What changed. We identified the aircraft type on 98 type codes that our records had been carrying as unclassified, and applied that identification to every operation back to 2020. The effect on the site is that operations move between the categories we display. No daily total changes, on any day, anywhere in the record. This is a reclassification of flights we had already counted, not the addition or removal of any flight.

What you will see. Across the whole record, 4,872 operations move out of “Other” — 4,849 of them into Jets, and 23 into Helicopters. Seaplane counts are unchanged. On the dashboard’s Last Month view for July 2026, that is 19 operations moving from Other to Jets, with the July total holding at 4,641. On the 2025 high-season jets page, the figure moves from 3,568 to 3,597.

Why “Other” and not something more specific. The site groups fixed-wing aircraft and unclassified aircraft together under “Other”. Underneath, 8,788 individual flight records were reclassified — but nearly half of those moved from “unclassified” to “fixed-wing”, which are both displayed as Other. The number you can check against the site is 4,872; the larger figure describes our internal records and will not reconcile against what is on screen.

Loud-jet counts. 533 operations are newly counted as loud jets, being aircraft now identified as jet types that exceed our published loudness threshold. The largest single group in this change — 5,761 records of one light-jet type — is not among them; it falls below that threshold.

Where the Noise Impact Score moves. The score is built from the helicopter and loud-jet share of operations, so days gaining loud jets will show a higher score. Days whose reclassified aircraft went to fixed-wing see no change from this.

Comparing across the change. Any comparison spanning this change is comparing two different classification bases. Figures from before it will show more operations under “Other” and fewer under “Jets” than the same days show now. This is the third such boundary in our record, alongside the 2024–2025 local-circuit change and the April 2025 classification boundary.

One further step remains. Our noise map is computed from aircraft type, so this reclassification also changes how the map is calculated. Those figures are being measured now and will be published when the next version of the map goes live; this entry is not complete until they are. We are publishing now rather than waiting because the operation counts described above are already live on the site and should not remain undescribed.

Earlier on August 19, 2026 we published two entries below — one on duplicate flight records, one on a test record. Both remain accurate about the individual days they list. Both were written while the work was still going on, and the totals in them were overtaken within hours. This entry supersedes their summary figures. We are leaving those entries in place unchanged, as this log always does, and correcting them here.

What the complete account is. Nine records, not seven, were found to be duplicates — the same physical flight recorded twice by our data provider — and removed from the counts. They fall on eight days. Two further records were removed for different reasons that are not duplication: a test record on February 25 (its own entry, below) and a departure on June 21 that our provider and its own alert both place at a different airport, not East Hampton.

The days that changed. February 25, 2026 from 1 operation to 0; June 21 from 209 to 207; July 9 from 140 to 139; July 25 from 190 to 189; July 30 from 143 to 142; August 1 from 147 to 146; August 4 from 144 to 143; August 16 from 207 to 206; and August 18 from 133 to 131. Eleven operations in total. Every one of those figures is live on the site as of this entry.

Two of the nine could not be settled the way the others were. We normally confirm a duplicate by comparing the two records’ flight paths: if they trace the same course at the same moment at the same speed and altitude, they are one aircraft. Two of these records have no stored flight path at all, so that test was unavailable. We settled them instead by asking where the aircraft actually was. In one case our provider’s own history shows the named aircraft airborne on a different route, hundreds of miles away, throughout the window it was supposed to be landing here. In the other, both our provider’s history and the original alert record the flight as departing a different airport entirely.

Why our daily check did not catch any of this. That check compares our records against our provider’s own accounting of the same day. When the provider supplies both copies of a duplicated flight — which is what happened every time here — the two sides agree and the day looks clean. August 4 reconciled as a perfect match while carrying a duplicate we removed later that evening. The check’s silence is not evidence that a day is free of duplicates, and we should not have treated it as such.

A downward step on these nine days reflects the removal of records that should never have counted, not a decline in real air traffic. We reviewed all of 2026 for this pattern by hand and corrected everything we could confirm. We have not reviewed earlier years, because the test that settles these cases needs stored flight-path data for both records and we mostly do not hold it before 2026. There is no automatic detector for this; the review was manual.

Why this entry exists at all. The earlier entries were published while corrections were still being applied. That is the mistake, and it is a straightforward one: a record of a change should be written after the change is finished and its figures re-read, not partway through. The individual day figures we published were right; the summary totals around them were not.

This site keeps a permanent, never-edited log of every change to how it counts, and our methodology commits to making the entry at the time of the change. Between June and August 2026 we made three such changes and did not publish the revisions here. We found this during an internal audit of our disclosure process, and we are publishing the entries now.

The corrected numbers have been live on the site throughout. What failed was the step that records them here — a step that until now depended on someone remembering to take it.

June 27, 2026 — we unified the definition of “an operation” across the site. Different pages had been counting slightly different things. We changed eighteen display surfaces to apply a single definition. Every one of them had been counting high: one page showed 71 operations where the site’s headline figure was 66, and on days with diverted flights a summary tile read 135 where the published total was 134. No published daily or monthly total changed — these were secondary displays that had drifted above the headline, and they were brought down to match it.

August 6, 2026 — we corrected the flight counts shown for individual addresses. The My Home page reports activity over a resident’s own address. Its underlying queries did not exclude duplicate records or internally generated test records, which every other count on the site does. We corrected the queries and added a separate operations count. No East Hampton address’s flight count changed. The page matches flights by their recorded flight path, and the internally generated test records have no flight path at all, so they could never have been counted over anyone’s home. Of the duplicate records in our data at the time, exactly one had a flight path, and near East Hampton it stayed further from the airport than the page’s widest search radius — its closest approach was 1.7 nautical miles, and at the three addresses we measured every total was identical before and after. That one flight also passed over the New York City area on its way here, where an address directly beneath it would show one fewer flight for June 10, 2026. What the change added is a second, smaller figure beside the total — the subset that meets our published definition of an operation — and a label on records we cannot confirm.

August 12–13, 2026 — we corrected how local round trips are counted, and restated published totals. When an aircraft departs from East Hampton and returns without landing elsewhere, our flight-data provider sends it as a single record. We split it into a departure and an arrival and count both, which is standard practice. We found cases where that split was being applied to flights whose own tracking data shows they did not begin at East Hampton. We added a check against the tracking data. The same review found the opposite error as well: three real East Hampton departures had been recorded as arrivals diverted in from another airport. Correcting those raised three days.

That change restated totals we had already published, and it moved them in both directions. Six days changed: July 27, 2025 went from 141 to 140; July 27, 2026 from 177 to 176; August 12, 2026 from 151 to 150; June 18, 2026 from 100 to 101; July 29, 2026 from 85 to 86; and August 2, 2026 from 201 to 202. The July 2025 monthly total went from 4,419 to 4,418.

One further day, August 5, 2026, carries a record of the same kind, and its published figure differs by one from what our morning reconciliation recorded at the time. We could not establish from our records whether that difference was caused by this correction or by flight data arriving late, which is common and ordinary. Rather than assert a cause we cannot support, we are noting it and leaving it out of the list above.

In April 2026 we analysed a problem in how we count local round trips, identified 99 specific records that should not have been counted as operations, wrote the correction up in full, and recorded the exact list. We then did not apply it. The records stayed in our published counts for a further 115 days, until August 19, 2026.

