The Fidelity Ledger

Why this page exists

I check the game's rules against the orders and write down what I find, including the places it falls short. If you control for a living, this tells you where the model stops, so you can trust that nothing outside those lines is faked. If you are studying for the ATSA, it tells you which behaviors are safe to learn from and which are game-shaped shortcuts.

Rules first, then data. I would rather have a made-up airport flying correct procedures than a real one flying wrong ones. Every simplification here was a choice, and the test each one has to pass is whether it teaches you something false about how the system works. If it does, it is a bug and it belongs in the last column.

Primary sources: FAA Order JO 7110.65 (current edition, BB), 14 CFR 91.117, and the AIM. Paragraph numbers follow the current 7110.65 and are stable across recent editions unless noted. This page audits the code as it stands; it does not change it.

VectorHeavy is not affiliated with, sponsored by, or endorsed by the FAA. Procedure and airport data are derived from publicly available FAA products.

Verdict key:

• FAITHFUL: matches the real rule within the game's resolution.
• SIMPLIFIED-OK: a deliberate abstraction; rationale given.
• NEEDS-REFINEMENT: a divergence worth a future batch; suggested fix + roadmap era given (v0.2 = wind/weather era, v0.3 = procedural depth, Academy = the strict-mode toggle).

1. Radar separation minima

2. Wake turbulence: in-trail (radar)

3. Wake turbulence: same-runway departure intervals (TIME)

Documented bounds (§3).

• Super interval, under-modeled. Our modeled Super interval is 180 s (3 min), shorter than the real 4-minute behind-a-Super rule. That is a known under-model, flagged for correction. It is reachable only via the dev spawn hook (no scheduled A380 at any modeled field), so no normal-play session ever exercises it.
• No intersection-departure model. The 3-minute intersection/opposite-direction case therefore does not arise (noted in code for when it does).
• Heavy behind heavy is flat. A following heavy-behind-heavy is held at a flat 120 s where the real value carries CWT-pair nuance.

4. Successive / simultaneous departure release

4A. Parallel-final exemption (simultaneous/dependent parallel approaches)

Several modeled fields land parallel runways: KSEA feeds three N/S parallels (16L/16C/16R, sub-1-nm apart, the center-to-outer spacing is on the order of 800 ft), and KSTL runs 30L/30R about 1,300 ft apart. The flat 3-nm/1,000-ft radar minimum is the radar-environment IFR standard for a single stream. Aircraft established on different parallel localizers are not separated by that rule: they are separated by the approach operation itself (simultaneous independent approaches, or dependent staggered approaches at close spacing), where the runway geometry plus the localizer discipline is the protection.

Holding two correctly-established parallel arrivals to the raw radar rule is wrong: at KSEA every established neighbor sits inside 3 nm of the next, so the whole final complex paints red and a sustained false "loss of separation" can end a session no player action could have prevented.

So a pair is exempt from the radar LOS rule when both aircraft are established on the final approach course of different parallel runways of the same flow: distinct runway ends, courses parallel within ~1°, each still inside a tight established corridor (~0.5 nm cross-track of its own centerline) and inside the approach segment (from the ~12-nm gate out to the threshold). The exemption drops the flat lateral and the flat vertical (the operation separates the pair on both axes) and the wake in-trail requirement (the cross-final wake matrix is not the governing rule between two independent localizers). While either aircraft is not established (converging, JOINing, or being vectored), full minima apply, so two ships at the same altitude still cannot be blitzed toward parallel finals; they must be established (or vertically split) first. The backstop is automatic: a ship that departs its corridor loses established status and the full minima (flat + wake) re-apply and catch the geometry.

This applies identically on both clearance paths: the legacy one-step landing clearance, and the two-clearance approach flow (CA then L), where the aircraft holds the onApproach phase from the approach clearance all the way to the numbers. Establishment is read from whichever latch the aircraft carries, so the exemption, the corridor backstop, and the "not established means full minima" rule all behave the same either way.

5. Same-runway separation (tower, arrivals/departures)

5A. VFR separation regime (see-and-avoid)

VectorHeavy's traffic is a mix of IFR airline/GA and VFR aircraft (light singles in the pattern, VFR full-stop arrivals, VFR departures). The 3 nm / 1,000 ft radar minimum is the IFR-to-IFR standard. A VFR aircraft is see-and-avoid: it is responsible for its own separation from other traffic, receives traffic advisories and safety alerts rather than radar minima, and the controller cannot even assign it an altitude (an altitude command answers "unable, VFR"). Holding a VFR pair to the IFR 3 nm / 1,000 ft rule is wrong: it ended sessions for a "loss of separation" the player had no tool to prevent. Pairs involving VFR therefore use a lighter proximity regime, and a pair of pilot-owned VFR tracks is resolved by their autonomous see-and-avoid rather than charged to the controller.

The regime covers closed-traffic pattern aircraft for their whole circuit, not only on the arrival legs: a ship flying upwind, crosswind, downwind, base, or final at a small field remains part of the live traffic picture. An inbound VFR pilot automatically reports the airport in sight inside a modeled 12 nm visual range; only then does L become available. Outside Class B the controller assigns the runway and landing/option clearance, while the pilot chooses the visual path, distance, spacing, pattern entry, and heading; extend downwind and option/full-stop sequencing remain available pattern tools. A pilot may approximately use a published final course as a convenient geometric reference, but it receives no RNAV/ILS clearance and does not comply with procedure fixes, markers, or altitude restrictions. Class B is the heading-vector exception. VFR never accepts controller altitude, speed, or fix routing. Once L is accepted it is a durable landing goal: the pilot may fly away or around to create a realistic final and does not claim the clearance was missing. Pilot-owned VFR tracks continuously predict nearby traffic, but select and hold one maneuver through a commitment and clear-confirmation window instead of changing heading every tick. A controller heading or hold removes the autonomous-pair exemption; wake turbulence always remains authoritative. A runway conflict or controller go-around can still break off the attempt.

6. Line up and wait (LUAW)

7. Speed

Documented bounds (§7, the airspace-class speed rules).

