# Check airline incident reports: NTSB data, 90 days, insure or wait

Riley Quinn · November 18, 2026

> Run a 90-day NTSB lookback after an airline incident; one event with no repeat findings supports non-refundable booking when refundable-versus-insured pricing w

| Takeaway | Detail |
| --- | --- |
| Run a 90-day NTSB lookback before booking after an airline incident. | The rule is a 90-day NTSB/aviation-incident lookback for that carrier. |
| A single event with no repeat findings means book non-refundable. | The condition is single event with no repeat findings. |
| Decide with refundable-vs-insured booking price, not headlines. | The thesis says the price of a refundable-vs-insured booking is part of the decision. |
| Do not use unrelated incident databases as airline evidence. | Grounded examples include the AI Incident Database, Marsh cyber incident reports, fire system failures, Suncorp insurance, and motor insurance DAS. |

 Check the airline’s 90-day NTSB event history, then compare refundable-vs-insured booking prices before deciding whether to insure or wait.

 The answer changes if the 90-day lookback shows a single event with no repeat findings, which triggers the non-refundable booking rule.

## What NTSB Records Actually Tell You

 The mechanism you need to understand is the NTSB's public CAROL Query database at ntsb.gov: it is the searchable record you run by airline name and date range, and it is the reason a short lookback returns a very different kind of document than the headlines suggest. This section explains how CAROL works and why a preliminary report and a final report are not interchangeable pieces of evidence.

 When you search CAROL by carrier and date range, the first thing to notice is the document type attached to each entry. A preliminary aviation report typically appears within days of an event. It is a factual snapshot assembled early, before the investigation has run its course. A final probable-cause report can take 12 to 24 months. That timing gap matters for your check: a 90-day lookback will mostly surface preliminaries, so you must read what you find as unverified. Treat the preliminary as a notice that an event is under investigation, not as a finding about what happened or why.

 The label on the record is your next check. An accident in NTSB usage involves substantial damage or serious injury. An incident is a lesser event. A preliminary report may carry no finding at all. Those are three different things, and the label tells you how much weight the entry can bear. Do not upgrade a preliminary with no finding into a probable cause just because the event made news. Read the label first, then read the document, and keep the two categories separate in your notes.

 Here is the practical reading rule. For each entry in your date range, write down the carrier, the event date, the label, and whether the document is preliminary or final. If it is preliminary, mark it unverified even if the narrative reads confidently. If it is final, you have a probable-cause finding you can rely on for that event. The distinction is about timing and evidentiary status, not about whether the airline was at fault.

 CAROL also rewards a consistent search method. Use the airline name exactly as the database indexes it, then set your date range and scan the returned labels before reading any narrative. Because final reports lag so far behind the event, a recent event will almost always show up as a preliminary. That is expected. It is not a gap in the record and not a sign the database missed something. It is simply how the reporting timeline works.

 One caution on sources: incident databases are not all the same. The AI Incident Database, for example, indexes harms and near harms from AI system deployment, which is a different domain from aviation safety reporting. For an airline check, stay with the NTSB record and read its document types on their own terms. That keeps your decision anchored to the carrier's actual aviation event history rather than to a general incident feed.

![Overcast runway beyond rain streaked window tarmac distant aircraft](https://screenshots.mightytravels.com/article-images-ai/check-airline-incident-reports-ntsb-data-ai-ecb590bd.jpg)

## The 90-Day Lookback Method

 This section defines the exact 90-day window procedure and the repeat-event threshold that separates isolated noise from a pattern. The check is deliberately narrow: you are not trying to decide whether an airline is “safe” in the abstract. You are trying to decide whether the carrier’s recent operating record gives you a reason to change how you book after a headline you have already seen. Start with the date of the incident that caught your attention, then look straight back 90 days from that date. Use that same window for every carrier you are comparing, so you are not accidentally giving one airline a longer or shorter review period than another.

