Resolved YES if, within the 14-day market window, there exists at least one 1-hour window in which ≥10 qualified sensors from the sensors.social Cyprus mesh each report PM10 > 100 µg/m³ (coarse dust at breathing height).
Qualified sensor pool (frozen at market open, 19 devices): all sensors.social PM10-reporting devices inside Cyprus (lat 34.5–35.7, lng 32.0–34.6) with ≥14 days continuous uptime. New sensors added during the market window do not affect resolution.
Data source: sensors.social — citizen air-quality mesh on Robonomics. https://sensors.social
Historical baseline (April 14–22, 2026): six oracle-triggered hours, all on April 18; peak 14 of 19 sensors simultaneously.
Full spec & reproducible chart: https://github.com/ensrationis/tamper-proof-weather-oracle
Resolution by market creator based on the public sensors.social data, audited against the frozen sensor pool. Any disputes can be independently re-derived from raw data.
1,000🏅 Top traders
| # | Trader | Total profit |
|---|---|---|
| 1 | Ṁ31 | |
| 2 | Ṁ4 | |
| 3 | Ṁ2 | |
| 4 | Ṁ0 |
# Final report — 14 of 14 · resolves NO
Frozen pool: 19 sensors. Of 349 observation hours, oracle triggered 0. Peak simultaneous over-threshold: 3 of 19 — reached on 4 separate hours, never higher. With ~8h to close, CAMS forecasts surface PM10 4-7, dust 0. Resolution locked.
The trigger rule discriminated cleanly across all three regimes seen:
| Regime | Triggered hours | Peak |
|---|---|---|
| Storm 1 (pre-market, localised) | 0 | 7/19 |
| Storm 2 (pre-market, basin-wide) | 6 | 14/19 (74%) |
| Market window (no event) | 0 | 3/19 |
| Wind/rain washout 3–4 May | 0 (anti-dust) | — |
Each weather regime got the answer it deserved. The threshold (10 of 19 sensors > 100 µg/m³ within one hour) is well-calibrated: never false-positive on local cooking smoke (peak 3/19), never false-negative on a real storm (Storm 2 cleared by a wide margin).
Robustness against single-source attacks: the two known noisy sensors in the pool (`4DjaKwvFCG…` balcony smoker, 4Et69cDhnvUqws… restaurant kitchen) regularly logged 100–400+ µg/m³ from purely personal sources. They never coordinated to push the network past 3/19 simultaneously. To force a YES dishonestly an attacker needs ~10 simultaneous personal sources owned by ~10 operators distributed across the island. Not feasible.
Cost: 0 (ran on existing sensors.social + Robonomics infrastructure).
Full report: https://github.com/ensrationis/tamper-proof-weather-oracle/blob/main/updates/2026-05-07-final.md
Spec stays live; the same rule will be retroactively scored against any future Cyprus dust event to keep building a sensitivity profile. Integrations welcome.
— Sergei
Day 12 of 14 — 2 days to close
Frozen qualified pool: 19 sensors (per the market description — sensors with ≥14 d uptime at open). The pool is committed as audit artifact: data/cyprus_oracle_qualified_pool.json.
18 of 19 sensors reported in the window (one offline 12 days)
0 oracle-triggered hours
Peak simultaneous over-threshold: 3 of 19 — well under the quorum of 10
Notable atmospheric event 3→4 May: SW wind storm (peak gust 103 km/h, 15.8 mm rain). South-westerlies = clean ocean air, not Saharan dust. The storm washed out the column. Ground network
registered the cleanest hourly medians of the entire window in the hours after.
CAMS forecast through close (~63h): AOD 0.09–0.17, surface PM10 4–7, dust 0, 0 hours over 100.
For YES to resolve, a Storm-2-class event has to land in ~48h despite an explicit CAMS forecast against it. Increasingly improbable.
Full update: https://github.com/ensrationis/tamper-proof-weather-oracle/blob/main/updates/2026-05-05.md
Day 9 of 14 — status update
Numbers
Cyprus mesh active sensors: 22 (was 19 at market open — three new devices joined the qualified pool over the window)
Oracle-triggered hours so far: 0
Peak simultaneous over-threshold (>100 µg/m³): 2 of 22 sensors — far below the quorum of 10
Days elapsed: 9 of 14 · days remaining: 5
The network is calm. No basin-wide event has occurred since market open. A single sensor in Limassol (4DjaKwvFCG…) is the noisiest in the cluster (24 h avg 15.7, max ~390) — a known local outlier rather than a signal of any wider dust influx.
More: https://github.com/ensrationis/tamper-proof-weather-oracle/blob/main/updates/2026-05-02.md
Day 3 of 14 — status update
Cyprus mesh: 21 qualified sensors active, 0 oracle-triggered hours, peak simultaneous over-threshold = 2 of 21 (far below the quorum of 10). Network calm.
This week we also started publishing CAMS surface PM10 forecast alongside live sensor data on every weekly check (https://x.com/SensorsSocial/status/2048014995569467600). Snapshots saved week-over-week so anyone can compute CAMS-vs-ground accuracy later — directly relevant to whether a satellite indicator could pre-warn markets like this one.
CAMS 24h-ahead Cyprus surface PM10: ~20–24 µg/m³ haze (well below dust-event level). 11 days left.
Full update: https://github.com/ensrationis/tamper-proof-weather-oracle/blob/main/updates/2026-04-26.md
Prt sc ^ ^ - storm detected by robonomics platform sensors netrwork, April 18, five days before market open.
Same trigger rule applied retroactively — six consecutive 1-hour windows in which ≥10 of the 19 qualified Cyprus sensors recorded raw PM10 > 100 µg/m³. Peak: 14 sensors simultaneously (74% of the pool)