What has been tested
Every method is run on frozen test days and checked against known events before anything is shown on a map. Failures stay on the record.
GNSS degradation from ADS-B
Data: 20 worldwide days: 9–25 Oct 2025, 11–12 Nov 2025 and 22 Sep 2026.
| Check | Result | Verdict |
|---|---|---|
| Known hotspots appear | Southern Black Sea off Turkey 37–59% of aircraft-hexes degraded; Gulf of Finland and the Estonia–Russia border 30–41%; Kaliningrad 13–50%; Caucasus 14–39% | Pass |
| Quiet controls stay quiet | Control regions near 0% | Pass |
| SBAS outage from aircraft alone | The 13–17 Oct 2025 EGNOS outage reconstructed to within an hour at both ends | Pass |
| Storms separate from jamming | G4, 12 Nov 2025: 38.5% of capable aircraft below 10 m against a 3.7% baseline; G2, 18 Oct: 12.4%. The 93 m measure stays flat in both | Pass |
| Minute-level timing | Bay Area, 22 Sep 2026: bursts at 18:00–18:08Z and 19:52–20:12Z. An FAA flight advisory (PMSRCA GPS 26-03) scheduled a Point Mugu Sea Range GPS test that day 18:30–20:29Z, reaching the Bay Area: the second burst falls inside it, the first starts 22 minutes before it | Pass |
| Announced GPS test (FAA flight advisory WSMRNM GPS 25-16, White Sands, Oct 2025) | Near the range, 10–27% of 10 m-capable aircraft fell below 10 m from 03Z on 7 of the 10 scheduled days, against about 2% at other hours; none of the 7 unscheduled days showed it. Onset matched the 03Z start every time | Pass |
| Faulty avionics removed without removing jammed flights | The peer test drops aircraft bad among healthy neighbours only; hotspot shares change by at most 1.8 points after the post-exposure tag | Pass |
| Nightly bursts over central Europe explained | The post-exposure tag cut them from 33 to 26 events and shrank the peaks, but catches only 38–47% of the aircraft in the bursts that remain | Partly |
A finding: downstream effects by aircraft type, and what's behind it
Over 17 days, A350-900s were 24.9% of aircraft-days still degraded after leaving interference, against 1.3% of traffic (19.4×). A380s were 19.2× and 777-300ERs 4.7×; most narrowbodies were near 1×.
Tested on 23 Oct 2026, this looks like a reporting effect. In 19 days of raw traces over the southern Baltic, A350s at cruise altitude stayed visible on 89% of their crossings on jammed days, while narrowbodies went dark on half of theirs; 74% of the positions the A350s kept sending were flagged as degraded. On quiet days every type was seen about equally, yet A350 degraded episodes lasted a median of 8 minutes against 2 for other types. A documented behaviour fits: Australia reported to an ICAO regional meeting in June 2025 that on A350s and A380s the ADS-B accuracy value is set to zero when the inertial reference unit flags a single satellite as degraded, even when the position is valid (CNS SG/29 IP/05). So the type pattern most likely shows how these aircraft report accuracy, not lasting harm. Next: count A350 and A380 accuracy-only drops apart and rerun the validated days.
These cases as charts · Full validation report
A year of data (1 Oct 2025 – 21 Oct 2026)
The daily pipeline was backfilled to 1 October 2025, so leads from the test days could be checked against a whole year. Some checks also used the raw adsb.lol traces, following each aircraft through an area.
