A briefing pack contains hundreds of NOTAMs. Three of them change what you do.
Investigators keep finding that the information was there — and that the way it was presented stopped it being used.
simplenotams is a deterministic triage engine. It sorts a pack into operational buckets, assigns a tier to every item, and shows you the reason. Nothing is summarised away, and nothing is graded by a language model.
The evidence
Presentation, not availability, is the recurring finding
Four extracts from official investigation reports. They are quoted because they describe the problem precisely — not because any of these bodies has any connection to this work.
“Acknowledging the known existing limitations of the NOTAM system, the NOTAM reviews by the flight crews of both aircraft were insufficient to identify that reduced runway length operations would be in effect.”
The NOTAM was fifth in a list of ten runway NOTAMs spread over two pages, in correct ICAO format, correctly delivered.
“…the flight crew’s misidentification of taxiway C as the intended landing runway, which resulted from the crewmembers’ lack of awareness of the parallel runway closure due to their ineffective review of notice to airmen (NOTAM) information before the flight and during the approach briefing. Contributing … Air Canada’s ineffective presentation of approach procedure and NOTAM information.”
The same report, at §2.3: “…the presentation of the information did not effectively convey the importance of the runway closure information and promote flight crew review and retention.” The closure NOTAM sat on page 8 of a 27-page release.
“…is not automatically obvious from the selection, but it becomes apparent if someone studies the NOTAMs package in detail.”
A picture that exists only in aggregate is invisible in a list. That is an argument for synthesis and for geometry, not for a longer document.
“…1.0 nm radius of the reference point but did not give the location of the crane itself.”
“On one day selected at random during the investigation, there were NOTAMs relating to 20 cranes in Central London standing up to 1,000 ft amsl.”
Twenty imprecise circles are not twenty usable obstacles.
On the use of these reports. No investigating body endorses this or any product. These extracts are reproduced with their report numbers so they can be read in full and in context. We make no claim that this or any tool would have altered the outcome of any occurrence — in several documented cases the information reached the crew and something else failed. The narrow, defensible claim is the one above: presentation, salience and delivery are repeatedly identified as factors, and that is the problem this work addresses.
What it does
Bucket, then tier, then generate the action
-
Parse
Every NOTAM in the pack is accounted for. Across six real briefing packs — 7,471 NOTAMs — the parser holds a 100% floor: N in, N accounted for, zero dropped. An item that cannot be understood is surfaced as unresolved, never silently discarded.
-
Bucket
Each item is placed by effect on the operation, not by Q-code convenience.
- AVOID
- CLOSED
- OBSTACLE
- PROC
- DEGRADED
- SERVICE
- SURFACE
- INFO
- GNSS
- COMPANY
- RESIDUAL
A separate bucket holds foreign-language bodies for translation, so they are visibly pending rather than quietly mis-sorted.
-
Tier
Fourteen named rules assign a tier, and the tier is phase-aware: what matters in the briefing room is not what matters airborne.
-
Generate
The tier is necessary, but it is not the point. What the tier produces is: the confirmation the crew is required to make, the item carried forward into the next phase of flight, the obstacle field drawn as bounded geometry on a chart rather than listed as twenty circles.
- Critical
- Changes a decision. Cannot be scrolled past.
- Operational
- Changes how the operation is flown or planned.
- Advisory
- Worth knowing; does not by itself change anything.
- Noise
- Retained, addressable, out of the way.
Every item keeps two tiers — one for the briefing room, one for airborne — so an item can be de-emphasised on the ground and re-raised in flight, or the reverse.
Why deterministic
Every tier traces to a named rule
There is no language model anywhere in the live classification path. Classification is rule-driven, so the same pack produces the same output on any day, on any machine, and each decision can be traced to the rule that made it and the text that triggered it.
This matters commercially as much as operationally. A tool that grades safety information by inference cannot tell you why an item was demoted, cannot be regression-tested against a fixed corpus, and cannot be audited after the fact. A tool built on named rules can be shown to a safety department, argued with, corrected, and re-run.
Correction is part of the design. Sixteen classification rulings have been taken from a qualified line pilot and encoded as rules; where a ruling contradicted the engine, the rule changed and the whole corpus was re-run.
| Measure | Figure |
|---|---|
| Real briefing packs used for calibration | 6 |
| NOTAMs parsed, in / accounted for / dropped | 7,471 / 7,471 / 0 |
| Conflicts against the corpus-mined expectation grid * | 443 → 81 (−82%) |
| Named tiering rules | 14 |
| Pilot rulings captured in the decision ledger † | 29 |
* A conflict is a disagreement between the classifier and an expectation grid mined from the corpus itself. Both endpoints are counted directly from the record sets, but this is not a like-for-like measurement: the grid was regenerated (v0.2 → v0.4) between the two counts, so the figure reflects the shipped system end to end rather than an improvement to the classifier alone. Nor was the path monotonic — conflicts rose from 153 to 210 at one point during the work before falling.
† Rulings made by the pilot and recorded with their reasoning. The ledger holds 33 numbered entries in total; the other four are engineering findings raised against the system itself, or questions still open, and are not counted here. Capture is not the same as encoding: each ruling is recorded with the evidence behind it, and most but not yet all are implemented as deterministic rules.
Status
In development
simplenotams is being built and calibrated against real briefing packs with a qualified commercial pilot. It is not available for general use, it has no operational approval of any kind, and it is not certified, approved or endorsed by any regulator or investigating body.
It is not a source of aeronautical information and is not a substitute for the official briefing. It reorders and annotates a pack you already hold; the pack remains authoritative.
If this is a problem you recognise
We are interested in talking to flight crew, flight operations and safety departments about how packs are actually read in the time available — particularly where you think the tiering above would get it wrong.