills contradict the screenshot; we cannot state how the display metric produced its figure, because its internal accounting is not published.
- Whether any forecast-vs-price edge exists in temperature markets generally: this page reconciles the cited wallets; it is not a study of the market class. Two cited wallets did profit — modestly, and neither in the passive form the guides describe.
- Operator identity and intent: out of scope. We reconcile published numbers against public transaction records; we make no claims about who runs any wallet or why any post was written.