 Run the search by carrier family, not by the brand name on the ticket. On the NTSB’s CAROL Query database, filter to the last 90 days for the airline’s mainline operation and its regional subsidiaries. A Delta search, for example, should include Endeavor Air and SkyWest-operated Delta Connection flights, because regional operators are listed separately and are where many turboprop and regional-jet events sit. The same logic applies wherever the mainline brand sells flights operated by a separate carrier. If you search only the marketing name, you can miss the aircraft actually showing up in the record.

 Count events; do not read tone. A preliminary entry, a final report, and a short narrative all get counted the same way at this stage: one reportable event is one event. Do not upgrade an item because the wording sounds alarming, and do not downgrade it because the write-up is dry. The point of the method is to turn a messy public record into a repeatable count before you attach money to the decision. Keep a simple list with the date, operator, aircraft type, and phase of flight, then stop adding detail once the event is logged.

 Use this threshold: 0–1 reportable events in the 90-day window means book normally. That is the isolated-noise bucket. One event can be a genuine anomaly, a reporting artifact, or simply the ordinary background rate of a large operation showing up in a public system. It is not a clean bill of health, but under this rule it is not enough by itself to change your booking posture.

 Two or more events of the same system or phase of flight means treat it as a pattern. Two unrelated items are not the same as two engine-related aborts, two landing-gear events, or two occurrences on approach. The repeat has to share the system or the phase of flight. When that match appears, the rule is to either wait or insure rather than assume the first headline was the whole story. That is the practical line this method draws: isolated events stay in the normal-booking lane; matched repeats move you into the protect-yourself lane.

![The 90-Day Lookback Method — Check airline incident reports](https://screenshots.mightytravels.com/article-images-pixabay/check-airline-incident-reports-ntsb-data-73b2f93c.jpg)

## Insure or Wait: The Cost Math

 Once the lookback is clean, the booking decision is arithmetic, not anxiety. Put the same itinerary into three postures and compare them side by side: buy the non-refundable fare today, buy that same fare with a trip-cancel policy, or wait and re-shop later. The only new claim this section makes is a simple one: for most readers whose check shows zero or one isolated event, booking now non-refundable wins on cost, because the saved premium and the avoided fare drift usually beat the protection you are buying.

 Posture one is the plain non-refundable ticket. The added cost at checkout is zero, and the tradeoff is explicit: if you cancel for a covered reason that the fare does not allow, you lose the ticket value. That is a real downside, but it is a known quantity you can see before you pay. It is also the baseline every other posture has to beat.

 Posture two is that same non-refundable fare plus a standard trip-cancel policy. Do not guess the premium. InsureMyTrip is a comparison site that gives live quotes, so pull an actual quote for your trip cost and dates instead of assuming a percentage. Then read what the policy covers: a cancellation benefit is not the same as coverage for a schedule change you simply dislike. If the quote is higher than the amount you would lose by canceling, the policy is not buying you much.

 Posture three is waiting. Waiting is not free either. Open Google Flights and use its price history for the exact route and dates to see where the fare has been moving. If the fare has been drifting upward, a delay can cost more than the insurance premium you were trying to avoid. If it has been flat or falling, waiting has a real argument, but you still need to compare it against the certainty of the fare in front of you.

 Here is the decision rule. Run the three postures on the identical fare, not on three different flights. Write down the non-refundable price, the live InsureMyTrip premium, and the Google Flights price-history direction. If your 90-day check produced zero or one isolated event and no repeat findings, book now non-refundable: you keep the premium in your pocket and you do not carry the drift risk. Insurance is a rational purchase when the quoted premium is small relative to what you stand to lose and your plans are genuinely uncertain; waiting is rational when price history supports it. Headlines do not enter the math.

![Insure or Wait: The Cost Math — Check airline incident reports](https://screenshots.mightytravels.com/article-images-pixabay/check-airline-incident-reports-ntsb-data-51a03096.jpg)

## What the Data Does Not Establish

 This is the one place where the 90-day count rule stops being the whole answer. The rule holds when the carrier is U.S.-investigated and the event trail is clean; it breaks at the edges, where the record you need either lives somewhere else or was never written in the first place. Here is the when-it-breaks versus when-it-still-wins frame, stated once so you can apply it without re-reading the rest of the guide.