• 91.117 is modelled as a PILOT rule, and that is the point. The section is a Part 91 operating rule — "no person may operate an aircraft…" — binding the pilot in command whether or not anyone is talking to ATC. So it lives in the profile the aircraft flies, beside the 250-below-10,000 auto-comply, not as a controller-assignable restriction. This is also why it does not contradict §16's "VFR altitude and speed are pilot-owned": a VFR pilot observing 91.117 is the pilot owning the speed. The only controller-side piece is the refusal to assign something 91.117(c) forbids.
• A charted procedure speed wins over the (c) clamp. A jet on a published descend-via whose procedure speed exceeds 200 while it happens to be beneath a Class B shelf keeps the procedure speed. That is a real divergence from (c), which has no authorization clause — and it is deliberate, on this section's own standard: the defect §7 was written to fix was applying a speed limit where it does not belong and making a published crossing unmeetable by construction. In the real system such an arrival is kept inside the Class B, which is an altitude-model question (the TRACON bubble is a cylinder and is not coupled to the shelves, §16/§17), not a speed one. NEEDS-REFINEMENT there, not here.
• No anticipation on the lateral edge. The 10,000-ft threshold is anticipated (the aircraft arrives at it on speed). The Class B shelf edge is not: an aircraft begins slowing when it is under the shelf, not before it crosses beneath. The edge is a lateral boundary rather than a planned descent point, so there is no equivalent lead computation; the effect is a few seconds of over-speed at the edge, disclosed rather than hidden.
• IAS vs TAS, unchanged. 91.117 caps indicated airspeed. Airspeed is a single number in a no-wind world here (§7, first row), so 200 is applied to the number the game carries. The same abstraction the 250-kt rule already runs on.
• "Above the surface" is field elevation. 91.117(b) measures AGL; the game has no terrain model (§8), so within 4 nm of the field the surface is taken to be field elevation. An approximation of the reference, not of the 200-kt number.
• VFR corridors are not modelled. 91.117(c) also names "a VFR corridor designated through such a Class B airspace area". No modelled field publishes one in the AIS dataset, and the game has no corridor concept, so that clause is simply not exercised.

8. Altitude floors: MVA / MIA

9. Approach vs landing clearance; missed approach

10. Holding

11. Handoffs / point-outs

12. ATIS

13. Phraseology spot-check

14. Published procedures: descend-via STARs (batch 2), climb-via SIDs (batch 3), two-clearance approaches (batch 4)

VectorHeavy flies published RNAV STARs with descend-via. A share of IFR arrivals at a field that has STARs (KSTL/KATL/KDEN/KSEA/KPDX) check in already established on a real charted arrival ("descending via the AARCH TWO arrival"), fly its leg sequence to the runway transition, and manage their own altitude/speed to the charted crossing restrictions until the controller takes them off it. The data comes from the CIFP cycle. The engine never invents a fix or a restriction.

Documented bounds (batch 2).

• Compressed altitude band → upstream crossings satisfied on Center. The game flies a compressed terminal band (e.g. KSTL ~2,000–12,000 ft), so crossings at or above the band ceiling (e.g. BOOSH at or above 14,000 at KSTL) belong to Center airspace upstream of the TRACON. The aircraft complied with them on Center frequency before the handoff and enters boundary-compliant, they are marked satisfied-upstream at assignment (deterministic, documented), never "ignored." That 14,000 crossing becomes the player's job in a future Center mode.
• An out-of-band MINIMUM still binds, clamped to the band ceiling. "Satisfied upstream" is the right treatment for a maximum the aircraft can never exceed inside the band. It is the wrong treatment for a minimum that is still ahead of the aircraft on the scope: dropping it lets the arrival descend straight through a published floor on its way to the next in-band crossing. So a floor published above the band ceiling is clamped to the band ceiling and held until its fix is crossed — the aircraft stays as high as the model can fly. What the player sees: at KATL (band ceiling 12,000) the ONDRE1 and CHPPR1 arrivals cross ONDRE / CHPPR at 12,000, where the chart says at 13,000 — 1,000 ft low, because 13,000 does not exist in KATL's modelled band. That residual gap is the band seam, and it is the whole of it: the aircraft is never lower than the highest altitude the model has. A floor sitting exactly on the band ceiling (KATL ENSLL / TIZZY at 12,000, KSEA AUBRN / PIKEZ at 12,000) is flyable exactly and is met exactly. Soak law starFloor asserts this at every field, and reverting either half of the rule makes it fire.
• Edge case, nothing in-band. A STAR whose every restriction is out-of-band at a given field still flies the lateral procedure (named, on the strip) but has no descend-via to fly: the arrival checks in normally ("with you, [altitude]") and is descended conventionally.
• RF-as-track (above) and speed windows flattened to their cap carry over from the batch-1 data bounds.
• Approaches / missed approaches are not this batch (two-clearance ILS + charted misses = batch 4; SID climb-via shipped in batch 3, below).

The departure side: published SIDs with climb-via (batch 3)

VectorHeavy also flies published SIDs with climb-via. A seeded share of IFR airline departures at the SID fields (KSTL/KATL/KDEN/KSEA/KPDX) file a real charted departure consistent with their exit fix / destination bearing; the takeoff readback carries "climbing via the COLLE SIX departure", and off the runway the aircraft flies the TRACON portion of the procedure. Same data doctrine: never an invented fix or restriction.

Documented bounds (batch 3).

• TRACON truncation. The flattened SID is cut at the exit-fix ring (~1.25× the exit distance + 5 nm); enroute legs beyond it, and every restriction at/above the band ceiling (including the charted top altitude), are satisfied downstream on Center. A future Center mode makes them the player's job.
• Edge case, nothing in-band. A SID whose every restriction is out-of-band still flies the lateral procedure and still reads back "climbing via" (the clearance is the clearance); the vertical machinery simply has nothing to level for, it climbs to the band cap as any departure does.
• Divergent-climb join (above): the dropped VA/VI/VM initial legs are stood in for by the existing divergent initial climb, then a nearest-sensible-leg join.
• Airline departures only; GA/VFR keep today's filed-direct behaviour (mirrors batch 2's arrival eligibility). Small fields carry no SIDs, correctly unchanged.

The two-clearance model, approach + landing clearances (batch 4)

VectorHeavy separates the approach clearance from the landing clearance, the core realism upgrade of the era. An approach clearance ("cleared ILS runway 30L approach", 7110.65 4-8-1) arms the approach; the tower's "cleared to land" is a distinct gate. A visual approach follows the "report field in sight" exchange. The family (ILS/LOC/RNAV) comes verbatim from the CIFP plates; RNAV(GPS) fields (KJEF/KSPI/KCOU) finally fly their real finals.

Documented bounds (batch 4).