Our internal notes recorded the opposite — that the correction had been made and only the public wording was pending. That was wrong, and nobody checked the database against the note. We found it while auditing our disclosure process for the entries above.

What the records were. When an aircraft departs East Hampton and returns without landing elsewhere, we count two operations. To do that we create a second, synthetic record for the departure half. The April analysis found two groups where that was being done wrongly: flights that were in fact ordinary round trips to somewhere else and back, and flights whose closing arrival message never reached us, which made an outbound flight look like a return to the field.

What changed on August 19. 99 records across 43 days between January 3 and April 26, 2026 no longer count. January 2026 falls by 40 operations across 17 days, February by 33 across 12, March by 2 across 2, and April by 24 across 12. The monthly totals now read 311, 321, 445 and 738. The 2026 year-to-date figure through August 18 falls to 14,107. Applying it now, rather than reconstructing the April reasoning from scratch, means we removed exactly the records April identified and no others.

The underlying condition has not been fixed. The rule that produced these records is still in place for new flights arriving today. Correcting it properly requires recovering the precise test the April analysis used, and we have not been able to reconstruct that from what was written down. Until we do, this class of record can still be created. We would rather say so than describe a correction and let a reader assume the cause was dealt with.

We have kept every corrected record rather than deleting it, marked so it no longer counts, so the correction stays auditable.

A single test record, created during software development in February 2026, was counted as a real arrival at East Hampton. It was the only record we held for February 25, 2026, so that day published one operation that did not occur. The day now correctly shows zero. February 2026 falls from 355 operations to 354, and the 2026 year-to-date figure through August 18 from 14,210 to 14,209.

A zero-operation day is unremarkable at that time of year: February 23 and 24 already stood at zero, and single-figure days were normal that fortnight. The correction removes a flight that never happened; it does not indicate a gap in our data for that date.

This extends our earlier entry on the 413 sample records removed from January and February 2026. That correction dealt with one batch of development data imported in a single operation, and this record was not part of it — it was created separately, two weeks later, and so was not caught then. We have now searched our full history for records bearing test-like identifiers and found no others.

One clarification is owed on that earlier entry. We wrote that a check now runs every night to confirm no generated record has entered a published total. That check verifies that records marked as generated stay out of published figures, and it does that reliably. It cannot identify generated data that was never marked in the first place — which is what this record was, and why it survived. It was found instead during a manual review of records that carry no aircraft registration. As with the earlier correction, the record has been marked and excluded rather than deleted, so the correction remains auditable.

FlightAware, the data source behind every figure on this site, sometimes issues two separate records for one physical movement — the same aircraft, the same takeoff or landing, recorded twice a few seconds or minutes apart. We found seven such pairs in 2026 and had been counting both halves of each. Each one inflated a daily total by one.

Six days are affected. July 9 falls from 140 operations to 139; July 25 from 190 to 189; July 30 from 143 to 142; August 1 from 147 to 146; August 16 from 207 to 206; and August 18 from 133 to 131, the only day with two. The year-to-date total through August 18 falls from 14,217 to 14,210. The largest single-day error was 1.5%. Corrected figures are live as of this entry.

These were hard to see. Our automatic daily check compares our records against FlightAware’s own accounting of the same day. When FlightAware supplies both copies of a duplicated flight — which is what happened in every one of these seven cases — the two sides agree and the day reconciles cleanly. The duplicates were found instead by comparing the recorded flight paths: two records describing one aircraft trace the same course at the same moment, at the same speed and altitude, which two genuinely different aircraft do not do. In these seven cases the paired records stayed within about half a nautical mile of each other for the length of the flight, while the closest pair of genuinely distinct aircraft we examined in 2026 was more than nine nautical miles apart.

We reviewed all of 2026 for this pattern and corrected everything we could confirm. We have not extended the review to earlier years: the test that settles these cases needs stored position data for both records, and for most flights before 2026 we do not hold it. Earlier totals may therefore contain duplicates of this kind that we cannot presently identify. We also do not yet have an automatic detector for it — the daily check remains unable to see a duplicate that FlightAware itself supplies twice, and this review was done by hand.

The Operations card at the top of the dashboard can be filtered by fleet: All, Helicopters, Jets, Seaplanes, and Other. The operation counts on every tab have always been correct. But the “vs. prior year” percentage and the small trend line beside it were only ever computed for All, Helicopters, and Jets. On the Seaplanes and Other tabs, both showed the all-aircraft figure instead of the figure for that fleet.

The two tabs were added on March 24, 2026; the year-over-year comparison predated them and was never extended. The error was therefore live for roughly five months.

The clearest case is the default view, which compares the last complete month to the same month a year earlier. For July 2026 against July 2025, the Seaplanes tab showed an increase of 5.1% — the all-aircraft figure. Seaplane operations had in fact fallen 24.9%, from 474 to 356. The published figure had the wrong sign: it reported a rise in seaplane traffic during a month when seaplane traffic dropped by about a quarter. On the same view, the Other tab showed 5.1% against a correct 8.5%. Over a trailing 90-day window, Seaplanes showed +12.5% against a correct −6.5%, and Other showed +12.5% against a correct +13.2%.

This went unnoticed because, in most windows, the Other category is large enough that its year-over-year change sits within about a percentage point of the all-aircraft change — the wrong number and the right number looked alike. Seaplanes are what exposed it, and only because they moved in the opposite direction to the rest of the fleet.

Each tab now computes its own comparison and its own trend line. Where the prior-year seaplane figure is unavailable, or is genuinely zero — seaplane service is seasonal and several winter months have no operations at all — the card now shows no percentage rather than a misleading one. A test now fails if any fleet tab reproduces another tab’s figure. In the same change we corrected the detail panel, which titled a list of seaplane flights “Other Operations.”

The Today page includes a list of operators with three or more operations in the voluntary curfew window (10 PM–7 AM) over the trailing 30 days. That list was built with a simpler filter than the site’s published counting definition. In particular, it did not exclude flights flagged for review — including flights whose ADS-B track shows they never came within 4 nautical miles of the airport, which are excluded from every other number this site publishes.

One operator was affected. From June 21 through July 20, 2026 — 24 days, in two stretches — the list showed the operator “CNS” with three curfew-period operations. Under the published definition the correct number was two: the third record, a late-night June 20 departure, is a flight whose track never came within 4 nautical miles of East Hampton Airport. Because the list requires three operations to appear at all, the operator should not have been on it those days. The listing rotated off screen on July 20 in the ordinary movement of the 30-day window. We have verified that every operator on the current list meets the threshold under the full definition.

The same audit found a milder form of the same gap on the Jets, Helicopters, and Seaplanes page totals and on parts of the Today page: those displays counted some operations the published definition excludes. Measured over a recent 30-day window, the difference across all categories was 56 operations out of 4,842 — about 1.2%, always in the over-counting direction. The headline counts — the homepage, the daily summary, the historical charts, and My Home — were never affected; they already applied the full definition, which is how the discrepancy was found.

As of this entry, every public count surface applies the same five-part definition, and an automated test now fails whenever a counting surface stops doing so.