| Check | Result | Verdict |
|---|---|---|
| Was 23 Oct 2025 unusual for the southern Baltic? | 12.9% degraded, 87th percentile of 383 days: a bad day, not an unusual one | Closed |
| What happens on the worst Baltic days? | On 9 May 2026 about as many aircraft crossed the southern Baltic as a week later (293 against 326), but three crossings in four lost their ADS-B position, against one in twenty. 2 Jan, 1 Feb and 24 Feb 2026 show the same, a little weaker. Strong jamming removes aircraft instead of marking them degraded, so the map now has a "Fewer aircraft than usual" view | Found |
| Why is the Baltic worse at night? | Over northern Europe the worst hour is 02Z or 03Z on 369 of 383 days. It isn't the altitude mix, and only partly aircraft arriving already degraded: aircraft that arrived healthy are degraded 3–10 times more often at night than by day | Partly |
| Malay Peninsula onset cluster | Never above 0.9% degraded on any day: no local interference | Closed |
| Kurils / Fukuoka FIR, Jul–Aug 2026 (pilot reports via OPSGROUP) | adsb.lol barely covers the area (median 2 aircraft-hexes a day) | Can't test |
| Jammertest 2026, Andøya, as a known positive | No aircraft degraded at any hour of the test week, including the 50 W windows | No effect |
| Spire's maritime jammer case, 14 Oct 2024 | Not distinguishable from the days either side. Spire counted 3 of 53 aircraft affected, below what these methods resolve | Missed |
| False look-alike: Kiruna weather-radar fault and exercise chaff, 5–6 Mar 2024 | No GNSS degradation over the area (thin coverage) | Pass |
| SBAS flags over India and Australia | Flagged on 162 and 229 of 344 days because most aircraft there don't use those services. Listed apart on the map as routine | Explained |
| Radio-horizon localizer, scored against a 2025 TDOA network (Gattis, Cydejko & Akos 2026), settings fixed in advance | On the four days with published fixes, the top peak was 20, 37, 48 and 45 km from the day's site: 1 hit at 25 km. The sites always ranked in the top 2% of cells, but the mean miss (38 km) is no better than naming Kaliningrad city, and the two sites are only 21 km apart | Fail |
| Do A350s stay visible while degraded where other types go dark? (raw traces, 19 days) | On jammed days, A350s were seen on 89% of crossings at cruise altitude against 43% for narrowbodies and about 60% for 777s and 787s; the highest share on 7 of 8 full jammed days. Their degraded episodes also last 4–8 times longer, which fits a documented avionics behaviour | Confirmed |
| Localizer, third version (every approach direction weighted equally), against the second on five more held-out days | Third version within 25 km of a TDOA site on 3 of 5 days (mean 23 km; 5.5 km on one day), second version 2 of 5 (31 km). But a single fixed point between the two sites, 21 km apart, was within 25 km every day (23 km), so neither version beats a constant guess | Fail |
| Localizer, second version, on five held-out days chosen and frozen before download | Within 25 km of one of the two TDOA sites on 3 of 4 testable days (mean 23 km, against 38 km for naming Kaliningrad city). But every peak sits about 20 km east of both sites, so this is a steady pointer to the right coast, not yet a site fix | Partly |
GNSS spoofing from ADS-B (validated)
Spoofed aircraft report positions they can't physically have flown: near-zero ground speed at altitude, circles or a fixed point at altitude, jumps to another track and back. Tested on 22 days, Feb 2024 – May 2026, against public reporting of where and when aircraft were spoofed:
| Check | Result | Verdict |
|---|---|---|
| Places aircraft were shown match published spoofing points | Beirut airport within 1 km (early 2024), Amman 3 km (Oct 2024, up to 695 aircraft a day), Cairo 9 km, Smolensk 9 km, Moscow Sheremetyevo 2 km; Beirut gave way to Amman between the 2024 dates, as OPSGROUP reported | Pass |
| Independent check with multilateration (MLAT) positions | 90% of the episodes MLAT can test are confirmed; at least 43% of the positions MLAT shows as false are caught. Offset tracks with no jump are missed | Partly |
| Quiet controls (six regions, 25.6 million aircraft-hours) | 0.002% of aircraft-hours flagged; one unexplained cluster over northern Florida on 6 Mar 2024 | Pass |
| Held-out days with the method frozen (12 days by date rule, checks committed before processing) | All five checks passed: controls 0.002%; Middle East 0.39% and Baltic 0.81% of aircraft-hours (194× and 407× the controls); MLAT confirmed 91.5% of testable episodes; Beirut on 20 Feb and 20 Jun 2024, Amman on 20 Oct 2024 | Pass |
Version 0.5 (the default from late Oct 2026) also lets indicated airspeed, from air data rather than GNSS, qualify an aircraft for these tests. A spoofer controls GNSS ground speed, so an aircraft spoofed for its whole visible flight was never tested before. Over 41 days it finds 13% more episodes, mostly at Russian points the published lists don't name; the controls stay at 0.002% and all five held-out checks still pass, though two of those 12 days were seen while designing the change.