• Localizer/glidepath model. The instrument intercept reuses the landing-phase localizer capture and a ~320 ft/nm glidepath, not a modeled ILS beam, angular sensitivity, or DA/MDA. That is the same geometric abstraction the landing phase already uses.
• Intercept geometry. The convergence leg is capped at the real 7110.65 5-9-1 intercept angle: ≤ 30°, and ≤ 20° when the join is close in. A firm minimum cut keeps an off-course aircraft driving decisively onto the centerline rather than creeping in parallel. A position needing a bigger course change first repositions to an altitude-sized energy gate, a shallow-join setup, instead of lunging at the localizer.
• "Maintain [alt] until established" is flown, since 2026-08-25. The clearance has always SPOKEN the preamble; until this date the aircraft ignored it and flew a continuous descent to a fixed 2,000 ft AGL shelf regardless of the altitude named. Measured before the fix: cleared at 4,000 ft from 14 nm, level at 2,000 ft within sixty seconds. Every arrival therefore converged on the SAME altitude, which removes the vertical split that makes simultaneous parallel approaches workable — reported by a player working the KSTL parallels, the field whose signature is three of them. The row above read FAITHFUL throughout, on the strength of the phraseology, while the aeroplane did something else. The aircraft now holds the altitude the clearance named until it is established, then the glidepath owns the vertical. It is clamped never to command a climb, so a ship already below the named altitude keeps descending as before.

• The established window. The aircraft is only tagged established (FINAL) once genuinely on the rail: within 0.2 nm cross-track of the extended centerline, aligned within 10°, and closing. The window is tight so that FINAL stays on the aircraft's own runway. At KSTL the 11/29 pair parallels the 30s ~10° off, and a looser window let a 29 arrival paint "established" while displaced onto the 30s' corridor. The same test gates the bare vs "…until established" clearance phraseology (4-8-1): a bare clearance is only issued when the aircraft is truly on course.
• Published offset finals remain offset. KCPS RNAV 12R flies the FAA CIFP JUPIS–RW12R rail at 107.3° true against the runway's 122.3° centerline, a genuine 15° offset. The procedure validator carries a named, procedure-specific 16° allowance for this one published rail; it does not rotate the runway or weaken the global 2° geometry law. PUBLISHED FACT / FAITHFUL.
• Charted missed approaches. Still the generic runway-heading climb (§9), a v0.3 depth item; the approach clearance's authorization of the published miss is modeled as that climb.
• Visual approach is pilot-managed own navigation that self-establishes a final via a gate: the aircraft maneuvers to a final-approach-course gate ~3.5 nm out on the approach side (identical geometry to the instrument recapture), sits at ~1,100 AGL there, then flies inbound on the final course to land, continuing straight in only when already on a plausible short final (aligned within ~45° of the course and below ~1,500 AGL). It is not a charted visual segment or a "follow the traffic you have in sight" sequencing, honest and simple, and it never crosses the field low without being established. A blow-through past the threshold obeys the landing-overshoot go-around rule below.
• Landing overshoot go-around. A cleared-to-land aircraft that flies past the threshold still airborne (blew through the final, or was cleared to land from the wrong side of the field) executes a pilot-initiated go-around rather than chasing runway heading outbound off the departure end while tagged FINAL. No score penalty (parity with the VFR own-energy / uncleared-arrival go-arounds); recapture is only modeled before the threshold, never from behind the field.
• Field-in-sight roll draws from a dedicated RNG stream (the delay + the distance-scaled success probability), so it never perturbs the STAR/SID or main seeded sequences and the replay gate is unchanged.
• VFR excluded. Approach and visual clearances are IFR-only; VFR aircraft are see-and-avoid and use the runway (L) directly, as before.

15. Radio voices: a platform capability note

This one is not an FAA-rules matter, it is a bound on where the game's audio runs, and it belongs here for the same reason the rest of the page does: so you know exactly where the model stops.

The radio speaks the full exchange, the controller's transmission and the pilot's readback, through a neural voice model that runs entirely on your own device. It loads and speaks on desktop browsers: Chrome, Firefox, and desktop Safari. It is not available on iPhone or iPad. On those devices the radio plays text-only: every controller call and pilot readback shows on the radio strip as text, the same full exchange the voices would speak.

15A. Shared regional geography

Every river and lake in the game is now generated from the USGS National Hydrography Dataset. None is hand-drawn, and none comes from a world-scale cartographic dataset any more. Nine canonical features, the Ohio, Missouri, Mississippi, Illinois, Arkansas, Kansas (Kaw), Columbia and Willamette rivers and Hillsdale Lake, are rendered by one generator (scripts/hydrography/genRivers.ts) from committed NHD extracts pinned by SHA256 in reference/usgs-sources/MANIFEST.md. Every field that shows a feature projects that one centerline through its own FAA ARP, so one physical river lands in one world place on every scope.

Rivers come from NHD "Flowline - Large Scale" (layer 6) and lakes from "Waterbody - Large Scale" (layer 12), queried by GNIS name. Public domain, cited as USGS. Nothing is fetched at build time or at runtime.

"Illinois" and "Illinois River" used to be two names for one river in two modules. NHD returns it as one connected chain of all 347 named segments from the Kankakee confluence to the Mississippi at Grafton, so it is now one canonical feature serving all six fields that draw it.

Still hand-drawn, deliberately, because no neighbouring scope reaches them and they are not claimed as sourced: a handful of small local features (KIXD's Clinton Lake, KICT's Cheney Reservoir and Little Arkansas, KTUL's Keystone and Oologah lakes, KSTL's Meramec, KSEA's water underlay, KDEN's South Platte). They remain OUR DISCRETION and are the natural next batch for this pipeline.

KPIA's official FAA field profile also identifies Army National Guard Chinook operations. Those helicopters are not represented: the sim has no rotorcraft flight or runway-use model, and substituting fixed-wing behavior would teach the wrong traffic picture. The documented 182d Airlift Wing C-130 operation is represented with the FAA's real C130 designator; its game performance figures, Reach callsign use, route choices and traffic weight are OUR DISCRETION.

16. Airspace class: the drawn boundary (the AIRSPACE map layers)

VectorHeavy draws the published Class B / C / D boundaries on the scope as selectable map layers — the field's own designation and its neighbours' surface areas, each toggled separately, both on by default — and the same geometry decides who ATC may vector. This section is about what that claims and what it does not.

What a real display does. A terminal radar display (STARS) draws its static background from configurable video maps (FAA Order JO 7910.1, Aeronautical Information Services Video Map Products), and those maps are selectable layers the controller turns on and off per position. Airspace boundaries are among the mapped features. So a toggleable airspace overlay is authentic in form. What we have not confirmed with a working controller is the real-world default state at a given facility, or exactly which boundary features a given facility's map set carries. Ours defaults ON; that is the founder's call, flagged here as a discretion call, not a claim about any facility.