We reviewed Rev. 4 (March 23, 2026) of the Town’s “Helicopter Arrival and Departure Routes” and updated our citations, which previously referenced Rev. 3 (March 4, 2025). Rev. 4’s one substantive change revises the Sierra departure procedure: the climbing turn is specified north of Runway 10-28, and departures are to cross Georgica Pond at 2,500 feet or above, replacing the prior best-rate-of-climb profile. Waypoint coordinates are unchanged, so route classification is unaffected.

In the same update we corrected two route descriptions to match the published document: we removed a statement that the November Route is intended to relieve congestion on Echo, and a statement that Sierra availability depends on the absence of fixed-wing traffic south of the airport — neither appears in the Town’s document.

FlightAware occasionally records one physical flight as two flight records (see the June 2026 “Duplicate-flight-record correction” entry below). Operation counts already exclude the duplicate record in each pair — but the Community Noise Map and My Home noise estimates had no such rule: every stored flight track fed the noise computation, so a duplicated record’s track contributed its noise energy in addition to the real flight’s. In the current season’s data, one such record qualified: a duplicate of a June 10, 2026 departure, whose recorded track (55 GPS positions) added a second copy of that flight’s noise energy at 1,478 grid cells along its path.

As of this revision, the noise computation excludes duplicate-marked flight records by rule, the same way the operation counts do; the genuine record of each flight is unaffected. The effect on the published surface is small: in the recomputed surface, published with this week’s regular season-to-date refresh (data through August 3, 2026), no grid cell’s DNL estimate fell by more than 0.8 dB, with the decreases concentrated along that flight’s June 10 path. Because DNL is a seasonal energy average, one duplicated flight moves even the cells directly under its path by well under a decibel.

The change is versioned as noise methodology 2026.08-duplicate-track-exclusion, and every grid computation is stamped with the methodology version that produced it. Duplicate records remain in the database — marked and excluded, not deleted — so the correction is auditable.

Added the East Hampton Town Airport Operating Procedures memo (Rev. 19, 7/17/2026) as a documented source on the About page. The memo corroborates the Town’s published 10 PM – 7 AM curfew (stated in the memo since at least Rev. 11, 9/21/2022) and states helicopters “are expected to fly” the published Noise Abatement Routes — stronger than the “strongly encouraged” phrasing in revisions through Rev. 14 (5/30/2024). No methodology, classification, or scoring change.

413 flight records were erroneously included in JPXWatch’s published operation counts for January 13 – February 11, 2026. They were sample records generated for software development that were mistakenly included in the live database in February 2026, where they were counted alongside genuine FlightAware data for roughly six months before being identified and removed on August 1, 2026.

Across those 30 days, the published figure falls from 768 to 355 operations. January 2026 moves from 617 to 351 and February 2026 from 502 to 355. On five days — January 18, January 25, January 26, February 7 and February 8 — every record was a sample operation, so those days now correctly show zero operations where JPXWatch previously reported 19, 22, 13, 17 and 19. Anyone who viewed those dates saw activity that did not happen.

No other period is affected. The sample records existed only in that 30-day window. Annual totals for 2020 through 2025 are unchanged: 13,537 / 18,722 / 15,097 / 15,820 / 17,273 / 18,763. The 2026 year-to-date figure falls by 413 operations, and the all-time total from 2020 onward moves from 111,075 to 110,662 — a reduction of 0.37%. The error was small against the full record and large within the month it touched, which is why it went unnoticed: it is invisible in an annual figure and unmistakable in a daily one.

What went wrong. Sample data was created for a local development database, which is normal practice. Nothing on those records marked them as artificial, so when data was later copied into the live system, they were indistinguishable from real flights, and every check that counts operations counted them. The failure was not that the sample data existed — it was that nothing identified it as sample data.

What has changed. Every flight record now carries an explicit marker for whether it came from FlightAware or was generated; the marker is applied at the moment records are created, and records marked as generated are excluded from every published count by rule rather than by any individual check remembering to exclude them. A check now runs every night to confirm that no generated record has entered a published total. The 413 records were kept rather than deleted — marked and excluded — so the correction remains auditable.

A flight that takes off from East Hampton and later lands back at East Hampton performs two operations — one takeoff and one landing. FAA air traffic activity counting tallies each landing and each departure separately, and JPXWatch has applied that rule to newly recorded flights since April 2026. But it was applied only to short pattern work. Longer round trips — the seaplane runs to the East 34th Street Skyport and back, training flights that leave the area and return, helicopter out-and-backs — were still recorded as a single operation, because FlightAware delivers them as one record when the aircraft’s stop at the far end produces no airport arrival event.

This revision applies the same rule to those flights for 2025 and 2026. 972 departure records were added, each paired with the arrival record already published. The 2025 total moves from 17,890 to 18,763 operations (+873). The 2026 year-to-date figure rises by 99 operations (as of July 28, 2026; 2026 is an open year and its total continues to move with daily traffic). Curfew-period operations for 2025 move from 164 to 165 — a single added takeoff at 6:29 a.m.

Of the 873 operations added for 2025, 384 are individually confirmed by ADS-B track data. The remainder rest on the counting rule applied to the flight record, without a track to confirm the return.

2020 through 2024 are not restated. Published as before: 13,537 / 18,722 / 15,097 / 15,820 / 17,273. The reason is coverage — ADS-B track data exists for only 0.2% to 0.8% of affected flights in those years, compared with 48.5% for 2025 and 100% for 2026. Restating would mean applying an estimated rate rather than a verified count. The practical consequence is that 2025 and 2026 are not directly comparable to 2020–2024 on total operations.

The Community Noise Map and the My Home noise estimates now show the current 2026 season to date, updated weekly through Labor Day, replacing the completed Summer 2025 surface that had been the default. The modeling is unchanged from the June 2026 propagation-and-window regime — the same per-flight source levels, atmospheric absorption, and track-derived exposure — but the seasonal DNL is now energy-averaged over the calendar days elapsed since Memorial Day (52 of the season’s 106 days, through July 15, 2026) rather than over a completed season. Because DNL is an average, this figure converges toward a full-season value as the season fills in; the map label always states the through-date, and every 2026 figure is marked “Estimated.”

As of data through July 15, 2026, the surface models 168 grid cells at or above 65 dB (maximum 73.6 dB), with a higher median across the grid than the prior surface (31.0 vs 28.3 dB). These figures are not directly comparable to the completed Summer 2025 surface (75 cells at or above 65 dB): this covers only the first 52 days of the season — peak early summer — while 2025 spans the full Memorial Day–Labor Day window, and a season-to-date average over a shorter, busier window reads higher and will move toward the full-season figure by Labor Day. They should not be read as a finished year-over-year change. The 2025 season data is retained for reference.

These remain modeled estimates, not monitor measurements or formal Part 150 noise contours.

Until July 11, 2026, the per-route “Altitude range” figures in the Seaplanes page’s corridor overlap analysis displayed raw FlightAware track altitudes — reported in hundreds of feet — as if they were feet, understating altitudes by a factor of 100. The November route’s 2025 high-season range read “0–56 ft” where the corrected figure is 0–5,600 ft; Echo read “−2–41 ft” (corrected: −200–4,100 ft) and Sierra “1–21 ft” (corrected: 100–2,100 ft). This is the same defect class as the helicopter altitude-adherence correction below, in a separate code path. Corrected ranges display reported values as-is, including occasional below-zero readings near water touchdown; figures in the precomputed summer view are computed from simplified GPS tracks, so live-computed views over the same period can differ modestly.