Spoofing and jamming are mostly separate: the Black Sea is mostly jammed, Egypt and Jordan mostly spoofed with clean accuracy values. Not on the public map yet: offset-track spoofing is missed, and a map layer needs Tom's go-ahead. Full spoofing report
L-band emitters from NISAR
| Check | Result | Verdict |
|---|---|---|
| Image method on the published Shiraz case (GPS World, 18 May 2026; jammer on 8 and 20 Jan 2026). The article's coordinate (27.32°N 52.87°E) is 257 km south of Shiraz, though it also says "about 50 km SW of Shiraz", so the known answer itself is uncertain | 56 detections over 6 passes, none within 50 km of the jammer; 21 on the date it was reported off | Fail |
| Processor tables on the same area, 40 passes June–October 2026 | Site in the swath on all 40 passes; no solid detection within 50 km. The jammer may have been off or moved, or the method misses it | Not confirmed |
| Broadband detector on an area with documented L1/L2/L5 jamming (30 days, 335 passes) | Solid detections on 12 of the 13 passes that covered it, 9 of them in the slice containing GPS L2. The tone detector saw nothing at the L2 centre there | Pass |
| False alarms on vegetation | Heavily forested regions with no reported jamming gave 1–4 detections each over 37–83 passes | Pass |
| Repeatability | Strong detections repeat from pass to pass within 9–24 km, and agree across ascending and descending geometries | Pass |
| Announced US GPS tests (pre-registered 26 Oct 2026, 16 sites) | 5 hits in 24 in-window frames (20.8%) against 93 in 507 control frames (18.3%); none of the hits at GPS L2. They fit other L-band emitters near the ranges | Fail |
| Known long-range radar sites (descriptive, not pre-registered) | Six published radars have a repeat site within 50 km: 3.3 and 13.2 km for sites seen from several orbit tracks, 19-39 km for sites seen from one; median about 17 km. Identities unconfirmed | Descriptive |
| Position accuracy against a surveyed emitter | Not yet tested | Open |
Since 28 Oct 2026 the repeat sites are on the map as leads, with this status beside them (how the layer works). Until a known emitter is confirmed and positions are checked against a surveyed one, treat every site as something to look into, not a finding.
