Methodology and limitations

Beyond GDP: the UN launches the country phase — information session, 21 October 2026

Source status. The readout uses a screen recording of the first information session (89.5 minutes). The transcript comes from local speech recognition. It is not an official UN record, and the organisers have not reviewed it.

Quote checks. All 11 quote blocks and 17 short quoted phrases were checked against a second transcription of the audio. Two passages gave different words in the two passes. Those passages are paraphrased and carry no quote marks.

What kind of readout this is

This was an information webinar. UN staff presented a support offer, and the audience asked questions. No member state took the floor. So the readout does not map positions, fault lines or engagement. It records what the UN said, and it marks what is new against the three earlier Beyond GDP readouts.

The meeting is classified as informal. No position score was run, because there are no delegation positions to score.

How the readout was made

Step	Method
Transcription	Audio extracted to 16 kHz mono. Transcribed with Whisper large-v3-turbo on a local GPU, with voice-activity detection and settings that reduce repeated text. Nothing left the machine.
Speakers	Voice-based speaker separation (pyannote). Each voice was named from the Zoom name label on screen, the speaker’s own introduction and the published programme. Six speaker changes were checked against the video. All six matched.
Slides	One video frame every 20 seconds (269 frames). Slide text was read from the frames. Figures marked “slide” in the readout come from this step. Frames that did not show the webinar were left out.
Companion documents	Programme, concept note and expression-of-interest note from the UNCTAD event page. Dates and names were checked against them.
“What is new”	Each item was compared with the readouts of 4 March, 7 May and July 2026. Items the earlier readouts already carry are left out or listed as repeats.
Quote check	Each quoted passage was cut from the audio and transcribed again with different decoding settings. The two texts were compared word by word.

Choices and limits

Audience questions are not attributed. The askers are private attendees, and speech recognition does not spell their names reliably. The questions are numbered and paraphrased.
Some audio is unclear. Parts of the recording have dropouts and crosstalk. Where a passage stayed unclear after the second transcription, it was left out. One answer about Spain’s role in the process is not reported for this reason.
Speaker slips are corrected from the slide. One speaker said “seven out of seven”. The slide behind him reads “7 in 10”, so the readout uses the slide figure.
External claims are reported, not verified. The statement that multidimensional poverty rose for the first time is reported as the speaker’s statement.
Data coverage figures come from the presenters’ slides. The presenters call them “preliminary and indicative”. They count public international sources only, not national databases.
One session only. The second session on 22 October had the same programme. Its questions may differ and are not covered.
Times in the readout are minutes from the start of the recording, which begins at the welcome.