The same mismatched values also fed a filter that excludes low-altitude positions near the airport from corridor counts — read in the wrong units, it excluded near-airport positions below 10,000 ft rather than below 100 ft. Corridor overlap counts were essentially unaffected: recomputation over the full 2025 high season reclassified exactly one position of roughly 1,150 (November: 937 to 938 inside corridor); flagged-flight counts and overlap percentages are unchanged, because the published corridors barely intersect the one-nautical-mile exclusion radius around the airport. Both affected paths — live-computed views and the precomputed Last Summer data — were corrected and the precomputed data regenerated on July 11, 2026.

From the launch of the Helicopter Route Activity page on March 19, 2026 through July 11, 2026, the “Altitude Adherence at Checkpoints” statistics compared GPS track altitudes against the published route altitudes in mismatched units. FlightAware reports track altitude in hundreds of feet; the comparison read those values as feet, so every recorded altitude evaluated at roughly one percent of its true value — thousands of feet “below” the published 2,500 ft and 3,500 ft checkpoints. As a result, the page reported 0% of flights within the published altitude at every checkpoint, for the entire period.

With the units corrected (July 11, 2026), the same underlying data shows a substantial share of checkpoint passages within tolerance: for the 30 days ending July 9, 33% at E2, 49% at E3, 43% at E4, and 58% at N1 were within ±300 ft of the published (voluntary) altitude — that definition is unchanged. Two things did not change. Route usage figures (Echo / November / Sierra / Non-Standard) were never affected: route classification matches tracks laterally against waypoint positions and does not use the altitude comparison, which recomputation over the full season’s flights confirms. And the corrected data still shows more checkpoint passages below the published altitudes than above them — the defect overstated how far below, not whether. Adherence statistics are computed live from stored GPS tracks each time the page is viewed; no stored data was involved, so all views, including of past periods, now show corrected figures.

Through this revision, operator cards showed a letter grade (A–F) computed over whatever time period a visitor had selected (default: the trailing 30 days), the fleet filter could further narrow the flights being graded, and any operator or registered owner above a small activity floor received a grade.

As of this revision, three things changed together. The window is fixed: every card is computed over the trailing 12 months ending on the most recent complete day (Eastern Time), shown on the card; visitor selections no longer affect any score. The letter grade is retired: cards now show a 0–100 Observed-Operations Score — higher means fewer deductions — computed from two inputs: curfew-period operations (a first operation costs the same for every operator; repetition escalates with the share of flights during curfew hours; up to 60 points) and fleet noise (the share of flights at or above the JPXWatch-defined 85 dB modeled threshold; up to 25 points). Shoulder-hour operations and average event-level noise no longer deduct points; both remain displayed. The previous per-event formula does not scale to a 12-month window — with fixed per-event penalties and caps, high-volume operators accumulate enough events over a year to hit the caps regardless of how rare those events are as a share of their flying, so the score would have measured volume, not behavior. Scores now attach only to demonstrated agency: operators with commercial flight attribution and at least 12 operations in the window. FAA-registered owners, unconfirmed identities, and privacy-masked attributions receive full statistics but no score — registry ownership alone cannot distinguish an owner-operator from a passive lessor.

Because the window moves daily, scores change daily — as operations enter and age out, and when historical records are corrected. The full formula, including every tier boundary, is published on the About page. Nine operator cards were also consolidated on documented registry evidence (same registrant, address, or aircraft appearing under trivially different spellings of one name); the audit record is committed to the project repository. No historical grades or scores were retained or republished under either formula; both have always been live computations.

On about 18% of all operations since May 2025 — roughly a quarter of arrivals — FlightAware’s feed recorded the landing time as identical to the origin takeoff time, stamping the arrival when the aircraft left its origin rather than when it touched down at East Hampton. The pattern followed air-traffic-control coverage, not aircraft type: VFR general-aviation traffic — fixed-wing, helicopters, and seaplanes — was affected, while IFR jets, tracked end to end, were nearly immune. The defect is upstream (FlightAware never marked these flights “arrived”) and eased after roughly May 2026. JPXWatch has repaired 3,278 of these arrivals from FlightAware’s own landing estimate or last observed track position — 3 with no recoverable time were flagged and excluded from time-sensitive statistics; repaired times are modeled estimates, and operation totals are unchanged.

The fix lowered reported curfew-period (10 PM–7 AM) arrival counts, because early-morning arrivals whose recorded landing was actually their origin’s pre-7 AM takeoff had been mis-counted inside the curfew. Curfew-period arrivals were revised down — for 2025 (May–December), 124 to 93; for 2026 (January 1–July 7), 75 to 69. The two periods are not directly comparable (2025 is a partial year, 2026 year-to-date); for the same May 1–July 7 window, corrected curfew arrivals run 30 in 2025 and 44 in 2026. Daily and monthly summaries were recomputed from the repaired data.

The Seaplanes page summarized the high season from a pre-computed snapshot built on a fixed May 1 – September 30 window, while the page itself was labeled and dated to the canonical Memorial Day–Labor Day window. The snapshot also predated several months of data backfills and duplicate-record corrections. It has been regenerated on the Memorial Day–Labor Day window (2025-05-26 – 2025-09-01) from current data.

The high-season seaplane operation count now reads 934 (previously 732); the monthly distribution, operator breakdown, curfew-period count, and corridor-overlap analysis were recomputed over the same window. The increase reflects both the corrected window and the more complete underlying data — the earlier snapshot under-stated the true Memorial Day–Labor Day figure.

The seasonal DNL grid has been regenerated on the Memorial Day–Labor Day window. The May 2026 “Propagation model update” entry (below) projected grid cells at or above 65 dB rising “from 10 to 83,” with seasonal DNL up roughly +1.8/+2.15 dB (median/mean). Those were pre-release projections; the grid was not regenerated then, so it kept showing the prior model until now. Measured: 75 cells at or above 65 dB (previously 4), maximum 73.0 dB, and a +1.60/+1.72 dB (median/mean) per-cell shift. This entry qualifies, but does not alter, the May entry.

The 65 dB contour now also encloses a second zone near Francis S. Gabreski Airport (KFOK), Westhampton — the grid models all KJPX operations wherever they fly, including the KJPX–Gabreski shuttle. The map and About page were updated to match.

These remain modeled estimates, not monitor measurements or formal Part 150 contours.

FlightAware occasionally splits a single physical operation into two flight records. When both records were stored as completed, JPXWatch counted one departure twice. An audit identified and removed nine such duplicate departures, each on a separate day, lowering that day's total operations by one: 2020-07-15 (76→75), 2021-04-02 (31→30), 2021-06-21 (56→55), 2021-07-29 (121→120), 2023-07-04 (65→64), 2023-08-20 (170→169), 2026-06-03 (64→63), 2026-06-11 (113→112), and 2026-06-12 (116→115). In each pair, FlightAware's own record marks one leg cancelled; the real operation is retained and only the duplicate is removed.

This correction was applied as a one-time audit. A small residual remains on records where FlightAware removed the aircraft's tail number, which our matching cannot yet pair automatically; these are tracked and corrected as identified. Because a duplicate adds a fixed one operation to a day's total, the effect is proportionally largest on the lightest-traffic days and smallest on the busiest days.