Where the geometry comes from. The FAA Aeronautical Information Services "Class Airspace" dataset — the FAA's own published boundary geometry, the same data the agency charts Class B/C/D from (ais-faa.opendata.arcgis.com/datasets/class-airspace). The pipeline is scripts/airspace/fetch-airspace.ts (npm run gen:airspace), which emits src/data/airspace/<icao>.ts. Nothing in that directory is authored by hand.

The attributes were cross-checked against an independent primary source. For St. Louis, the Federal Register legal description (83 FR 14392, 2018-04-05, Amendment of Class B Airspace Description; St. Louis, MO, incorporated into FAA Order JO 7400.11) lists Areas A–M with floors SFC / 1,700 / 2,000 / 2,500 / 3,000 / 3,500 / 3,500 / 4,500 / 4,500 / 5,000 / 5,000 / 5,000 under a uniform 8,000 MSL ceiling. The dataset returns exactly those areas with exactly those limits, and the drawn Area A ring measures 6.00 nm from the published Point of Origin (38°45'10"N, 90°21'39"W) — the published 6-NM radius.

Coverage: every modelled field, real geometry, none invented.

Every shape above is the published shape. No field draws an approximated ring, and no field with unsourced geometry draws anything — that case does not arise today, but the data module is built so it degrades to drawing nothing rather than to drawing a guess.

Documented bounds (§16).

• The clearance does NOT extend vectoring authority. CB authorizes the flight to ENTER; it does not let the controller vector it while it is still outside. The containment rule above still owns that — a heading is accepted only while the aircraft is physically inside a Class B volume, cleared or not. Real 7-9-2 work routinely vectors a VFR onto a Bravo transition before it crosses the boundary, so this is an under-model, disclosed. NEEDS-REFINEMENT.
• The entry timings are OUR DISCRETION. No order says when a pilot keys the mic or when it turns. The pilot asks about two and a half minutes from the boundary on its present track, holds clear at ninety seconds, and re-asks every 90–150 s while it waits; the first ask follows the trigger by 4–12 s. Look-aheads are expressed in TIME and converted against the aircraft's own speed, because a fixed mileage that suits a Skyhawk is a few seconds of warning for a bizjet. Only the request delay draws randomness, from a dedicated seeded stream (bravoRng), so test:determinism and test:replay stay bit-exact.
• Holding outside is an orbit, not a published hold. The pilot uses the game's present-position hold. It is bounded and already modelled; steering along the boundary instead walks a light aircraft into the first concave shelf it meets and corners it there, which is a worse lie than an orbit.
• The entry model does not run for IFR. An IFR clearance authorizes Class B operation, so no IFR flight is asked for one. That is correct, and it means the §16 entry rows describe VFR behaviour only.
• The boundary projection is a level/target-altitude track, not a trajectory. It marches the aircraft's present heading at its present altitude AND at the altitude it is descending to (Class B is shelved, so a pilot can enter through the floor as easily as through the side). It does not model a turn radius, so an aircraft already within a few seconds of the boundary at high speed can cross before it can comply — which is the case the unauthorized-entry advisory reports.
• The 91.117(c) speed rule beneath the shelves now runs off this geometry (§7). The 91.117(b) Class C/D surface-area case is written but not yet flown; §7 carries the row and the reason.
• Vertical band edges are inclusive at both ends. An aircraft exactly on a published floor or ceiling is treated as inside that volume. The real limits are floor-inclusive and ceiling-exclusive ("up to but not including"). At the game's 100-ft altitude resolution this touches only an aircraft sitting exactly on a ceiling, and the effect is to resolve it to the MORE controlling class.
• Part-time airspace is drawn as though it were continuous. Several of these designations carry published effective hours; the game has one static picture and does not model a tower closing.
• Currency. The geometry is a snapshot with its retrieval date stamped in each data file, not a live feed. Re-run npm run gen:airspace to advance it.
• Class C's inner/outer structure is drawn as published, not explained. The overlay draws the two charted areas; it does not teach the outer area / outer circle distinction.

17. The departure model: SID mix, radar vectors, and the TRACON bubble

This section documents a model we researched and built, not a rule we can cite. Where a row says OUR DISCRETION it means exactly that: we chose a number that produces an honest-feeling operation and we are stating it plainly so a working controller can tell us it is wrong. None of the numbers in this section are presented as published FAA fact. The parts that ARE evidence are labelled.

What it replaced. Every outbound aircraft used to fly to a pre-determined exit node and then call for a frequency change. Two things were wrong with that. First, the node was picked from every named fix in the airport data, including fixes that exist only because they sit on a published STAR — so at KSTL a westbound departure filed SNYDR (7.6 nm, 254°), a point on an arrival descent, and the same fix carried an inbound and an outbound at once. Second, the exit node was the only way out, so "reach the fix" and "leave my airspace" were the same event, and the entire middle of a departure's life — the part where a controller actually works — did not exist.

17.1 Per-field SID mix

Bounds. The share governs airline/freight departures only. GA departures carry no SID at any field — that is the batch-3 eligibility bound carried forward, and it is a documented under-model: real bizjet traffic off KOJC or KIXD does fly the published procedures. Freight is not special: a cargo flight is an IFR airline departure and takes exactly the same path through this model.

17.2 Non-SID IFR departures fly radar vectors

17.3 Departures never route through arrival fixes

17.4 The TRACON bubble handoff

Bounds (§17).

• No modelled field currently spawns a non-pattern VFR departure. The VFR GA share goes to full-stop arrivals and closed-traffic pattern work. The VFR departure path is implemented and proven by a staged soak scenario (no node of any kind, exits by distance), but it is not organic traffic today.
• The bubble is a cylinder, not the shelved, sectorised shape of real delegated airspace, and it is not coupled to the drawn Class B/C/D geometry of §16.
• GA departures still take no SID (§17.1 bounds).
• The vector heading is a model, not a published departure heading for any runway at any field. A real facility's standard departure headings come from letters of agreement we have not sourced.

18. The comms compression ladder

How a pilot readback gets shorter when the frequency gets hard. Implemented in src/game/compression.ts; enforced by commsCompressionLaw in scripts/soak.ts.