GNSS interference from orbit (CYGNSS)
NASA's CYGNSS satellites listen to GPS reflections from the sea. The noise in each measurement, ahead of any reflection, rises when a jammer is in the antenna beam. Tested on six days in 2026 against the aircraft map:
| Check | Result | Verdict |
|---|---|---|
| Separates aircraft-jammed areas (10% or more of aircraft degraded) from clean ones (under 2%) | AUC 0.98 over 11,463 cell-days; at a 1 dB threshold fixed in advance, 98% of jammed cells caught and 16% of clean cells flagged (5% at 2 dB). The positives come from four regions, three-quarters from the Gulf | Pass |
| Quiet controls (central Pacific, South Atlantic, interior Brazil, Australia, southern Africa) | 0.8% of control cells read 1 dB or more | Pass |
| Held-out days: 21 more days chosen by date rule, method frozen, checks set in advance | AUC 0.974 (224 jammed, 39,713 clean cell-days); 98.7% of jammed cells caught at 1 dB, 15.8% of clean cells flagged; controls 0.7%; leaving any one region out, AUC stays at 0.959 or above | Pass |
| A full year, method frozen (383 days, descriptive) | AUC 0.974 over 746,753 paired cell-days; 97.7% of jammed cells caught at 1 dB, 16.1% of clean cells flagged. The March 2026 Gulf surge shows in both sensors (aircraft 63% degraded, CYGNSS 1.5–1.9 dB rising to 3.9 dB) | Holds |
| Tracks change over time | Persian Gulf: CYGNSS follows the spring rise in aircraft degradation (rank correlation 0.82 over 21 days), but from late June to October it stays at 2.8–3.8 dB while aircraft fall back to 1–5% degraded. Why is open | Partly |
| Why the Gulf stays raised after June: does aircraft spoofing persist? (check set before download, 7 days by date rule) | Aircraft spoofing in the Gulf was 0.000% of aircraft-hours before the March 2026 surge and 0.178% from July to October, present on all four days. It persisted, though it may not explain the whole CYGNSS level | Supported |
| Announced US GPS tests (36 sites, 39 days) | Overall not shown (AUC 0.57). White Sands and China Lake read high on about 40% of in-window passes against 5% outside, but strong tests are heard by neighbouring sites hundreds of km away, which blurs the comparison | Partly |
| Position of an emitter | Not tested. The beam is a few hundred km wide, so the map shows exposure, not a transmitter site | Open |
About 80% of the cells CYGNSS flags have no aircraft coverage: the Red Sea, Iran, Iraq, Myanmar, Sudan and the Arabian Sea. It covers only about 38°N to 38°S, so not the Baltic. Its satellites also stop returning data for minutes at a time over Iraq, Iran and the Levant; the cause isn't confirmed.
GPS interference from orbit (Sentinel-6A's own receiver)
The Sentinel-6A satellite (66° inclination, about 1,340 km up) tracks GPS for its own navigation. When a jammer is in view, every link it tracks weakens at once. Each second, Noisefloor compares each link's signal strength with what the receiver usually logs at that elevation, and takes the median across links. A cell about 300 km across is scored by how often the receiver lost strength while it could see the cell, from up to 3,800 km away. Method and pass marks were fixed on six days, then tested on 21 held-out days against the aircraft map:
| Check | Result | Verdict |
|---|---|---|
| Separates aircraft-jammed cells from clean ones (21 held-out days) | AUC 0.956 (407 jammed, 16,874 clean cell-days); 0.957 on the six tuning days | Pass |
| Leaving any one region out | AUC stays at 0.918 or above (Eastern Europe holds 348 of the 407 jammed cell-days) | Pass |
| Quiet controls (central Pacific, South Atlantic, interior Brazil, Australia, southern Africa) | 0.0% of 6,216 control cell-days at or above the threshold | Pass |
| Agreement with CYGNSS where both judge | Rank correlation 0.504 over 76,369 cells | Pass |
| Tracks change over time (Persian Gulf, spring 2026; reported, no pass mark) | Gulf cells averaged 0.201 on the spring days and 0.207 on the winter days: Sentinel-6A rates the Gulf raised all year and does not show the spring rise the aircraft and CYGNSS saw | Not shown |
| Rises on the days aircraft see interference, within the same region (events check, 21 days, set in advance) | 59.8% of 224 regions scored higher on their event days (mark 60%; sign test p = 0.002): borderline. It places jamming regions (38% of aircraft observations raised against 15% of random land points) but adds little day to day | Weakened |
| Position of an emitter | Not tested. A cell's score draws on everything within 3,800 km, so the map shows regions, not transmitter sites | Open |
It reaches 66°N, so it covers the Baltic, which CYGNSS can't see. Myanmar is weaker (own AUC 0.80).
