Methodology — UN80 Initiative Readout, 29 July 2026

The source, and its status

The readout is built from one source: the official UN transcript of the meeting, obtained from the UN's transcripts portal.

Known transcription artefacts in this meeting, left uncorrected in the underlying transcript but not carried into the readout: the initiative's name is repeatedly rendered “UNAT”, “UNAD” or “UNHCR” instead of “UN80”; Jens Wandel is introduced once as “James Wondal”; Elizabeth Spehar appears as “Speja”, “Sbar” and “Bihar”; the Peacebuilding and Peace Support Office appears as “PBPSO”, “PPPSO” and “PBBSO”. Names and acronyms in the readout are normalised to their correct forms.

The pipeline

Why three separate analyses of one meeting

The meeting covered three reform packages on a single speakers' list. Each delegation had one three-minute turn and most used it to address all three. The readout could have treated this as one debate. It does not, for two reasons.

• The packages will diverge. The Secretariat committed on the record to separate consultation tracks for packages 18 and 21, and a separate September deliverable for 22. Future meetings will be package-specific, so each analysis is built to anchor its own sequence over time.
• The three axes measure different things. A delegation pushing hard on funding reform is not thereby pushing hard on a new results mechanism. Scoring them together would average away the distinction.

Because the packages were debated together, the readout also carries a cross-package comparison the separate analyses cannot produce — which delegations change posture between packages. Twenty of thirty-four do.

How positions were scored

The plots place each delegation by a score, so the horizontal position has to mean something defensible. It is produced by a fixed rubric applied to each delegation's own words.

The three axes

Each package has one scored axis, always “how hard did this delegation push toward one named pole of the package's central disagreement”. The pole is chosen before any delegation is scored.

Order of operations

• The normal analysis is done first — consensus, fault lines, poles. Scoring decomposes an existing claim; it never substitutes for the analysis.
• The rubric is instantiated from this transcript, with anchor quotes for each band, before any delegation receives a single point.
• Each delegation is then scored against those anchors only, reading its full statement, without looking at other delegations' totals or any expected ordering.

The rubric

Seven positive indicators totalling 100 points, plus negative markers subtracted. The indicators are categories of speech behaviour, which is what lets the same rubric work for a funding debate and a peacebuilding debate.

Reading the numbers

• Score is the displayed total out of 100, floored at zero. Net is the pre-floor sum including negatives, and is what orders the dots. Both are shown in the “Scores at a glance” table.
• The scale is ordinal, not interval. It supports “argued harder than”. A 76 is not twice a 38. Dots are spaced by rank, not by score value.
• Ties are kept and marked with =. A tie between delegations with very different profiles is a finding, not an error — El Salvador and Russia tie at the bottom of package 21 by entirely different routes.
• Briefers are not scored. The three Secretariat officials authored the proposals, so their support is definitionally maximal. They appear as dashed, dotted-outline dots pinned to the right.

Verification of the scores

How to read the plots

• Left to right is the score, banded into three fixed thirds — low (0–39), medium (40–66), high (67–100). Left is weakest, right is strongest, and this means the same thing on all three plots. Cut points are fixed thirds of the scale rather than quantiles, so a lopsided split is a finding about the debate rather than an artefact of the method.
• Colour is the grouping, and nothing else. Grouping and intensity are different questions and they frequently disagree, which is the point of separating them.
• Up and down is whichever dimension is selected in the dropdown. A delegation's column never changes when the vertical dimension changes — the column comes from its score.
• Click a dot to see every indicator, its points and the verbatim quote that earned them, including indicators recorded as “not stated”. Click a second to compare two delegations side by side.

The groupings

Five groups, coloured consistently across the prose, tables and all three plots so a reader can follow one delegation through the page.

• Western donors — EU member states, the UK, Switzerland, Japan, Republic of Korea
• CANZ and Pacific Islands Forum
• G77 and China members speaking nationally
• Structural sceptics — Russia, Iran, Cuba, China, who challenged the design of the proposals themselves
• Secretariat briefers — unscored

Groupings reflect how a delegation positioned itself in this meeting, not formal membership alone. The palette was checked under simulated deuteranopia and protanopia; the closest pair separates at ΔE 10.4. Colour is never the only carrier — dots print country codes and the legend names every group in words.