The authority for this section is docs/research-ntsb-comms-patterns.md — ~45 primary sources, being NTSB dockets plus five FAA/Volpe/CAMI voice-tape studies (~250 hours of routine frequency tape with denominators, so the rates below are true rates rather than incident-biased ones). Its §6 BUILD SPEC is the specification this was built to.

It supersedes docs/doctrine-heimlich-reply-comms-model.md where the two disagree. That is the fidelity doctrine working as intended: a retrieved primary source beats recollection, including a working expert's. Rich Heimlich gave us the right shape — compression is real, it is measured in transmission length, borders drop and the core survives — and the wrong order. His example ladder ("Turn left 40 degrees for weather" → "Turn left 40 degrees" → "Left 40") sheds the reason clause first. The transcripts show the reason clause is the most durable token in the utterance. We did not implement his ordering.

Explicitly OUR DISCRETION in this model

Listed separately because these would fail an authenticity audit if presented as sourced, and the three-bucket rule requires them to be named:

• Any speech-rate change. There is no source for one, in any magnitude. The "~5% faster at one load threshold, ~10% at the next" figures in the expert doctrine are unvalidated — research §4.7(j) searched for them specifically and found nothing in either direction. Only one qualitative statement exists (Cardosi 2016: saturation "can result in controllers increasing their speech rate", no magnitude). The 120–180 wpm figures circulating online trace to forum posts, not studies. General cognitive-load speech literature actually leans the other way — load decreases articulation rate. We therefore implement no speed change at all in this batch. If one is ever added it is a designed feel choice and must be labelled as such.
• Every threshold number in the pressure terms: the complexity brackets (≤2 / ≤4 / ≤7 / 8+), the load brackets (45 / 70, inherited from the existing Contact Load bands), the time-available brackets (600 / 300 / 120 s), the +1 threatened step, and the 300-second courtesy boundary. The direction, precedence and endpoints of each are evidenced; the boundaries are ours.
• The spread of the measured "partial" mass across rungs 1–6. The studies report a single "partial" bucket; how that mass distributes over the individual drop rungs is our construction.
• The ~25% advisory-token rate is read off a single 12-case tally (Austin: 3 of 12 replaced by a token) and is treated as a design coefficient, not a rate.
• Everything about the EXPANSION half's shape. §6.3 names three constructions; it does not order them into rungs, does not say which party speaks them, and does not say when each fires. Ours: expansion is controller-led (justified above from §6.3a, but a judgement call); the three rungs are ordered callsign → restatement → explicit negative; the restatement is withheld from any instruction that mentions a clearance, a runway, a hold or a go-around; and the explicit negative is confined to a runway clearance for a runway the sim has verified is empty. That last confinement is stricter than the source — the source simply says "where doubt exists" — and it is deliberately so, because the failure mode of a generous reading is the game telling a crew a runway is clear when it is not.
• The 60–90 s TERMINAL SILENCE band. The corpus gives exactly one number, US1549's 75.7 s. A single anchor is not a distribution; the band around it is a design choice. So is the decision to measure it in seconds-to-ground (the nearer of the fuel deadline and range-over-groundspeed) rather than in fraction of the emergency, and the approach-commitment gate on the geometric term, which exists so a downwind abeam the threshold does not read as short final.
• Keeping the 7700 squawk at the moment of the event rather than at the declaration. During onset silence the transponder is therefore ahead of the radio: the player sees the emergency code before anybody speaks. The corpus does not settle the ordering. We chose it because it is the honest legibility cue — something has gone wrong and nobody is talking about it yet — and because moving the squawk would have quietly re-tuned scoring.
• NO TX as the legibility device. The research says nothing about how a game should surface a silence. Our rule was: no chatty "the crew is too busy to answer" line, because narrating the silence destroys it. A badge, the help panel and an AAR factor instead.
• The 2:1 bank-collapse tolerance in the soak law is a chosen engineering threshold, not an observation.

Named as UNESTABLISHED by the research, and therefore not modelled

The research doc searched for these and found no usable source. We have not invented numbers to fill the gaps:

• Speech rate versus traffic count — no magnitude exists anywhere (§4.7j).
• Disfluency density — no measured "ah"/"uh" rate for controllers or pilots in any source reached (§4.7k).
• Blocked / stepped-on transmission rate — effectively unmeasured. Cardosi 1993 found zero error-causing blocked transmissions in 47 hours (§4.7k).
• The multi-aircraft weather-saturation case — "the big one". No transcript was found showing six to ten aircraft all requesting deviations on one sector, which is precisely the scenario a compression ladder is about. Every transcript reached follows one aircraft (§4.3).
• The QUIET end of the duty cycle — the 1990s Volpe samples deliberately excluded very low workload periods as "yielding little interesting data". For a game built on the boredom-to-terror arc, half the arc has no literature at all, and the expert's memory remains the only source for it.
• Acute stress versus fatigue as separable speech signatures — no aviation-specific source separates them, and the Volpe studies state plainly that voice-tape analysis cannot assess workload or fatigue. We therefore use the better-evidenced THREATENED vs MERELY BUSY axis instead, and treat stress and fatigue as response flavours rather than as the primary driver.

19. Terrain is not modelled, and at KLAS that is the headline

§8 records the general omission: there is no terrain grid, no MVA/MIA, no MSAW, and the altitude floor at every field is a flat number rather than a terrain-derived one. That row has been in this ledger since the beginning and it applies everywhere. This section exists because at one modelled field the omission stops being a footnote and becomes the single most misleading thing on the scope, and burying it inside a general row would be the kind of quiet smoothing this page exists to refuse.

KLAS — Harry Reid International, Las Vegas. The field sits on the floor of the Las Vegas Valley, a bowl. The Spring Mountains rise to the west, with Charleston Peak at 11,916 ft — roughly 9,700 ft above the field and inside 25 nm of it. Frenchman and Sunrise Mountains stand immediately east of the airport at around 4,000 ft. The McCullough Range closes the south and the Sheep Range the north. Terrain is not scenery here; it is the thing that determines where aircraft may be put, which arrivals exist, how high the minimum vectoring altitudes are in each sector, and why the published chart set has no instrument approach at all to runways 08L or 08R. The FAA's own capacity profile for the field names it twice as an operational constraint — "high terrain to the west", and "military airspace, high terrain, and proximity to other airports reduce arrival and departure flows at LAS."