Full Sentinel-6A validation report
GPS interference from orbit (COSMIC-2 receivers)
The six COSMIC-2 satellites (24° inclination, about 550 km up) track GPS on two navigation antennas each, one looking forward along the track and one backward. UCAR, which runs the mission's data centre, computes an interference index for each antenna every second. A cell about 300 km across is scored by how often that index was raised while a satellite could see the cell, from up to 2,500 km away, each antenna counting only its own side. Method and pass marks were fixed on six days, then tested on 21 held-out days against the aircraft map:
| Check | Result | Verdict |
|---|---|---|
| Separates aircraft-jammed cells from clean ones (21 held-out days) | AUC 0.919 (200 jammed, 13,092 clean cell-days); 0.922 on the six tuning days | Pass |
| Leaving any one region out | AUC stays at 0.920 or above | Pass |
| Quiet controls (central Pacific, South Atlantic, interior Brazil, Australia, southern Africa) | 0.0% of 6,216 control cell-days at or above the threshold | Pass |
| Agreement with CYGNSS where both judge | Rank correlation 0.265 over 76,369 cells | Pass |
| Tracks change over time (Persian Gulf, spring 2026) | Gulf cells averaged 0.333 on the spring days and 0.273 on the winter days (+0.059 against a +0.05 mark): the spring rise shows, narrowly | Pass |
| Rises on the days aircraft see interference, within the same region (events check, 21 days, set in advance) | 46.2% of 171 regions scored higher on their event days (sign test p = 0.86): it places jamming regions (39% of aircraft observations raised against 19% of random land points) but doesn't track which days | Fail |
| Position of an emitter | Not tested. A cell's score draws on everything within 2,500 km, so the map shows regions, not transmitter sites | Open |
It covers about 45°S to 45°N, so not the Baltic. It is strongest over Myanmar (own AUC 0.97), where Sentinel-6A is weakest. Only three to five of the six satellites delivered data on the test days.
Full COSMIC-2 validation report
L-band radars from orbit (SMAP)
NASA's SMAP satellite listens in the protected 1400–1427 MHz band and flags each look that carries interference. Long-range air-surveillance radars transmit just below the band and can spill into it. For each 36 km cell, Noisefloor counts the share of looks flagged over 7 days and how far it stands above the cells 100–300 km around.
| Check | Result | Verdict |
|---|---|---|
| US long-range radar sites from a 2010 FAA file (31 sites, one week, set in advance) | AUC 0.502: cells near the sites were flagged no more often than cells far away (6.7% against 7.1%) | Fail |
| North Warning System long-range radars (14 active sites, two unseen weeks, set in advance) | AUC 0.988: cells near the radars were flagged on a median 80% of looks, far cells 2%; 10 of 14 sites heard in both weeks; 1.2% of far cells flagged half the time or more | Pass |
| Crowded regions | Most looks are flagged across Europe, western Russia, East Asia and much of the Middle East, so single emitters can't be told apart there | Partly |
| GPS jammers | Not tested and not expected: GPS uses other parts of L-band | Out of scope |
The second check was not blind: the idea came from always-flagged cells noticed in Labrador. The site list is complete, the weeks were unseen and the pass marks were committed first.
Ground reference stations
Permanent GNSS reference stations log every satellite's signal strength. Six European stations 110–486 km from the Kaliningrad coast were tested on six jammed days and two quiet days, with the rule fixed before any download.
| Check | Result | Verdict |
|---|---|---|
| Do stations near the southern Baltic see the interference that aircraft see? | No flagged minutes on any of 48 station-days, while aircraft within 150 km were degraded for 1,945 minutes. A ground antenna sees a coastal jammer only within about 35–60 km, against about 410 km for an aircraft at cruise | Null (expected) |
The detector itself works: it caught a local drop at one Latvian station on 16–17 May 2026 that its neighbours didn't see. Full ground-station report
C-band interference from Sentinel-1 (prototype)
Sentinel-1 radar products carry the processor's own reports of radio interference in each burst. Read alone, without the imagery, they show where C-band emitters such as military radars were heard.