The My Home page computed its summary statistics from a flight list that truncated to 1,000 rows whenever an address, radius, and date-range combination matched more than 1,000 flights — a threshold some views began crossing as summer traffic built in early June. For those views, the headline count was understated and frozen at 1,000, and other measurements on the page were computed from the truncated list.

The count, fleet and curfew breakdowns, busiest day, and Closest Approach are now computed over the complete matched set; the noise measurements (loudest, high-noise, average) are still drawn from the most recent flights and now state when they reflect a subset, with the flight list doing the same. Views matching fewer than 1,000 flights — including the public demo — were unaffected. The issue was found by an internal audit of address-page numbers.

A terminology pass across address-level and summary displays, continuing the April 2026 unit-convention and curfew-terminology work. No metric, calculation, or threshold changed.

• The My Home high-noise summary now reads “flights with modeled event levels of 65 dB or higher at or near this address” (previously “flights above 65 dB at your address”), and its tooltip now states explicitly that this 65 dB is a JPXWatch-defined single-event threshold, unrelated to the FAA's 65 DNL cumulative threshold shown in the Estimated DNL card. The two figures share a number but measure different acoustic quantities.
• The total-flights denominator was removed from that card pending an internal review of how the underlying counts are scoped; the count of qualifying flights is unchanged.
• Remaining uses of “compliant / compliance” describing voluntary procedures (altitude, corridor, and curfew contexts) were replaced with adherence- and operations-based phrasing, consistent with the April 2026 terminology entries. The Today page's repeat flag now uses neutral frequency language.
• The “Est.” prefix and explicit dB units were added to remaining bare modeled values (e.g., Loudest Flight, Avg Event Level).
• The DNL seasonality link text was rewritten descriptively (“why seasonal concentration isn't reflected in annual DNL”).
• The comparison-population label for percentile rankings was unified to a single formulation across NoiseRank and DNL displays.
• The navigation label “Accountability” was renamed “Scorecards,” completing the April 2026 accountability-terminology cleanup, which had not reached the navigation label.

JPXWatch's aircraft noise source data is now maintained in a public, auditable registry rather than solely in application code. For each of the 57 mapped aircraft types, the registry records the served noise values, the EPNdB certification triple where published, certification basis (Part 36 appendix or EASA TCDSN), native unit, data source, confidence level, and a citable source document. An automated check verifies on every change that the registry and the production code remain identical; the noise model's constants and release notes are now versioned, and noise-grid computations are stamped with the methodology version that produced them. The basis-of-estimate indicator introduced in April 2026 is now driven directly from this registry.

No displayed value changed. This is the system-of-record infrastructure for the source-data provenance commitments described in the methodology section.

In populating the source registry, every cited source document was checked against public archives. The four FAA report designators long carried in the codebase for nine helicopter types' field-measured values — DOT/FAA/CT-84-2, FAA-AEE-01-04, FAA-AEE-09-01, FAA-AEE-15-01, as referenced in the April 2026 “Unified aircraft noise source-level data” entry — could not be matched to any public FAA, ROSAP, or DTIC record as of June 9, 2026.

Impact on published values: none. The helicopter noise values themselves are unchanged and remain in use. What changed is how their sourcing is recorded: the registry rows for the affected types now describe the values as recorded-as FAA LAmax measurements with provenance unconfirmed, carry medium rather than high confidence, and include EASA EPNdB certification data alongside for comparison. The values most plausibly trace to the FAA's INM/AEDT helicopter noise database, which descends from FAA measurement programs; source recovery is tracked, and citations will be upgraded if the values are matched to a resolvable document. This entry qualifies, but does not alter, the April 2026 entry — entries in this log are never retroactively edited.

Seaplane operations are now reported as a distinct category in the fleet-mix breakdown. Previously they were counted within fixed-wing.

What changed. We refined how an aircraft is identified as a seaplane, moving the determination to the airframe (tail) level: an aircraft is recognized as a seaplane from a confirmed operation at a water-only seaplane base, or from a known seaplane operator, and all of that aircraft's operations in eligible types are then counted consistently as seaplane. An amphibious aircraft does not change configuration between trips. Aircraft that cannot be confidently identified as seaplanes remain counted as fixed-wing.

Magnitude. Approximately 6,200 historical operations (2020–2026), previously included in the fixed-wing category, are now shown as seaplane. Fixed-wing category counts decrease by the same amount. Total operation counts are unchanged, as are the helicopter, jet, and curfew-period counts — only the fixed-wing / seaplane split is affected.

Composition correction. The refined method also surfaces seaplane activity by a second seaplane operation that had previously been folded into fixed-wing — much of its activity carries no operator label in the upstream data, so the earlier operator-based method did not recognize it. The seaplane breakdown therefore reflects more than a single operator.

Auditability. Every reclassified flight record is individually logged with its prior category and is reversible.

The “Today at JPX” / “Today so far” panels and the How Wind Direction Affects Runway Usage explainer describe which runway fixed-wing traffic is likely using and which way departures and arrivals fly. That description has been corrected and rebuilt on a single canonical definition of the airport's runway geometry, so the panels and the About page can no longer drift apart.

Correction. An earlier version of the live copy stated reversed departure/arrival directions for one runway case and asserted a runway as definitively “active.” Both were fixed: every runway-direction fact now derives from one source, and the directional facts were re-verified against the published FAA airport diagram.

Inference, not observation. JPXWatch does not receive a live feed of which runway is in use; it infers the likely runway from the current wind and the airport's published fixed-wing preference for Runway 28. The wording now reflects that — it hedges the runway (“likely using Runway 28,” “traffic may shift to Runway 16”) while stating the resulting overflight direction plainly. JPXWatch infers conditions; it does not direct traffic.

Crosswind runway now represented. The logic and the explainer now express the 16/34 crosswind runway, used in strong north–south winds, which the prior wind→runway description omitted entirely.

Favored vs. tolerated. The copy now distinguishes winds that genuinely favor a runway (aligned, a headwind) from winds a runway merely tolerates as a crosswind. In particular, the prevailing summer southwest wind sits at a wide angle to Runway 28 — more crosswind than headwind — so Runway 28 carries that traffic as the standard fixed-wing runway by preference, not because the wind favors it. The previous text implied the latter.

Stale-observation guard. When the most recent wind observation is more than about 90 minutes old, the panel now shows the observation time (“winds as of H:MM ET”) rather than presenting a confident runway claim on stale data.

Scope: fixed-wing aircraft only. Helicopters follow their own published routes (Echo, November, Sierra) and are not inferred from runway logic. No operation counts, aircraft classifications, or noise estimates are affected — the change is to the live conditions description and the methodology explainer.

The main dashboard's operations and curfew counts now apply the same definition of a counted operation that the public homepage and Today views already used, and that this page has described since April 2026: operations ADS-B shows never came within 4 nautical miles of KJPX (phantoms), and flights filed to KJPX that diverted and landed elsewhere, are excluded from the totals and the Helicopters / Jets / Seaplanes / Other breakdowns. Previously the dashboard's headline “Operations” card and its category and curfew breakdowns still counted those operations, so the dashboard read slightly higher than the homepage for the same period. The two now agree.