None of that is in the model. Concretely, at KLAS as at every other field:

• There is no terrain surface. The mountains are not drawn, not stored, and not consulted by anything.
• There is no MVA/MIA grid. The 5,000 ft band floor is arithmetic on the field elevation, not a vectoring altitude. It is not derived from, checked against, or reconciled with any real minimum vectoring altitude anywhere in the Las Vegas TRACON, and in several sectors the real MVA is far higher.
• There is no MSAW and no low-altitude alert. Nothing warns you, nothing goes red, and nothing ends the session. An aircraft descended to the band floor west of the field is, in the game, simply an aircraft at the band floor.
• The geography layer draws water only (Lake Mead, Lake Mohave). Its silence everywhere else is the absence of a terrain layer, not a statement that the ground there is flat.

What this means for a prep-oriented player, stated plainly: VectorHeavy's KLAS must not be read as teaching Las Vegas terrain avoidance, Las Vegas MVAs, or terrain-driven vectoring judgement of any kind. A controller who knows Vegas will notice the mountains are missing within a minute, and they are right to. Sequencing, separation, phraseology, the published procedures and the runway crossing this field is built around are all modelled and are all worth practising. The vertical picture is not. Treat every altitude you assign here as a number in a game, because in this model that is exactly what it is.

We are stating this rather than fixing it because the fix is a real subsystem (§8's roadmap item: an MVA polygon layer plus a low-altitude alert), and shipping a field this terrain-defined behind a quiet footnote would have been the dishonest option. The disclosure is the price of the field.

The same omission matters at KTBN in the Ozarks. Forney Field's surrounding ridges are not drawn, and the game has no terrain surface, MVA grid, or MSAW. The runway, Class D and procedure geometry are sourced; their presence must not be read as a claim that an off-procedure vector is terrain-safe. The omission is also operationally important at KBBG. FAA airport remarks explicitly warn of hills and uneven terrain near Branson and note that terrain avoidance may activate on final. Those hills, terrain alerts, MVAs and MSAW are not modeled. The sourced runway, Class D and procedures must not be read as a claim that an off-procedure vector is terrain-safe.

The omission is operationally important at KPDX as well. The Cascades and the Columbia River Gorge shape the real Portland terminal environment, but the game draws only the sourced river routes. It has no terrain surface, MVA/MIA grid, MSAW or low-altitude alert. KPDX's published procedures and runway flows are modelled; an off-procedure vector must not be read as terrain-safe.

KPDX's exact airline weights, 4% GA share, twelve-aircraft cap, spawn cadence, ramp and 90% STAR/SID assignment rates are OUR DISCRETION. The Port of Portland's 2025 statistics and current nonstop roster establish the carrier and market basis, not those game coefficients. Its east/west runway flows are published FAA capacity-profile facts. The Columbia and Willamette centerlines are USGS NHD facts; their displayed widths are OUR DISCRETION.

20. Aircraft types: what the FAA publishes, and what we made up

Every aircraft on the scope resolves through one table (src/data/aircraftTypes. generated.ts). That table decides the wake category, the slowest speed a pilot will accept, how long the aircraft holds the runway on rollout, how fast it can descend and change speed, and how many souls it reports on a 7700. It used to be hand-authored, and hand-authoring it cost us the worst honesty failure on this page.

The failure. Airline fleet lists and ga.types lists name type codes as bare strings, and nothing checked those strings against the table. Nine codes were referenced and never defined: A319 and A220 (KICT, KTUL), B739 (KATL, KDEN, KSEA, and the center's overflight mix), C25A/C25B/C25C (KICT, KIXD, KMKC, KOJC, KTUL), DH8D (KSEA), E170 (KMCI) and TBM9 (KICT, KIXD, KMKC, KTUL). Every lookup in that module falls back on a miss — wake class to large, approach-speed floor to 160 kt, landing distance to 5,000 ft, souls to 100–175 — so those aircraft flew as generic large jets. A nine-seat Citation CJ2 and a single-engine TBM held large-class wake separation, refused any speed below 160 kt, occupied the runway for a widebody's rollout, and could report 163 souls on board. This was live in production. It is fixed, and the fix is a pipeline rather than nine more hand-authored rows.

The source. The table is now generated from FAA Order JO 7360.1K, "Aircraft Type Designators" (date issued 2025-04-10, effective 2025-06-12; cancels JO 7360.1J), a US Government work in the public domain, retrieved from faa.gov and pinned by SHA256 in reference/faa-sources/MANIFEST.md. Appendix A of that order is the authoritative list of designators approved for use in the NAS, and it carries per designator the aircraft class, engine number-type / FAA weight class, ICAO wake turbulence category, Consolidated Wake Turbulence (CWT) category, Same Runway Separation category, LAHSO group, and every MANUFACTURER, Model that decodes to it. All 2,691 designators are transcribed verbatim into reference/faa-sources/jo7360-1k-appendix-a.tsv, which is committed — the pipeline is reproducible without re-downloading the PDF.

KBLV adds four sourced air-mobility designators to the live table: K35R (KC-135R), C17, C5M and LJ35 (the C-21 family). Their identities and wake categories come from FAA Order JO 7360.1K. Their game speeds, climb modifiers, landing distances and soul ranges remain OUR DISCRETION, under the same rule as every other type.

KPIA adds the sourced C130 designator for the 182d Airlift Wing operation at Peoria; its identity and wake category come from the same FAA order, its performance figures are game tuning like every other type.

Two codes we had been using were not designators at all. The check caught them the moment it existed: E175 is not in the order (the ERJ-170-200 US regionals fly is E75L, long wing; E75S is the short wing), and PA28 is not in the order (the fixed-gear Cherokee/Archer/Warrior family is P28A). Both were corrected everywhere they appeared.

What the source confirmed rather than contradicted. The CWT mapping reproduces every wake class the hand-authored table carried, including the awkward ones: the Gulfstream IV is CWT G and therefore large while the other business jets are H/I and therefore small, which the old table had reasoned its way to by MTOW. The hand work was right; it just had no gate.

21. Which airports are towered, and which Center owns them

The lobby groups every playable field under an ARTCC, and each Center section carries a "N of TOTAL" brick-progress note. Until now both halves of that claim were hand-typed: the field-to-Center mapping in src/data/centers.ts, and the total in docs/ZKC_BRICK.md, assembled by reading one airport's Chart Supplement at a time. The method was sound and nothing checked it.

