August 27, 2006, lagging airport, atc and aircraft design system caused Comair flight 5191 to crash while attempting to depart on a runway that was too short in Lexington, Kentucky. Forty-nine lives were lost. NTSB has the Federal mandate to determine probable cause, the Air Line Pilots Association does not. Remember Reason's swiss cheese? The Reason/Swiss-cheese reframe that the party submissions (ALPA especially) pushed against the NTSB's "flight crew failed to..." probable cause. The Board's finding put the pilots as the proximate cause; the party process documented the latent conditions that pre-loaded the trap and stripped away the redundant defenses, so that a normal-sounding taxi error became fatal. Here's the map, by layer.
Stack them and the picture the parties drew is unmistakable: a first-encountered, unlit, crowned, chart-mismatched runway, reached by a freshly-changed taxi route, with the external cross-check (a second controller) administratively removed and no cockpit position-alert installed — every redundant defense degraded before the crew ever moved the thrust levers. Under Reason's model, the pilots didn't cause the accident so much as inherit an already-holed system and become its last, failed defense. That's precisely the substitution-test result: swap in a rested, competent crew and, on a bad enough morning, many would take the same first runway.
The active failure (the last slice of cheese)
Comair 5191, CRJ-100, Blue Grass (KLEX), 0606 local, 27 Aug 2006. Cleared for Runway 22 (7,003 ft); taxied onto and departed Runway 26 (3,501 ft); overran, struck terrain/trees; 49 of 50 fatal, FO Polehinke sole survivor. NTSB probable cause: failure to use available cues to confirm position + failure to cross-check; contributing: non-pertinent conversation during taxi (sterile-cockpit breach).
That's the last line of defense failing. Everything below is what the parties argued had already failed upstream.
Airport design / surface geometry
Reconfigured taxi route from recent construction. The airfield had just been repaved and Taxiway A realigned. The new routing took aircraft across the Runway 26 threshold to reach Runway 22 — 26's entrance came first and sat right where the crew's mental model expected the run-up to the active. The geometry itself created the ambiguity.
Humped/crowned Runway 8-26. The crest hides the far end from the threshold, so the "this runway is too short" cue was not visually available from where they lined up. You can't see the deficiency you're about to launch into.
Signage/markings degraded or transitional during construction — hold-position and surface-painted guidance were part of the party critique of positive-position cues.
Runway lighting
Runway 26 edge lights were OFF (it wasn't the active); Runway 22's were on. Taxiing into a black hole was itself an anomaly — a crewmember audibly remarked on the darkness/lack of lights — but it was rationalized and pressed. ALPA's HF read: the lighting state was a negative cue (absence of expected light) rather than a positive alert, and humans are poor at catching missing cues under time pressure and expectation bias.
Charts / Jeppesen currency
The airport diagrams (both government and Jeppesen) lagged the construction. The chart in the cockpit did not fully depict the as-built taxiway state the crew was driving through, so the one tool meant to rebuild ground picture disagreed with the pavement. Chart-to-world mismatch is a classic latent condition — you can brief perfectly off a chart that's already wrong.
NOTAMs / ATIS
The taxiway/construction change was covered by NOTAM, but buried in the volume and format — the parties' recurring complaint that safety-critical NOTAMs are undifferentiated from noise (this case is often cited in the later NOTAM-reform push). "Right information, wrong salience" — which is your "right info, wrong time" refrain for this case.
ATC staffing / the tower (the redundant human defense that wasn't there)
One controller doing two jobs. A single controller worked both tower and radar/TRACON positions, contrary to an FAA staffing directive requiring two controllers on that midnight shift. He issued the takeoff clearance, then turned away to perform administrative/radar traffic-count duties and did not monitor the takeoff roll — so the wrong-runway lineup went uncaught. NTSB judged that even a second controller might not have been looking at the right moment, but ALPA/NATCA argued the staffing violation removed the last external cross-check and reflected fatigue (the controller had very little sleep). That's a management/latent failure, not a controller-skill failure.
Aircraft / cockpit design
No runway-awareness technology. RAAS (Honeywell Runway Awareness and Advisory System) and moving-map/position-awareness tech existed and could have annunciated "on 26" or a short-field warning — not installed. The absence of an engineered alert meant the only wrong-runway detector left was the two tired humans.
Heading cross-check not procedurally forced. The HSI would read ~260 vs assigned 220 — a 40° disagreement — but nothing in the SOP required a heading-to-clearance confirmation call at the hold. The design/procedure didn't make the cheapest cross-check mandatory.
Organizational / physiological
Early-morning, back-of-clock departure — circadian low for both crew after a minimum rest period, at the time 8 hours vs current 10 and the lone controller; fatigue as a performance-degrader across all the humans in the loop simultaneously.
Sterile-cockpit deviation as normalized practice, not a one-off character flaw — the HF framing is that non-pertinent taxi chatter was industry-common, i.e., a latent normalization, which is why the fix is systemic (procedure/culture), not "these two talked too much."
The through-line
Stack them and the picture the parties drew is unmistakable: a first-encountered, unlit, crowned, chart-mismatched runway, reached by a freshly-changed taxi route, with the external cross-check (a second controller) administratively removed and no cockpit position-alert installed — every redundant defense degraded before the crew ever moved the thrust levers. Under Reason's model, the pilots didn't cause the accident so much as inherit an already-holed system and become its last, failed defense. That's precisely the substitution-test result: swap in a rested, competent crew and, on a bad enough morning, many would take the same first runway.
Or you could argue that the runway was too short, a universal design flaw.