Underlying flight records are unchanged — diverted and flagged flights remain in the database and in flight lists; only the aggregate counts exclude them. As part of the same update, historical daily summaries predating the April 2026 ADS-B phantom audit were rebuilt so cached totals match the live definition, and a nightly recompute keeps them in sync going forward.

Following the Phase 2 wire-up of real METAR data, two user-visible items on the /weather page were inaccurate and have been corrected. First, the “Operations by Weather Condition” chart was binding to total flights per condition rather than the per-day rate; readers were comparing buckets of unequal denominators. Bars now show average operations per day of each condition with the total surfaced in the tooltip, matching the per-day framing already used in the Key Insights below the charts.

Second, a paragraph at the bottom of the /weather page still described historical weather correlation as “estimated weather pending verified historical-NOAA ingestion” — stale text from before the Phase 2 wire-up. The paragraph now describes the real KJPX-primary / KFOK-fallback sourcing consistent with the inline disclosure card above it.

No underlying data or methodology changes; the corrections are to chart bindings and attribution text. Companion clarity improvements landed in the same change — the Wind Rose title was extended to acknowledge that lobe length encodes operations frequency while color encodes average noise (both true), the Daily Weather & Operations Timeline gained a fixed “Ops” y-axis reference and bottom-anchored bars to remove visual ambiguity, and the per-chart KJPX/KFOK source-attribution tooltip now explains that 100% / 0% reflects observed coverage during the visible date range rather than implying KFOK is non-functional.

The Weather Correlation page at /weather previously rendered charts using deterministic estimated weather, with an inline disclosure acknowledging the limitation (see the prior “Weather correlation provenance” entry below). The page now queries real historical METAR observations from KJPX (East Hampton Airport, the primary source) with KFOK (Francis S. Gabreski Airport, Westhampton) as fallback for gaps in KJPX's reporting. Each chart shows the source distribution as “Data: KJPX X% / KFOK fallback Y%”.

Wind rose join. Operations are now bucketed into the wind rose by the wind direction at the actual hour each operation occurred, not by the day's aggregate dominant direction. The hourly join is more accurate when intraday wind shifts are large.

Source. Observations are sourced from the Iowa Environmental Mesonet, an academic archive maintained by Iowa State University that ingests from NOAA's authoritative feeds (Unidata IDD, NCEI ISD, MADIS One Minute ASOS). The backfill spans 2020-01-01 onward; KJPX covers approximately 98% of days and KFOK fills the remaining ~2%.

Coverage gaps. The Weather Correlation page's summary stats strip dropped its “Avg Humidity” cell because the daily summary view does not aggregate humidity. The other four cells (Avg Temp, Avg Wind, Avg Visibility, Period) are sourced from the real daily summary.

Live chip unaffected. The “Current” chip in the page header continues to be powered by NOAA Aviation Weather's live METAR feed (a separate path from the historical observations) and is unchanged. The Raw METAR tooltip is also unchanged.

The dashboard's Weather & Operations summary card was also updated to use the real hourly wind direction at each operation's hour for the “X winds dominant” subtitle, replacing the deterministic estimated wind direction the card previously used.

The “Show Noise Footprint” feature on individual flight details applied no unit conversion to FlightAware altitude data, producing dB values 11 to 16 dB higher than reality at typical cruise altitudes. A named helper at the boundary plus a cross-check test prevent recurrence.

The Weather Correlation Analysis card on the /weather page (Wind Speed vs Noise Level, Temperature vs Operations, Operations by Weather Condition, Wind Rose, and Key Insights) uses deterministic estimated weather for the selected period rather than observed historical NOAA data. The page header subtitle, the page-bottom source line, and the About page Data Sources entry previously implied the analysis used NOAA observations. Those claims have been replaced with honest disclosure: an inline notice on the correlation card identifies the data as estimated, the page header subtitle and source line describe each data path accurately, and the About page methodology subsections each note the estimated-weather provenance.

The live current-conditions chip in the WeatherCorrelation card header continues to use real NOAA METAR data and is unaffected. The Raw METAR tooltip continues to use real NOAA METAR data and is unaffected. The Estimated Carbon Emissions block uses a separate data path (FlightAware operations and ICAO emission factors) and is unaffected.

The Key Insights section now requires a minimum of 10 days per subset for any percentage-difference insight, and uses descriptive rather than causal language. Insights below the threshold are suppressed.

Verified historical-NOAA ingestion is planned. When implemented, the correlation analysis will use real observed weather and the disclosure language will be retired.

JPXWatch's various noise-modeling surfaces previously used inconsistent ground elevation defaults: the live noise trail visualization used 30 feet (a regional average for the South Fork coastal plain), while the footprint calculator, DNL grid pipeline, and altitude-compliance checks all used 55 feet (the airport field elevation per FAA Form 5010). The trail visualization has been updated to use the same 55-foot baseline as the rest of the pipeline, and to apply the same location-specific elevation overrides for characterized higher-elevation zones (currently the South Fork moraine ridge near Town Line Road in Wainscott, 132.5 feet MSL).

Impact on published values. Trail-visualized noise estimates increase by approximately 0.5 dB over baseline-terrain locations and by approximately 1.5 dB over moraine-override zones, reflecting more accurate AGL calculations. The DNL grid pipeline, footprint calculator, and other modeling surfaces are unaffected — they already used the 55-foot baseline and (for the DNL grid) location-specific overrides.

Planned improvements include extending location-specific elevation data to additional flight-path-sensitive areas (East Hampton proper, Sagaponack, North Fork analogs) and eventually integrating per-location terrain data from the USGS National Elevation Dataset.

JPXWatch's noise-estimation pipeline was updated to use track-derived event exposure duration and atmospheric absorption in the DNL grid pipeline.

Track-derived exposure. Sound Exposure Level (SEL) is now estimated using the time an aircraft spent within the listener's noise-impact radius, derived from FlightAware track data, rather than a fixed 15-second flyover default. For events with multiple track points within the radius, exposure is calculated from the observed track-point span plus a 30-second nominal interval; single-point events receive a 30-second nominal duration.

Atmospheric absorption. The model now applies a static atmospheric absorption adjustment as a simple linear function of total slant distance, approximately 0.5 dB per 1,000 feet. This adjustment was previously implemented but not applied to DNL grid calculations.

Lateral attenuation deferred. Lateral attenuation remains gated off in the live DNL grid pending a separate methodology review. The open questions are the provenance of the simplified attenuation table and the angle convention used by the lookup. This work is deferred pending external expert engagement.

Impact on published values. Estimated seasonal DNL values across the KJPX impact zone increased by approximately +1.8 dB at the median, with a mean increase of +2.15 dB across 335,016 grid cells. Grid cells modeled at or above 65 dB seasonal DNL-style exposure increased from 10 to 83. This increase is concentrated near the airport and appears to be driven primarily by the track-derived exposure-duration change at short slant distances, where atmospheric absorption is comparatively small. These values remain modeled estimates, not monitor measurements or formal Part 150 contours.

Documentation reconciliation. Public methodology documentation was updated to reflect the current live model. The About page now describes JPXWatch as a first-order source-propagation model informed by aviation-noise modeling principles, not as an AEDT implementation or regulatory contour model. The heatmap language was also revised to refer to the FAA Part 150 reference threshold, rather than implying that JPXWatch's seasonal-DNL estimates are equivalent to formal annual Part 150 contours.