What it cost. The hand-built ZKC roster said 26 civil towered fields. The FAA's own data says 33. Nothing in it was wrong — every one of the 26 is a real ZKC towered field, and every exclusion it argued (KLWC, KUIN, KAAO, KFLV all untowered; KOKC/KDSM/KOMA/KXNA/KDEC and the rest in neighbouring Centers) matches the FAA exactly. It was wrong by omission: seven fields were never enumerated, so they were never checked. Two of them (KPIA Peoria, KBMI Bloomington-Normal) sit inside Chicago Center's published boundary while Kansas City is the responsible Center — the exact case a by-eye boundary read cannot get right. One more (KBEC, Beech Factory) was excluded as "private use" when the FAA records it as privately owned but public use.

The fix. src/data/toweredAirports.generated.ts is generated from the FAA NASR 28 Day Subscription (effective 2026-07-09; pinned by size and SHA256 in reference/faa-sources/MANIFEST.md §7, committed extract at reference/faa-sources/nasr-2026-07-09-towered.tsv). Tower status is NASR's TWR_TYPE_CODE; the Center is NASR's responsible ARTCC, the value the Chart Supplement prints. 712 US towered airports across 25 ARTCCs, per-Center brick rosters derived rather than counted.

The gate was checking fifteen fields out of seventeen, and said PASS. validate:towered kept its own hand-typed array of airport modules while the registry carried seventeen. KSUS and KSGF were not in it, so their field elevations had never once been compared against NASR; validate:types carried the identical fifteen-entry array, so those two fields' airline fleet and ga.types lists had never been checked against the type table either. Both rosters are now derived from scripts/airportRoster.ts, and on their first run the two newly covered fields passed: KSUS 463 ft vs NASR 463.3, KSGF 1268 ft vs NASR 1268.3, both inside the ±1 ft rounding band, and neither field references a type the order does not define. Nothing was wrong; nothing had been looked at. The recurrence law is in validate:registry, and it now forbids the pattern rather than policing a list of files: any file under scripts/ that enumerates airport modules by hand fails the build unless it is registered as an explicit exception with a reason and a closed code list.

21b. What the public pages claim about the real world

The per-airport pages at /airports/<icao>/ carry hand-authored prose. validate:landing-copy holds every sentence about our data to that data, and it is honest in its own header about the half it cannot reach: a claim about the real airport that our data does not carry is invisible to it. Those claims are this section's job, and the standard is the same three buckets as everywhere else — a PUBLISHED FACT carries a citation, OUR DISCRETION says so, and UNKNOWN is never asserted.

A sweep on 2026-07-28 found seven real-world claims in the seventeen blurbs. Three are derivable from NASR (document 7, already pinned by SHA256 in reference/faa-sources/MANIFEST.md; the copy the derivation was run against hashed identical to the pin). Four could not be sourced to any primary FAA document and were cut, not softened into a vaguer number.

Three further sentences were examined and deliberately left: Wichita's "Air Capital of the World" and Tulsa's "Green Country" are civic nicknames rather than assertions, and Kansas City Downtown's "the metro's original airport" and New Century's "the old Olathe Naval Air Station" are historical background carrying no quantity. They are recorded here so the next sweep knows they were judged rather than missed.

A second sweep on 2026-07-28, with KALN, KBLV, KBMI, KCPS and KPIA, added one claim that is deliberately NOT a published fact and is labelled here instead of being cut.

This is a human sweep, and it does not become a gate. No token matcher can tell a sourced world claim from an invented one, and shipping something that looked like a gate would be worse than the honest gap.

21c. FAA facility levels (data carried, nothing driven by it yet)

src/data/registry.ts now carries faaLevel on each airport: the real FAA air traffic facility level, 4 through 12, the scale that ranks a facility by traffic count and complexity and sets its controllers' pay band. It exists for the career ladder in docs/CAREER.md, which wants real rungs rather than an invented difficulty number.

Nothing in the game reads this field today. The lobby's STEADY / BUSY / HEAVY tags are still the only difficulty grouping a player sees, and they are our own three-bucket UI shorthand, not the FAA scale. This section exists so that a sourced-but-unused field is disclosed as such rather than discovered later.

21. The shift clock: what time it is on your scope

Every shift opens at a local time, and the turnover briefing reads it out ("0614 local, morning shift"). Pilots greet you against it, and handoff sign-offs follow it. That clock keeps running on sim time, so a long session really does walk out of the morning push and into the afternoon.

Where the time comes from depends on the mode, and the game says so.

None of this weakens the determinism guarantee, and that is structural. The real clock is read exactly once, in the host, before the sim exists; the resulting minute is passed into the sim as an input and recorded on the proof tape, so a replay of an evening sandbox session is still an evening session years later. Nothing under src/game/ may read a wall clock at all — a gate greps for it and fails the build. Tapes recorded before the clock became a session input replay exactly as they always have, because their recorded time is the seed derivation, and a tape old enough to omit it falls back to that derivation. Both directions are mutation-tested in scripts/replayDeterminism.ts.

21a. How a shift ENDS: position relief

The clock's bookend. A shift opens with a colleague handing you the position; it now also closes with you handing it to the next controller (END SHIFT).

22. Regional geography is one world, not one sketch per scope

KMHK consumes the same canonical Kansas River centerline used by the Kansas City-area fields and projects it from Manhattan's FAA ARP; no field-specific river was drawn. The current canonical source trace ends east of Manhattan inside the widest KMHK scope, so the validator reports that visible endpoint. That is a DOCUMENTED SOURCE-EXTENT LIMITATION rather than permission to invent the missing upstream reach.

The St. Louis fields overlap. A bend in the Mississippi seen from Lambert must occupy the same real-world location when seen from St. Louis Regional, Spirit, Downtown, or Scott. Redrawing a river independently for each airport can look plausible in isolation while moving that bend several miles between scopes.

KSTL, KSUS, KALN, KBLV and KCPS therefore share one definition per river — the Mississippi, the Missouri and the Illinois River, each owned by exactly one module and generated from USGS NHD (§15A). Each centerline is converted through latitude/longitude into the airport's local nautical-mile plane, then the same smoothing and bank-width routine builds the displayed polygon. npm run validate:geography converts the finished polygons back to world coordinates and fails if any field disagrees, if two modules define one feature name, or if a field draws a locally adjusted copy of a shared feature.

22. Regional geography is one world, not one sketch per scope

The St. Louis fields overlap. A bend in the Mississippi seen from Lambert must occupy the same real-world location when seen from St. Louis Downtown, Spirit, Alton, or Scott. Redrawing a river independently for each airport can look plausible in isolation while moving that bend several miles between scopes.

KSTL and KCPS therefore use one canonical St. Louis regional Mississippi centerline. It is converted through latitude/longitude into each airport's local nautical-mile plane, then the same smoothing and bank-width routine builds the displayed polygon. validate:registry converts the finished polygons back to world coordinates and fails if the two fields disagree.