The “differs from camp” filter

Each Positions by delegation grid carries a filter that isolates the delegations whose score sits furthest from the rest of their grouping on that package. A delegation appears when its score is 20 or more points above or below its camp’s median for that package, and it carries a badge reading Pushed harder than camp or Pushed less than camp. Twelve of the eighty-nine delegation cards qualify: five on package 18, three on package 21, four on package 22.

The filter is computed from the scores already printed on the cards. It adds no new coding and infers nothing: a delegation is flagged only on the arithmetic distance between two numbers shown on the page.

What the badge claims, and what it does not. Each package scores a different axis, so the badge is read against that package’s own pole and nothing else. A delegation above its camp on package 18 pushed harder for structural change to the funding architecture than the rest of its camp did; on package 21 the pole is a genuinely new system-wide results mechanism, and on package 22 it is consolidated cross-pillar peacebuilding capacity. Because the three axes measure different things with different ceilings, the badge never compares one package to another, and a delegation flagged on one package carries no implication for how it argued on the others.

A low score has more than one cause. The axis is signed: a delegation arguing against the scored pole accrues negative markers and lands low, and so does a delegation that engaged the package lightly. Japan on package 18 (27) and Russia on package 18 (28) sit near each other for opposite reasons. The badge marks distance from a camp, not the reason for it. The stance line on each card carries that distinction, which is why the badge is never shown without it.

The brevity marker

A delegation whose statement fell below 70% of the meeting's median length carries a brief chip beside its name. The median was 369 words; the threshold 258. Two delegations qualify: Brazil (248 words) and Ethiopia (242). A short statement has fewer chances to trigger indicators, so the marker is a caveat on the score — but it is a fact about speaker behaviour, not a defect in the data. Ethiopia scored in the high band on package 18 precisely because it spent its whole short turn on one package.

Quote verification against the meeting audio

Because the transcript is machine-produced, every quote that carries analytical weight was checked against the recording rather than trusted.

Method. The English-language audio track was downloaded from the UN Web TV video platform. For each quote, the word-level timestamps in the transcript JSON gave a time window; that window was sliced out with ffmpeg and independently re-transcribed with a local speech model (whisper.cpp, large-v3-turbo, with anti-hallucination settings). The two texts were then compared.

The correction

The portal transcript rendered the Special Adviser's closing line on package 21 as “because the concept is It’s only generated as a concept” — a machine garble. The audio is unambiguous: “the concept is only generated as a concept”. The readout quotes the corrected wording. The claim it supports is unchanged.

What the near-misses were

Six quotes failed an initial automated pass and all six resolved as method artefacts rather than substantive errors: two were spelling conventions (the speech model writes “programs” and “Recommendation 5” where the portal wrote “programmes” and “recommendation five”), two were windows that cut the phrase at the slice boundary, and two sat 10–30 seconds away from their recorded timestamp and were found with a wider probe.

Limitations

• Interpretation. Cuba, Mexico, France, Switzerland, China and Senegal spoke through interpretation. Indicators E and G judge verb strength, which for those delegations reflects the interpreter's word choice as much as the speaker's.
• Statement length. Speaking time was capped at three minutes (five for groups) and enforced with an automatic microphone cut-off — Angola, Senegal and Pakistan were cut off mid-sentence. Group statements had the longer slot and a structural advantage on the specificity and material-commitment indicators. Flagged rather than corrected, because correcting would mean inferring what a delegation would have said.
• Package 22's compressed ceiling. The office being scored already existed, so the strongest available ask is smaller than in the other two packages.
• One meeting. This is a first waypoint for all three packages. It establishes a baseline; it does not show movement. Positions here are a reading of one session, not of a delegation's settled policy.
• No verbatim record. Informal plenaries produce none, so there is no official text to check the ASR against. The audio is the highest authority available, which is why quotes were checked against it directly.
• Not an official document. This is an independent analysis. It is not produced by, endorsed by, or affiliated with the United Nations.

Underlying files

The analysis behind the readout, for anyone who wants to check it:

The pipeline method itself is documented in the project's position-scoring and quote-verification guides, which govern every readout in this series.