| Check (set before any area was run) | Result | Verdict |
|---|---|---|
| Quiet control (Gibson Desert, Western Australia) | 56 products, 3,276 bursts, none flagged, no sites | Pass |
| Sites line up with radar locations published by Bellingcat (2022): Rostov-on-Don, Pogonovo/Liski, Negev | A site lay within 30 km of the published point in two areas (16 and 25 km), but the points themselves ranked low within each busy area; Pogonovo/Liski was quiet. Reports place an emitter only to tens of kilometres | Fail |
Region-scale interference shows clearly: 82% of bursts were flagged over the Gulf, 21–35% in the two busy test areas, none in the control. Site positions stay private.
Worldwide, four weeks (2–29 Oct 2026; descriptive, not a test). Every IW product of each day, 563–1,000 a day, scored by region (cells about 110 km across):
| Measure | Result |
|---|---|
| Cells with four or more passes | 13,437 of 17,856 covered (75%) |
| Median share of passes with interference, known regions | Persian Gulf 1.00, Levant 0.83, Iran (interior) 0.60, Red Sea 0.60, South China Sea 0.57, Iraq 0.56, Egypt / Sinai / Suez 0.37 |
| Median share, quiet controls (central Pacific; Brazil, Australia and southern Africa interiors) | 0.00 in each |
| Against the aircraft map, cells with 140+ aircraft over the same 28 days | AUC 0.747 (47 jammed, 3,717 clean); median share 0.48 jammed, 0.00 clean |
The map's "C-band radar activity" layer shows these 28-day shares. It shows radars and other C-band transmitters by region, not GPS interference and not transmitter sites.
Interference spokes in weather radar (prototype)
A transmitter in a weather radar's band paints a thin line along one bearing. One day of open German and Finnish radar data (23 Oct 2026, 29 radars):
| Check | Result | Verdict |
|---|---|---|
| Finds a known source: the sun | Found in 43 of 54 sweeps with the sun in the beam, to within 0.23° of its computed bearing; 8 of the 11 misses were in rain | Pass |
| Detections are real spokes, not weather | Of 36 random clear-air detections reviewed by eye, 31 were spokes, 4 ambiguous and 1 rain (one reviewer, not blind) | Partly |
| Locates an emitter by crossing bearings from several radars | One three-radar crossing, against 2.7 expected by chance | Not yet |
Graded events (events v1.1, rules not yet validated)
What the methods above observe in one place and time window is joined into an event, and every event is graded on five fields: signal type, support, localization, attribution and verdict. The rules were frozen as version 1.0 on 26 Oct 2026; version 1.1 (27 Oct) added EASA's weekly FIR tables as a source for "deliberate suspected" and changed nothing else. They are stated choices, not yet tested against held-out cases. The register of events stays private; only these monthly counts are published.
- Support. "Multiple sources" needs two independent sensor families agreeing in place and time (aircraft and CYGNSS). "Partially corroborated" is one sensor family plus an official notice consistent with it, such as an announced FAA GPS test or EASA's weekly listing of the FIR. Support says how well an event is evidenced, not what caused it.
- Verdict. Planned: every effect lies inside an announced FAA test window and its FL250 radius, the one the map draws. Benign natural: a 10 m-only effect during a geomagnetic storm (poleward of a latitude gate, or inside a storm-flagged SBAS area) or with ICAO-moderate ionospheric scintillation in ESA's nowcast; worse-than-93 m loss is never excused this way. Deliberate suspected: an official notice reporting GNSS jamming or spoofing covers the event, or EASA's weekly table lists spoofing in the event's FIR that week (FIRs are approximated by country outlines). Spoofing is deliberate by nature, so the listing, not our data, is the evidence of intent, at the scale of a FIR and a week; EASA's table may itself draw on ADS-B data. EASA tables exist for 13 weeks of this period (its current table, and earlier copies kept by the Internet Archive), so this count follows which weeks have a table, not how much spoofing there was: months without one show zero. Open: none of these.