Summary of open refinement items

This ledger is maintained alongside the code. When a refinement item ships, its row moves to FAITHFUL. Corrections from the controller community are welcome.

Sources

• FAA NASR (National Airspace System Resources) 28 Day Subscription, effective 2026-07-09 — the tower status (TWR_TYPE_CODE), responsible and boundary ARTCC, public/private use, surveyed ARP and field elevation behind every airport in the towered roster and every per-Center brick count (§21), and — via APT_RWY.csv joined to APT_BASE.csv — the runway counts and runway lengths behind the three sourced real-world claims on the public airport pages (§21b). Retrieved from nfdc.faa.gov and pinned by SHA256 in reference/faa-sources/MANIFEST.md; the extract is committed at reference/faa-sources/nasr-2026-07-09-towered.tsv. No third-party airport dataset was used.
• ACI World (Airports Council International), 2025 final global airport rankings, released 2026-07-15 — the two world traffic claims on the public airport pages (§21b): Atlanta 1st by passengers (106,302,208) and Chicago O'Hare 1st by aircraft movements (857,392). The only non-FAA source behind a factual claim in the game, cited as ACI and never as FAA. Retrieved from aci.aero. ACI collects these returns from its member airports; no FAA publication ranks world traffic, which is why the claims were cut before this source was retrieved.
• FAA Order JO 7360.1K (Aircraft Type Designators), effective 2025-06-12, Appendix A — the designator, manufacturer, model, ICAO wake turbulence category and Consolidated Wake Turbulence category behind every aircraft type in the game (§20). Retrieved from faa.gov and pinned by SHA256 in reference/faa-sources/MANIFEST.md; the extract is committed at reference/faa-sources/jo7360-1k-appendix-a.tsv.
• FAA Order JO 7110.65 (Air Traffic Control), current edition, chapters 2 (phraseology, ATIS 2-9-3), 3 (3-9-4 LUAW, 3-9-6 same-runway, 3-9-7 intersection wake), 4 (4-8-1 approach clearance), 5 (5-5-4 minima/wake, 5-5-7 passing/diverging, 5-6-3 MVA, 5-7-1/-3 speed adjustment, 5-8-3 successive departures). Text via faa.gov ATC publications and the pointSixtyFive 7110.65 wiki mirror.
• 14 CFR 91.117 (Aircraft speed) and 14 CFR 91.131 (Operations in Class B airspace), eCFR.
• FAA Notice JO 7110.677 (Procedures for A380-800 and An-225 Aircraft) for the Super class handling and the 4-minute same-runway departure interval behind a Super.
• AIM 5-3-8 (Holding), 7-4 (Wake Turbulence), 3-2 (Controlled Airspace), 3-2-3 (Class B Airspace), 4-1-18 / 4-2-3 (initial radio contact).
• FAA Order JO 7110.65, Chapter 7 Section 9 (Class B Service Area, 7-9-2 VFR Aircraft in Class B Airspace) — the CLEARED TO ENTER / REMAIN OUTSIDE BRAVO AIRSPACE phraseology (§16).
• FAA Aeronautical Information Services, Class Airspace dataset (ais-faa.opendata.arcgis.com/datasets/class-airspace) — the boundary geometry drawn by the AIRSPACE map layer (§16).
• USGS National Hydrography Dataset, high-resolution "Flowline - Large Scale" (layer 6) and "Waterbody - Large Scale" (layer 12), public domain, via the National Map NHD MapServer: the route of EVERY river and lake the game draws as a shared regional feature, the Ohio, Missouri, Mississippi, Illinois, Arkansas, Kansas (Kaw), Columbia and Willamette rivers and Hillsdale Lake (§15A). Queried by GNIS name; the nine responses are committed under reference/usgs-sources/ and pinned by SHA256 in that directory's MANIFEST.md. Cited as USGS, never as FAA.
• Natural Earth, 1:10m Lakes + Reservoirs, public domain: KLAS's and KORD's lake geometry. No longer behind any river — the Illinois and the Ohio were sourced from Natural Earth until 2026-07-29 and now come from USGS NHD.
• FAA Order JO 7400.11 (Airspace Designations and Reporting Points), incorporated by reference into 14 CFR 71; 83 FR 14392 (2018-04-05), Amendment of Class B Airspace Description; St. Louis, MO — the legal description the KSTL geometry was cross-checked against.
• docs/research-ntsb-comms-patterns.md — the comms-pattern research pass behind §18, and the authority for it. Roughly 45 primary sources. The load-bearing ones: NTSB DCA23FA149 (Austin tower, the 13 clearance/readback pairs that settle the drop order), AIR-24-01 (JFK), DCA10IA001 (Northwest 188), DCA18MA142 (Southwest 1380), DCA09MA026/AAR-10/03 (US Airways 1549), AIR-18-01 (Air Canada 759); and the five voice-tape studies with denominators — Cardosi 1993 DOT/FAA/RD-93/11 (en route), Cardosi 1996 DOT/FAA/AR-96/19 (tower local) and DOT/FAA/AR-96/66 (TRACON), Bürki-Cohen (tower ground), Prinzo, Hendrix & Hendrix 2006 DOT/FAA/AM-06/25 (the key quantitative source), and Cardosi 2016 DOT-VNTSC-FAA (Kansas City ARTCC).
• FAA Order JO 7110.65 2-4-3 (Pilot Acknowledgment/Read Back) and 2-4-20 (Aircraft Identification), plus AIM 4-2-4 — the legal boundary between sanctioned compression and drift (§18).
• 123atc.com (facilities roster), retrieved 2026-08-26: the FAA facility level, 4 through 12, carried by eighteen of the forty modelled airports (§21c). 313 FAA facilities with identifier, type and level; its sibling contract-tower and DoD rosters carry no level column, which is the evidence that the other twenty-two of our fields have no level to carry. A secondary source, cited as 123atc and never as FAA, and the only reason it is used is that the FAA's own published figure is a band rather than a number.
• FAA Air Traffic Controller Workforce Plan 2026-2028 (published 2025-09-20): the primary source the 123atc levels were corroborated against before use (§21c). It groups facilities into level 4-9 and level 10-12 bands; all eighteen of our levels fall inside the band it assigns them, and validate:registry asserts that on every run.
• FAA Order JO 7910.1 (Aeronautical Information Services Video Map Products) — the selectable-video-map form the map-layer system mirrors (§16).