CODING-GUIDE v2 — gold corpus annotation (2026-08-02)

Supersedes the round-1 guide issued with the 2026-07-25 coder packets (FOR-PROFESSOR-2026-07-25/CODER-N/CODING-GUIDE.md). The label taxonomy is the frozen 2026-07-16 taxonomy and is unchanged — v2 adds the decision rules the PI ruled on 2026-08-02, incorporated verbatim from:

• docs/pi-ruling-function-axis-2026-08-02.md (legitimize vs inform-descriptive)
• docs/pi-ruling-escape-thresholds-2026-08-02.md (escape-label thresholds)
• docs/pi-ruling-artifact-rows-2026-08-02.md (platform-artifact eligibility)

Round-1 labels remain immutable; v2 governs adjudication of the round-1 disputes and any future coding round. Disagreement remains a measured quantity, not a failure: escape labels are valid substantive outcomes when the applicable evidence threshold is not met, and coders must not commit merely to increase agreement.

General evidence rule

Judge only evidence observable in the supplied text. Count a cue only when it adds independent information: repetitions, synonymous restatements, the mere presence of an actor, geography, topic, source signature, or an attributed actor’s statement without authorial adoption do not create additional cues.

actor_alignment

Select the story's primary beneficiary or alignment:

• russian-aligned
• aes-government-aligned
• france-west-aligned
• international-organization-aligned
• domestic-opposition-aligned
• other-local-partisan
• other-external-actor-aligned
• mixed-contested
• unaligned-unclear

Score evidence separately for each substantive alignment label:

• 2 points — direct cue: explicit endorsement or denunciation, an unambiguous call to support or oppose, authorial self-identification with an actor, or explicit assignment of political credit, legitimacy, blame, or illegitimacy.
• 1 point — indirect cue: a text-internal but inferential signal such as asymmetric favorable/hostile evaluation, hero–villain construction, selective causal credit/blame, or an approving/derisive visual-caption frame.
• 0 points: actor mention, bilateral contact, reported quotation, outlet or institutional signature, hashtag, language, location, or topic alone.

Apply the labels as follows:

1. Commit to a substantive alignment when one label reaches at least 2 points and no competing label that also reaches 2 points is within one point of it.
2. Use mixed-contested only when at least two substantive labels each reach 2 points and their scores differ by no more than 1 point. Co-presence, diplomacy, or cooperation between two actors is not by itself mixed alignment.
3. Use unaligned-unclear when no substantive label reaches 2 points. Text length is not a criterion: one explicit cue can justify commitment, while a long neutral bulletin can remain unaligned.

function

Select the story's primary communicative function:

• discredit
• legitimize
• mobilize
• intimidate
• deflect-deny
• polarize
• inform-descriptive
• unclear

Code the post’s dominant communicative function; factual form does not make clearly valenced framing inform-descriptive, but alignment or likely effect alone does not establish legitimize or discredit.

Decision rule: Code factual reporting as legitimize or discredit only when the post itself contains clear textual or visual evaluation, justification, normalization, celebration, blame, ridicule, or delegitimization that makes this the dominant communicative function; when valence is absent, incidental, or inferred only from the source, selected topic, beneficiary/alignment, or an attributed actor’s statement, code inform-descriptive.

Use unclear when no primary function can be determined reliably.

theme

Select the story's primary substantive theme:

• security-conflict
• foreign-presence-geopolitics
• governance-elections-law
• economy-sanctions-resources
• identity-religion-history
• information-media
• health-education-social
• migration-humanitarian
• environment-climate
• other-unclear

Theme is determined by the text’s primary subject, not by its actor alignment or by any secondary mention:

1. Commit to a listed theme when it governs the headline or lead and at least half of the text’s substantive propositions, or when it is the sole unambiguous proposition in a short post.
2. Use other-unclear when (a) the dominant subject falls outside the listed taxonomy, including ordinary sport, entertainment, advertising, or personal chatter; (b) the text contains no interpretable substantive proposition; or (c) no listed theme reaches half of the substantive content and none is clearly primary.
3. A passing reference to violence, a foreign actor, religion, health, or an institution does not determine theme. Do not force out-of-taxonomy material into the nearest available label.

Use foreign-presence-geopolitics when the foreign actor's role is itself the main subject; security-conflict when operational violence, military activity, attacks, or casualties are the main subject.

Required escape note

Whenever a coder selects mixed-contested, unaligned-unclear, or other-unclear, the note must identify the failed threshold in one line:

• mixed: <label A cue> / <label B cue>;
• unaligned: no label reached 2 points; or
• other: <dominant out-of-taxonomy subject> / unclear: <missing or ambiguous evidence>.

Eligibility — platform artifacts (ruling of 2026-08-02)

A row is ineligible for content-label scoring when the observable record does not contain an original, interpretable story proposition because it is either:

1. a platform-generated unavailability notice (tombstone-law-violation or tombstone-copyright); or
2. a footer/branding-only record whose remaining text consists only of the channel name, promotional furniture, flags/emojis, isolated hashtags, or a greeting and therefore supplies no substantive story to which all three coding axes can validly apply.

Adjudication instruction (round-1 disputes)

Apply these rules row by row to the 560 actor and 230 theme escape disputes; do not bulk-award disputed cases to the coder who committed. Preserve the two original labels, the adjudicated label, and the threshold note. Recalculate reliability only after adjudication, and report the observed result rather than treating the published ceilings as expected scores.

The 235-row function cell (coder 1 legitimize × coder 2 inform-descriptive) is adjudicated under the function decision rule above.

notes: optional free text outside the required escape note (e.g. why a row was hard, or a minority reading you considered).