 **When it still wins: the same carrier, a U.S.-investigated event, no repeat findings.** Nothing in the foreign-carrier exception changes the isolated-event path. If the event falls inside the NTSB’s remit and the lookback shows a single occurrence without a repeat finding, the rule keeps its force. Do not let the existence of an exception elsewhere in the world talk you out of a clean result at home.

 **When it breaks: maintenance and near-miss signals.** NASA’s Aviation Safety Reporting System (ASRS) and FAA service-difficulty reports capture events that never become NTSB cases. A zero in the NTSB count can therefore coexist with a maintenance or near-miss trail that never crossed the threshold for an NTSB docket. The check is to run those secondary sources alongside the primary count and read them as a separate column, not as a replacement for it. A report in ASRS is a self-reported safety observation; an FAA service-difficulty report is an operational record. Neither is a finding of cause, and neither should be counted as a repeat event under the threshold rule.

 **When it still wins: the secondary trail is quiet too.** If the ASRS and service-difficulty checks return nothing that changes your read, the single-event conclusion stands. The point of the edge-case list is not to make the rule fail more often; it is to name the two places where a clean count can mislead you — a different investigator, or a different kind of record — and to tell you which substitute check restores the same decision logic.

![What the Data Does Not Establish — Check airline incident reports](https://screenshots.mightytravels.com/article-images-pixabay/check-airline-incident-reports-ntsb-data-b355d33f.jpg)

## Worked Example: Run the Numbers

 Illustration only—replace every symbol with your own quotes. Scenario: one carrier, one round-trip itinerary, a party of four, and a 90-day lookback that is already complete and shows a single isolated event with no repeat findings. No elevated pattern is assumed, so the decision turns on price rather than on a forecast of disruption.

 Let B be the lowest non-refundable round-trip fare for the whole party, R be the refundable premium (refundable fare minus B), I be the insurance cost for the same trip, and W be the expected fare change if you wait to book. The totals if you book now are: non-refundable B; insured B + I; refundable B + R. If you wait, the expected total is B + W, where W may be positive or negative.

 Step 1: compare the two protection products. Insured is cheaper than refundable when I < R; refundable is cheaper when R < I; they are tied when I = R. Step 2: compare either protection product with non-refundable. Non-refundable remains the lowest certain cost, so it wins when the expected cost of a covered cancellation is below both I and R. Step 3: compare waiting with booking now. Waiting wins only when W is negative and its absolute value exceeds the cheaper protection price; otherwise fare risk outweighs protection cost.

 For this example, with one isolated event and no repeat findings, the winner is the non-refundable booking: nothing in the lookback raises cancellation risk above the base rate, and B is the lowest certain cost. The break-even trigger is I = R—once insurance costs as much as the refundable premium, the two protections are interchangeable, and any wait decision should be tested against that common price.

| Posture | Paid now | Break-even trigger |
| --- | --- | --- |
| Non-refundable | B | Expected cancellation cost < min(I, R) |
| Insured | B + I | I < R |
| Refundable | B + R | R < I |
| Wait | B + W | W < -min(I, R) |

![Worked Example: Run the Numbers — Check airline incident reports](https://screenshots.mightytravels.com/article-images-pixabay/check-airline-incident-reports-ntsb-data-8c0bb9d4.jpg)

## And Booking Checklist

 This section alone walks the full five-step check on a hypothetical booking, with the cost of being wrong in each direction attached to every step. Picture a $780 round-trip fare, which is about $390 per direction, on a carrier with one NTSB preliminary incident report dated 40 days ago. The goal is not to decide whether the airline is “safe” in the abstract. The goal is to decide whether to book non-refundable, add insurance, or wait, using the same sequence you can run before any booking.

 Step 1 is the CAROL query. Run the airline name and the relevant date range in the NTSB’s public CAROL Query database at ntsb.gov, as described in the earlier section on what NTSB records actually tell you. In this hypothetical, the query returns one preliminary incident report from 40 days ago. Record the count as 1. Do not treat the headline as the finding; treat the database entry as the finding.