- No grade names a perpetrator, and no event has a position fix yet.
Effect events, 1 Oct 2025 – 25 Oct 2026 (UTC). Besides these, 756 announced tests had no measured effect matched to them, and 74 geomagnetic storm periods are recorded as events of their own.
By support
| Month (UTC) | Days | Effect events | Single-source | Partially corroborated | Multiple sources |
|---|---|---|---|---|---|
| Oct 2025 | 31 | 11,861 | 11,126 | 223 | 512 |
| Nov 2025 | 30 | 12,925 | 11,695 | 634 | 596 |
| Dec 2025 | 30 | 12,599 | 11,706 | 238 | 655 |
| Jan 2026 | 31 | 13,009 | 12,244 | 128 | 637 |
| Feb 2026 | 28 | 11,276 | 10,177 | 600 | 499 |
| Mar 2026 | 31 | 13,919 | 12,050 | 1,025 | 844 |
| Apr 2026 | 30 | 12,313 | 11,170 | 551 | 592 |
| May 2026 | 30 | 12,411 | 11,399 | 301 | 711 |
| Jun 2026 | 30 | 12,112 | 11,230 | 226 | 656 |
| Jul 2026 | 31 | 12,610 | 11,653 | 94 | 863 |
| Aug 2026 | 31 | 12,365 | 11,448 | 134 | 783 |
| Sep 2026 | 30 | 12,503 | 11,597 | 130 | 776 |
| Oct 2026 | 25 | 9,414 | 8,474 | 432 | 508 |
| Total | 388 | 159,317 | 145,969 | 4,716 | 8,632 |
By verdict
| Month (UTC) | Open | Benign natural | Planned | Deliberate suspected |
|---|---|---|---|---|
| Oct 2025 | 11,729 | 57 | 39 | 36 |
| Nov 2025 | 12,590 | 102 | 44 | 189 |
| Dec 2025 | 12,427 | 62 | 7 | 103 |
| Jan 2026 | 12,799 | 185 | 25 | 0 |
| Feb 2026 | 10,942 | 52 | 54 | 228 |
| Mar 2026 | 13,364 | 136 | 80 | 339 |
| Apr 2026 | 11,983 | 85 | 102 | 143 |
| May 2026 | 12,237 | 41 | 53 | 80 |
| Jun 2026 | 11,972 | 19 | 53 | 68 |
| Jul 2026 | 12,519 | 44 | 47 | 0 |
| Aug 2026 | 12,249 | 45 | 71 | 0 |
| Sep 2026 | 12,410 | 19 | 74 | 0 |
| Oct 2026 | 9,178 | 58 | 42 | 136 |
| Total | 156,399 | 905 | 691 | 1,322 |
Most effect events are 10 m-only departures from a region's own recent baseline, a weak signal; the method report splits them by measure. Data: events-monthly.json, derived from the adsb.lol archive and published under ODbL 1.0 with attribution; also NASA CYGNSS, FAA flight advisories, the GFZ Potsdam Kp index (CC BY 4.0), ESA's scintillation nowcast: data provided by the European Space Agency (ESA) as part of its Space Safety Programme (S2P), © ESA 2026; and EASA's GNSS outages and alterations FIR tables (source: EASA), placed with Natural Earth country outlines. Method and full results
Validation still to come
- A localizer that resolves sites, not just areas. It failed its first scored test against the 2025 TDOA fixes, and its positions stay off the map until it passes one: two-emitter fits and fix-window timing were tried and dropped, and a heading-balanced third version improved on the second but didn't beat a constant guess. Next: days with a single known active site, scored at about 10 km.
- More announced test events as known positives. The FAA advisories are already a layer on the map. Jammertest 2026 showed no aircraft effect.
- More false look-alikes, beyond the Kiruna radar fault, as cases the methods must not flag.
- A surveyed emitter or announced test to measure NISAR position accuracy.
Each case is checked against its source before it's scored.