Week 11: Transcribe and diagnose

The method, stated plainly: how to transcribe honestly, and how to write a finding somebody who disagrees with you can check

17 May 2027

A methods lecture. Less argument, more craft, because the audit is due this week and the difference between a good one and a bad one is almost entirely method.

Transcribing honestly#

Mark what you cannot hear. [unintelligible], with a duration. This is the single most important habit and the one students resist, because a transcript full of gaps feels like a failure. It is the opposite: a gap is data, and a confident guess is contamination that the next reader cannot detect.

Transcribe what was said, not what was meant. If a speaker said thirteen thousand, write thirteen thousand. Normalising it to ONE THREE THOUSAND destroys the finding.

Separate the transcript from the analysis, physically. Two documents. Once you are annotating as you transcribe, you start hearing the annotation.

Say how many passes you made, and at what speed. A transcript is a measurement and measurements come with methods.

Timestamp everything. Overlaps are only visible with timings, and overlaps are this course’s subject.

Diagnosing against a standard#

The failure mode is diagnosing vibes. The discipline is three steps, in order, every time:

1. Quote the transmission. Exactly.
2. Name the standard form it departs from, and where that standard lives — the glossary, ICAO phraseology, a local supplement.
3. State the risk in this exchange. Not the general risk of the deviation. What could have gone wrong here, given who was on frequency and what they were doing.

If step 2 comes up empty, you do not have a deviation. You may have something real — a delivery problem, an equipment problem, a workload problem — and you should say which, but do not call it a phraseology finding.

Step 3 is where honest analyses admit that a genuine deviation risked very little on the day. Write that. It is a stronger finding than an inflated one, and it is marked as such.

Writing a checkable finding#

A finding is checkable when a reader who thinks you are wrong can locate everything you relied on and evaluate it without asking you.

That means:

• The quote is there, not summarised.
• The standard is cited, not gestured at.
• The inferential step — from what was said to what it risked — is visible as a separate claim, so it can be attacked separately.
• Your uncertainty is stated where you have it.

The last one is the mark of the thing. “I could not make out the level in the readback, so I cannot say whether it matched” is a better sentence than any confident alternative, and every accident report worth reading is full of sentences like it.

Findings versus blame#

Annex 13 separates investigation from liability: the purpose is prevention, not attribution. This course borrows the posture wholesale.

Write about the transmission and the system that produced it. Do not write about the character of the person transmitting. This is not squeamishness — it is that “the captain was arrogant” has no design consequence, while “the only available word for stop contained the word takeoff” has one, and the second is what a report is for.

Applying it to your own domain#

The audit asks you to do all of this somewhere nobody has done it for you. There is no glossary for clinical handover, which means step 2 has no standard to name — so you must first establish what the current protocol is, from observation, before you can call anything a departure from it.

That is the hard part of the audit, and it is where the 30% for evidence sits. Do not skip to proposing phrases. A proposal built on what you imagine people say is worth nothing, and it reads exactly like what it is.