 Step 2 is the repeat-event check. Apply the 90-day lookback method already defined: count qualifying events inside the window and compare that count with the repeat-event threshold. Here, the count stays at 1, with no repeat finding in the window. Step 3 is the operational follow-up: check whether there is an FAA emergency airworthiness directive or a fleet pause tied to the issue. In this example, there is no FAA emergency AD and no fleet pause. That combination points to an isolated event, not a pattern that changes the booking posture.

 Step 4 is the insurance quote. A policy quote of roughly 5–7% of the $780 trip cost comes to about $40–$55: 5% of $780 is $39, and 7% is $54.60. If you insure and the outcome is that the event was isolated, you have spent roughly $40–$55 for coverage that would not pay on a nervousness cancellation anyway. That is the cost of wrongly insuring in this example. It is not a claim that insurance is useless in every case; it is the specific cost of buying it when the check resolves cleanly.

 Step 5 is the wait-cost check. On a route with rising prices, waiting can add $50–$150 to the fare. That is the cost of wrongly waiting here: you avoid an unnecessary insurance premium but risk paying more than the $780 round-trip fare you had in hand. The comparison is concrete: about $40–$55 spent on insurance versus $50–$150 of possible fare drift. On those figures, the clean result — count of 1, no repeat finding, no emergency AD, no fleet pause — supports booking non-refundable.

 Copy-usable checklist, in order, about 15 minutes: (1) run the CAROL query by airline and date range; (2) count qualifying events inside the 90-day window; (3) compare the count with the repeat-event threshold; (4) check for an FAA emergency AD and a fleet pause; (5) price the insurance quote against the fare-drift risk, then book non-refundable when the result stays isolated. Keep the arithmetic visible: $780 round-trip, about $390 one-way per direction, roughly $40–$55 for insurance at 5–7%, and $50–$150 of possible fare drift if you wait.

Also worth reading
 [Airline incident reports: 3 years](https://www.mightytravels.com/2026/11/airline-incident-reports-3-years-of-maintenance-flagsbook-or-switch/)
·
 [Stop booking flights on Tuesdays](https://www.mightytravels.com/2026/05/stop-booking-flights-on-tuesdays-because-new-data-shows-a-cheaper-day-to-travel/)
·
 [Taiwan airline award seats: verify](https://www.mightytravels.com/2026/10/taiwan-airline-award-seats-verify-2026-transpacific-premium-cabin-space-before-booking/)

## What to do next

## Frequently Asked Questions

 **What lookback period should I run in NTSB records before booking after an airline incident?**

 Run a 90-day NTSB/aviation-incident lookback for that carrier.

 **What finding triggers the non-refundable booking rule?**

 A single event with no repeat findings means book non-refundable.

 **What should I compare before deciding whether to insure or wait?**

 Compare refundable-vs-insured booking prices.

 **Which database should I use as airline evidence instead of unrelated incident databases?**

 Use the NTSB's public CAROL Query database at ntsb.gov.

 **How is CAROL searched for a carrier's event history?**

 Search CAROL by carrier and date range.

 **What is the caveat about preliminary and final NTSB reports?**

 A preliminary report and a final report are not interchangeable pieces of evidence.

## Quick answers

| What should you do before booking after an airline incident? | Run the NTSB/aviation-incident lookback for that carrier for the period specified above, then compare refundable-vs-insured booking prices before deciding whether to insure or wait. |
| --- | --- |
| When is it appropriate to book a non-refundable ticket? | It is appropriate to book a non-refundable ticket if the lookback shows a single event with no repeat findings. |
| What should guide your decision instead of headlines? | The price of a refundable-vs-insured booking should be part of the decision, not headlines. |
| What should not be used as airline evidence? | Unrelated incident databases, such as the AI Incident Database, Marsh cyber incident reports, fire system failures, Suncorp insurance, and motor insurance DAS, should not be used as airline evidence. |
| What is the NTSB's public CAROL Query database? | It is the searchable record at ntsb.gov that you run by airline name and date range. |

Canonical: https://www.mightytravels.com/2026/11/check-airline-incident-reports-ntsb-data-90-days-insure-or-wait/
Markdown: https://www.mightytravels.com/2026/11/check-airline-incident-reports-ntsb-data-90-days-insure-or-wait/index.md
