Methodology — UN80 Initiative Readout, 3 September 2026

1. Source and its limits

What the ASR gets wrong here, and what we did about it

The recognition is reliable on structure and speaker order, and unreliable on acronyms. In this transcript the name “UN80” is rendered variously as UNAD, UNAT, UNATI, UNIT, UNITA, UNHCR, UNAIDS and UNIDIR. These are all the same initiative. Quotes used in this readout avoid the garbled tokens where a clean alternative sentence exists; where a quote must carry one, the surrounding sentence makes the referent unambiguous and the garble is not the load-bearing word.

Every quotation in the readout, and every quotation behind every score in the evidence panel, was matched character by character against the transcript, after normalising whitespace and curly quotation marks. That is 217 evidence fragments plus 10 talking points, all matching exactly and appearing exactly once. This confirms the readout reproduces the transcript faithfully. It does not confirm the transcript reproduces the room faithfully — see the next section.

2. Audio verification: attempted, not available

The problem the portal itself flags

The transcript JSON carries a warning field: “UN Web TV re-edited this video after the transcript was created; timestamps may no longer match the current video.” The recording was cut from 11,805 seconds to 9,226 seconds after the transcript was made, so every timestamp in the transcript points at the wrong moment in the published video.

That part was solved

The offset was derived empirically, by transcribing short probes at four fixed points across the recording and finding each probe's text in the transcript:

The offset is a constant front trim of about 1,260 seconds, stable from the start of the recording to the end. It matches the transcript's own first statement, which begins at 1,266 seconds — the cut removed the pre-meeting padding before the chair called the meeting to order.

The part that was not solved

The published recording carries one audio track. Requesting the English track, the Interlingua (floor) track, or any of the other 30 listed language tracks returns the identical file — same byte length, same checksum. That single track is an interpretation relay, not floor audio.

Probes on four delegations known to have spoken English on the floor confirm it. At the United Kingdom's slot, the audio renders as “for the preparation of the research and college personnel of the UN” where the transcript has the UN Institute for Training and Research and the UN System Staff College. The content is right, the speaker is right, the position in the meeting is right — and the words are an interpreter's paraphrase.

Comparing English quotations against an interpreter's rendering cannot confirm wording. It can only manufacture disagreements that look like transcription errors and are not. Correcting a quote on that basis would replace an accurate quotation with an inaccurate one, so no quote is marked verified and the readout says so wherever quotes appear.

3. Why this is an informal readout

Readouts in this series come in two shapes, and the shape is chosen before any analysis starts. A formal meeting — a Security Council session, a negotiation, a debate over contested text — exists to resolve disagreement, so divergence is the finding. An informal briefing or consultation exists to present work and collect reactions, so convergence is the finding.

This was a briefing. There was no text on the table, no outcome to negotiate, and no vote in the room. So this readout asks where the membership agreed, and treats what it does not agree on as open questions rather than as camp boundaries.

That choice has one concrete consequence for the numbers. A formal readout scores two mirrored rubrics — one for the case made for a position, one for the case made against it. This readout scores one. No delegation argued against UN80, so a for-and-against axis would impose a division the room was not having.

A related rule applies throughout: a question is never counted as opposition. A delegation asking how something will work has not taken a position against it. Detailed questions usually mean the opposite — that the delegation read the documentation closely enough to find the gaps.

4. The engagement rubric

Each delegation's score answers one question: how much did it put on the record about UN80 and what happens to it next? The score is not agreement, not support, and not influence. A delegation that argued at length for slowing the reform down scores high, because it argued at length.

Seven indicators, each scored only where a verbatim quotation earns it. Where a delegation did not address an indicator, it reads “not stated” and scores nothing. Nothing is ever inferred from silence, from bloc membership, or from what a delegation said at a different meeting.

Anchor quotations for each indicator were drawn from this meeting's own transcript before any delegation was scored, so the scale is set by the debate rather than by the scorer's expectations. Each delegation was then scored against those anchors without reference to any other delegation's total.

The brevity marker

Delegations speaking well below the meeting's median length carry a brief marker, because a shorter statement has fewer chances to touch an indicator. Word counts were counted mechanically, not estimated, and computed before any score was read.

The median across the 31 scored delegations is 394 words and the threshold is 276 words. One delegation falls below it: Tanzania, at 264 words. The chair enforced a three-minute limit with automatic microphone cut-off, which compresses the range and makes the comparison meaningful.

5. How the plot is built

The horizontal axis is quantity, never direction

Left to right means how much was put on the record: 0–39 points, 40–66, 67–100. Those cut points are fixed thirds of the scale, identical in every readout in this series, so a column means the same thing across meetings.

Direction never goes on this axis. A delegation arguing hard for slowing the reform sits on the right, because it argued hard. Which way it argued is carried by the vertical bands and by colour.

Colour is grouping, position is quantity

The four bands describe what each delegation asked the incoming session to do. They are not a scale from good to bad, and they do not correlate with the score.

These colours are reused unchanged from the July 2026 UN80 readout, where they were checked for colour-vision deficiency across every pair. Colour is never the only carrier of meaning: every dot prints a country code, every band is labelled in words, and the key names each group.

Who is not scored, and why

The Secretary-General, the two working-group co-chairs and the President of the General Assembly are shown on the plot with a dotted outline and no score. They present the material that the axis measures engagement with, so scoring their engagement with their own proposal would be meaningless.

Dot placement

Within a cell, dots are ordered by score and spaced evenly. Spacing is by rank, not by score value, because the rubric supports “more than” and “less than” rather than exact distances. There is no randomness anywhere in the placement; the same data always draws the same plot.

6. Attribution and naming

• One statement is misintroduced in the transcript. The chair introduces “the distinguished representative of Libya”, but the portal's own speaker record for that statement is Latvia, the statement aligns itself with the European Union group statement, and the chair thanks Latvia immediately afterwards. Libya did not take the floor. The introduction is a recognition error and the statement is attributed to Latvia.
• Russia and Belarus are treated as one speaker. The statement was delivered on behalf of both, under the Treaty on the establishment of the Union State, and says so in its opening line.
• Group statements are attributed to the group, with the speaking delegation named: Uruguay for the Group of 77 and China, Nepal for the Least Developed Countries, Australia for Canada, Australia and New Zealand.
• Partial personal names appear in the transcript for some delegates. The readout uses country and group names only.

7. What this readout does not claim

• The score is not a verdict. It measures how much a delegation put on the record, not whether it was right, influential, or representative of anyone beyond itself.
• Longer statements score higher. A statement with more sentences has more chances to touch an indicator. This is disclosed rather than corrected, because correcting it would mean guessing what a delegation would have said with more time. The brevity marker flags the delegation most affected.
• The scale is ordinal. It supports “more than” and “less than”. It does not support the claim that 80 is twice 40.
• Interpretation stands between several delegations and this text. Statements delivered in languages other than English reach the transcript through an interpreter, so the wording is the interpreter's.
• Silence is not a position. A delegation that did not address something is recorded as not having addressed it, never as opposing it. Delegations that did not speak at all do not appear.
• The quotes are not audio-verified. See section 2.
• This is a first political reading, not an official record. It is meant to be checkable: every claim traces to a named speaker and a quotable line, so a reader who knows the file can test it rather than take it on trust.