This revision was implemented in three phases (A0, A1a, A1b) over April 2026, addressing inconsistencies in how the dashboard labeled aircraft noise values across surfaces.

Background. Aircraft noise certification under FAA/EASA Part 36 uses different units depending on aircraft category. Small fixed-wing aircraft (Appendix F, G) and small helicopters (Appendix J) are certified in A-weighted decibels (dB(A)). Jets (Appendix B) and heavy helicopters (Appendix H) are certified in Effective Perceived Noise Level (EPNdB), which accounts for the tonal character and duration of jet noise. JPXWatch's dashboard aggregates noise data across all aircraft categories. Prior to April 2026, the dashboard inconsistently labeled aggregated values as “dB(A)” — technically incorrect for the EPNdB-derived portions of the data.

A0 (April 2026 — PR 195). Per-flight value displays, tooltips, alert messages, CSV exports, and Mapbox text fields were updated to use unit-neutral labels (plain “dB” rather than “dB(A)”) in mixed-fleet aggregation contexts. The unit-neutral approach reflects that aggregated values combine measurements derived from different certification units and cannot be assigned to a single weighting convention.

A1a (April 2026 — PR 196). Five band legends across the dashboard (Aircraft Type Breakdown, Map Overlays Flight Density, Map Overlays Live noise trails, Today Hero Map compact, Weather Correlation wind rose) were unified to a four-band scheme: < 65 dB Quiet, 65–74 dB Moderate, 75–84 dB Loud, ≥ 85 dB Very Loud. A shared utility (lib/noise/bands.ts) was introduced to centralize the band definitions and color tokens. The DNL legend on the Noise Exposure Map was not changed — it uses FAA Part 150 land-use compatibility categories, which are appropriately distinct from the perceptual bands.

A1b (April 2026 — this revision). The About page methodology section was updated to describe the unit-convention approach explicitly. Several factual claims were corrected:

• AEDT NPD methodology is documented in Section 4.2 of the AEDT 4a Technical Manual, not Section 11.3.4 (which covers BADA 4 aircraft performance, not noise). The previous citation was incorrect.
• The seasonal-airport DNL passage was rewritten to argue from the underlying mathematics (averaging dilutes individual loud events at low traffic volumes) rather than from a specific JFK operations figure that was not directly verifiable.
• The High-Noise Events methodology in the My Home feature was repositioned as provisional pending methodology review. The previous framing conflated single-event modeled noise at a listener address with FAA Part 150's 65 dB DNL cumulative threshold; these share a number but measure different acoustic quantities. Refining the threshold value and its acoustic basis is a planned methodology task.

A1b also extended the unit-convention work to aircraft cards on the /seaplanes, /helicopters, and /jets pages. Each aircraft card's modeled-takeoff stat block now displays the unit that matches the data the dashboard carries: EPNdB for jets (computed from Part 36 Appendix B EPNdB cert readings); dB(A) for helicopters and small fixed-wing (data is LAmax at 1,000 ft, an A-weighted measurement, either field-measured via FAA ROSAP technical reports or converted from cert EPNdB).

AEDT subsetting detail. JPXWatch's current calculation — Part 36 reference noise plus inverse-square geometric spreading plus a linear approximation of atmospheric absorption — is a defensible subset of AEDT 4a methodology. Three specific AEDT adjustments are not yet implemented:

• Noise Fraction Adjustment (AEDT §4.3.3): accounts for the fraction of the flight path contributing to noise at a given listener location.
• Duration Adjustment (AEDT §4.3.4): accounts for variation in event duration relative to the reference duration used in NPD curves.
• Lateral Attenuation Adjustment (AEDT §4.3.5): accounts for ground-effect attenuation at lateral distances from the flight path.

Implementation of these adjustments is tracked as a planned improvement, awaiting expert review of how to subset AEDT methodology appropriately for KJPX's small-airport operating profile. Two AEDT-acknowledged limitations are inherited by JPXWatch's modeling and are documented on the About page (Section 11): under-prediction over acoustically hard surfaces such as water, and the inability to separately account for airframe noise as distinct from propulsion noise.

Renamed user-facing “Curfew Violations” to “Curfew Operations,” and “Curfew Violators” / “Repeat Offenders” to “Curfew Operators” / “Repeat Curfew Operators,” across the main dashboard, mobile app, analytics surfaces, narrative insights, alert display labels, and CSV exports. The rename aligns short-form UI labels with the descriptive framing already in use on this page and in the Grading Formula: the dashboard reports observed operations during the Town's published 10 PM – 7 AM curfew, not formal regulatory findings.

The word “violation” is retained where it refers to formal regulatory findings rather than dashboard metrics — specifically, in the operator scorecard disclaimer that distinguishes project-defined community transparency metrics from FAA regulatory determinations, and in the About-page parenthetical characterizing what the site's data is not (legal evidence of route violations; the published routes are voluntary procedures, not regulatory mandates). Those uses are legally precise and deliberate.

No metric, calculation, or filter behavior changed. String-literal alert-rule trigger types and internal identifiers / API field names are deferred to a subsequent migration so that saved alert rules continue to work.

The corrections log section of the About page was migrated to a standalone page at /about/revisions and renamed “Methodology & Data Revisions” to distinguish it from the existing /corrections submission portal. All prior entries are retained with identical substantive content; changes are limited to headline formatting, impact tagging, ordering, and the addition of the “About this log” preamble and “Summary of material changes” section. No entries were removed, softened, or materially revised.

Scorecard methodology and labels. Refined the scorecard methodology and supporting terminology. The “Event level” column replaces the previous “Est. dB” header; “high-noise” replaces “loud” throughout. The numerical compliance score is no longer displayed on scorecards; only the letter grade appears. The underlying scoring formula and grade bands are documented in the scorecard methodology section.

Noise estimation fallback logic. The site's fallback logic for aircraft types not in the type-specific certification mapping now routes through a category-average path (helicopter, jet, or fixed-wing) rather than a single generic default. This matches the fallback approach described in the methodology section and produces more accurate estimates for a wider set of aircraft. Existing records have been updated to reflect the new logic.

Basis-of-estimate indicator. Each operator scorecard now displays a small indicator next to the Event level showing whether the value is derived from type-specific certification data, a category average, or a generic fallback. The indicator and its explanatory legend make the provenance of each value visible at a glance.

Expanded type-specific mapping. Additional aircraft types observed at KJPX have been added to the type-specific noise profile mapping, with values sourced from FAA and EASA Part 36 certification data. This moves the noise estimates for these types from category-average derivation to values specific to their aircraft type.

Within-type variation disclosure. The About page's noise methodology section now includes a subsection on within-type variation, acknowledging that aircraft sharing an ICAO type code can produce different real-world noise depending on generation, configuration, modifications, and operating technique.

Flights filed to arrive at KJPX but diverted to another airport before landing are no longer counted as KJPX operations in public totals. This aligns with the physical-presence methodology already stated for outbound diversions: only aircraft that actually touched the KJPX runway are counted. Inbound-diverted flights are retained in the database for transparency (operator patterns, weather-related diversion clusters), and FlightAware applies the same convention on its airport pages.

Five genuinely-missing operations were recovered and added to the public record. Two transient webhook-ingestion bugs caused 39 flight events to fail during two narrow windows: 27 on April 8 (null-status constraint) and 12 between April 11 and April 13 (status-check constraint). Most of the affected flights were already captured by the next-morning batch pull from FlightAware's history API, so the operational record was never substantially incomplete. Replaying the dead-letter queue restored 5 genuinely-missing operations (N9934Q local-circuit training loops at KJPX on the afternoon of April 8).

As part of this work, a latent issue was also corrected: out-of-order webhook events could previously overwrite populated flight fields with null values. Flight records are now merged rather than replaced on re-ingest.

198 flights (1.13% of the 17,453 flights in the database with full ADS-B track data) were tagged as KJPX operations by FlightAware but never came within 4 nautical miles of the airport. These phantom operations are concentrated in two operators (Cutter Aviation and NetJets, approximately 65% combined) and are predominantly fixed-wing and jet rather than helicopter. The phantom rate has declined over time, from 5–14% in 2020 to under 1% in recent months, reflecting improvements in upstream data quality.

Going forward, every newly ingested operation is cross-checked against ADS-B position data. Operations whose closest approach to KJPX (40.9594°N, 72.2518°W) exceeds 4 nautical miles are flagged for review and excluded from public counts. The 4 nm threshold was set after sampling 25 borderline flights and finding a clean break: every flight whose closest approach was within 3.75 nm was a verified KJPX operation, and every flight at 4.33 nm or greater was a phantom.

Approximately 85% of historical operations come from earlier batch imports without ADS-B track data and cannot be retroactively verified by this method. Historical counts remain as originally recorded; year-over-year comparisons should account for the fact that data from this point forward is more rigorously verified.

JPXWatch now compares the filed destination on a flight's departure record against the actual landing destination on its arrival record. When the two differ, the flight is flagged as diverted and a “Diverted” badge appears in the operations table and flight detail view, matching the treatment in FlightAware's app. Diverted flights remain counted as KJPX departures (the aircraft did physically depart KJPX), but the badge gives a complete picture of the flight's outcome. A historical backfill identified one prior diversion in the database.

A trigger on the ADS-B position table had an incorrect search path and was silently rolling back point insertions for a subset of flights, leaving 337 flights without denormalized track data. The trigger has been corrected and the affected flights backfilled.

A methodology audit identified that the approach-phase noise offset for helicopter arrivals was being applied in a direction inconsistent with the underlying source data. The offset has been corrected. Approach-phase modeled estimates, which primarily affect helicopter arrival operations, now reflect the corrected calculation.

The underlying source measurement (lamax at 1,000 ft from FAA technical reports) is a flyover measurement; the approach value is derived by applying a −2 dB offset consistent with the EASA-derived convention used for all other aircraft types in the model. Details in the investigation report and methodology audit.

Helicopter noise estimates decreased by 1–5 dB depending on type; the C56X (Citation XLS+) decreased by 2 dB; 19 previously-unrecognized aircraft types now receive per-type estimates instead of a generic 80 dB fallback. JPXWatch previously maintained two independent noise lookup tables — a 28-type client-side table and a 47-type EASA-derived mapping — and different views (e.g., live noise trails vs. track-based analysis) could produce different dB estimates for the same aircraft type. This inconsistency has been eliminated.

All code paths now use a single lookup cascade: FAA field-measured data (where available for helicopters) → EASA certification data (47 types) → category-average fallback.

For 9 common helicopter types (R22, R44, R66, S76, EC35, A109, B06, B407, AS50), source noise levels now come from FAA technical reports containing actual ground-level measurements at 1,000 ft (DOT/FAA/CT-84-2, FAA-AEE-01-04, FAA-AEE-09-01, FAA-AEE-15-01). Measured values are lower than the prior certification-derived estimates.

For the C56X, takeoff decreased from 86 to 84 dB and approach from 82 to 80 dB, adopting the EASA certification value (84/80) over the prior un-sourced estimate.

Nineteen additional aircraft types (including S92, GLF6, CL30, CL35, CL60, C750, B350, and others) now receive per-type estimates; most are louder than the generic 80 dB default they previously received.

Two related changes to the Noise Impact Score:

Loud-jet correction. Historical Noise Impact Scores previously treated all jets as ≥85 dB in the Fleet Mix component because per-flight noise estimates were not preserved in monthly aggregates. JPXWatch now stores a per-day “loud jets” count in the daily summary table, populated from the same per-flight estimation pipeline used elsewhere on the site. Historical scores have been recomputed under the corrected methodology and may show modestly lower Fleet Mix subscores than previously published, particularly for periods with significant light-jet traffic.

SEVERE tier boundary adjusted. The SEVERE tier boundary moved from ≥80 to ≥85, narrowing the top tier to be more selective and better match the underlying score distribution. The HIGH tier now covers 65–84. This change applies prospectively and to historical period displays as scores are re-rendered.

The public corrections log is now live, providing a structured public channel for submitting and evaluating specific corrections to data points, classifications, methodology, and operator identity claims. All submissions and their resolutions are documented transparently. Accepted curfew-period exceptions (e.g., mechanical diversions, medical emergencies) are excluded from grading-formula penalties.

No changes to calculations, data sources, or displayed values; this update improves how the methodology is described.

Added explicit description of the regulatory pedigree linking Part 36 certification data to the FAA's AEDT, ICAO Doc 9911, and the Aircraft Noise and Performance database. Corrected the description of certification measurement geometry (three reference points per Part 36, not a single reference distance).

Added peer-reviewed validation citations: Rindfleisch et al. (2024, JASA) large-scale AEDT validation at SFO; Meister et al. (2021, Aerospace) three-model comparison at Zurich and Geneva; Simons et al. (2022) on certification-category correlation with measured noise trends at Schiphol.

Added comparative context noting other airport noise platforms (WebTrak, Casper, Noise-Map.com) that use the same aircraft-type-based methodology. Added “relative vs. absolute” framing clarifying that JPXWatch's primary analytical value is comparative ranking of aircraft types and operators, not precise decibel certification at specific locations.

Added prominent disclaimer that displayed decibel values are modeled estimates derived from type-certification data rather than on-site sound measurements.

Renamed glossary term “Repeat Offender” to “Repeat Curfew-Period Operator” to better match the project's framing of curfew operations as observed events against a published standard rather than legal violations.

Expanded the operator scorecard disclaimer to explicitly note that grades are not judgments about safety, legality, service quality, or overall company performance.

Softened narrative language about “gap between voluntary commitments and actual behavior” to “differences between published standards, voluntary commitments, and observed operations” to align tone with the project's measurement framing.

Renamed “Curfew violations” to “Curfew-period operations” within the Grading Formula to align terminology with the project's framing of curfew operations as observed events against the Town's published standard rather than legal violations.

Expanded the Compliance Grading Note to explicitly state that grades are not judgments about safety, legality, service quality, or overall company performance, bringing the About page Note into parity with the corresponding disclaimer on operator scorecards.

Replaced undefined uses of “accountability” with “scorecards” or “transparency metrics” in navigation, section headers, and narrative framing. This change does not affect any metric, calculation, or data display — it aligns framing language with the project's measurement-and-documentation stance rather than an adversarial one. Defined terms of art (“compliance grade,” “compliance rate”) are unchanged, as these are mechanically computed against the Town of East Hampton's published 10 PM – 7 AM curfew standard.