Political discourse reading lab
2026-04-09 — ML and Cursor Ai
This lab is a first pass—not a definitive system and not one-size-fits-all. It is meant to join an ongoing conversation about how we read what circulates around us, rather than close it.
Take a moment before the outcomes and prompts: settle attention, then look at the wording itself—who speaks, what is narrowed, what is implied. The aim is not to pretend language is bloodless; it is to notice which words do the pushing, and to hold them in view with something like even weight.
What this lab leaves open on purpose. It does not tell you what “really happened” from a sentence alone. It trains you to see more than one honest reading of the same line, and to name what would decide between them when evidence exists. That gap is the workspace.
Process, briefly. Older mistakes here were simpler: taking lines at face value, and giving too little weight to other moves actors might make—moves that can look illogical until you widen the board. Much public talk now speaks past other public talk: people argue from headlines of outcomes while the substance of claims stays unjoined. One hears, for example, U.S. messaging that the Strait of Hormuz is shut and closed for traffic, while other sources track rising transit—both strands can be in play at once, and the reader still has to ask what each sentence is doing, on what evidence, for which audience.
An encounter. A friend once said there was a high probability troops would go in. The sentence landed as certainty. What felt missing was room for countermoves—for an opponent’s logic that might look “wrong” from one seat and forced from another—and for a picture of the situation rounded enough that probability wasn’t doing covert work. This lab is partly built from that kind of friction: not to win the argument, but to slow the leap from phrase to world.
What this lab is not. It is not a fact-check drill and not a machine for instant verdicts. It is a way to deepen thinking—especially for political and risk-flavored reading of macro global affairs—constructively, from the text outward.
Tempo. Move fast where the job is recognition (patterns: attribution, scope, fault verbs). Move slow where the job is tolerating ambiguity—interpretive forks, hinge, debrief, evidence notes—without collapsing them into a single takeaway.
If an LLM is in the room: mark clearly what is model-generated (lists, paraphrase, draft forks) and what is your judgment—where you agree, edit, or reject. The point is not to outsource thinking; it is to separate retrieval from decision.
Self-contained document. Everything needed for teaching and advanced analysis is in this file: learning goals, concepts, student materials, worked and practice examples, rubric, instructor notes, and all copy-paste prompts (lite, standard, compact, full 18-section analyst, and compact analyst).
Nature of the exercise: interpretive close reading of wording—not fact-checking unless you add a separate assignment with primary sources. The unit also teaches source in three senses (§2.1): provenance of the clip, journalistic sourcing in the wording, and primary evidence for verification—so “source” is not used only to mean “the official who spoke.”
Time: about 45–60 minutes (demo 15 min, guided
practice 20–25 min, debrief 10–15 min).
Materials: this file only; LLM or partner optional;
highlighters optional.
A note on humility: teaching and reading in public are humbling—how a line will land depends on others’ histories and stakes. Any analysis here risks sounding thinner than the world it points at. The lab still asks for care with the words on the page.
After this lab you should be able to:
| Idea | Plain language |
|---|---|
| Attribution chain | “The paper says that X said that Y”—each link can break or blur. |
| Scope restrictor | Words that shrink what counts (only, merely, not part of). |
| Cognition / fault verb | Verbs about minds and mistakes (misunderstood, denied, claimed)—often compete with blunter alternatives (disagreed, rejected, violated). |
| Interpretive fork | Two readings both allowed by the text; hinge = what you would need to know to choose. |
| Sense | What it means | Questions to ask |
|---|---|---|
| 1. Provenance of the clip | Where this text came from as an object you are reading | Outlet (who published)? Date and time? Headline / section? Byline or wire (Reuters, AP, pool)? Original language? URL or print page? Without this, you cannot judge timeliness, desk, or republication chains. |
| 2. Journalistic sourcing | How the story backs its claims (named official, unnamed official, document, eyewitness) | Does the sentence show on-the-record attribution, “sources said”, leaked material, or press conference? Who could have planted or shaped the line? (Still read the wording—but know what kind of relay it is.) |
| 3. Primary evidence | First-order materials that could verify or falsify the substance | Treaties, laws, full transcripts, verbatim quotes, court filings, data releases, original video/audio. Your evidence note is about this layer—not the same as “outlet X published it.” |
This lab still prioritizes close reading of the sentence; provenance and sourcing sit beside that, not instead of it.
Rule: Do not invent events, dates, or “what really happened” unless you have primary evidence in front of you. The skill is reading the wording—while knowing what you are missing if provenance or sourcing metadata was stripped off.
Paste into an assistant with your passage at the bottom, or use the same structure on paper.
You are helping someone read news-style reported speech carefully. Analyse ONLY the passage the user gives. Do not invent facts, quotes, or events.
Return these sections in order:
1. **Parse** — (a) **Provenance:** If the user supplied metadata (outlet, date, headline, byline, wire/agency, language, URL), summarize it and note what it implies for **timeliness** and **relay** (e.g. wire vs staff). If **not** supplied, list what should be recorded for real clips and state that reliability of the **chain** cannot be judged from the naked sentence. (b) **Journalistic sourcing:** What does the passage show about how the claim is backed (on-record speaker, paraphrase, “sources said”, document—only what the words support)? (c) Who the **narrator/reporter** is (if any). (d) What is **directly asserted** vs **attributed** (“X said …”). (e) Each **distinct claim** in logical order.
2. **Key words and phrases** — 4–8 items: dictionary sense plus **why they matter for meaning or blame** (e.g. scope: “only,” “not part of”; cognition/fault: “misunderstood”; agency: who is cast as mistaken vs informed).
3. **Fair interpretive forks** — **Reading A** and **Reading B** (and **Reading C** if clearly warranted), each defensible without smuggling in off-text facts. For each, say what it implies about facts, blame, or obligations. End with **what would hinge the difference** (definitions, missing text of a deal, prior statements, etc.).
4. **How to read similar passages (learner checklist)** — **8–12 numbered habits** (one sentence each). Each habit must be reusable on *other* news lines, and must cover at least: **provenance** (record outlet/date/wire); **journalistic sourcing** type if visible; **attribution** (who speaks in the text); **atomic claims**; **scope** words; **cognition/fault verbs** vs neutral rewrites; **who defines** loaded nouns; **implied audience**; **omissions**. After each habit, add **one** italicized line: *Here: …* citing **specific words** in *this* passage (or *Here: not visible in clip* for provenance if stripped). **Transfer**, not recap of sections 1–3.
5. **Evidence note** — What **primary** material would reduce ambiguity (documents, full transcript, verbatim quote, data)? What **corroboration** from **other outlets or official channels** would you seek? One short paragraph.
Source text (optional: paste **provenance** above this line — outlet, date, headline, byline, URL):
[PASTE PASSAGE HERE]Provenance (deliberately omitted for this demo) — In real work you would record outlet, date, headline, and whether the line is wire or staff copy. Here we only have the sentence; do not infer a publication.
Source text
Vice President JD Vance said the cease-fire focused only on Iran. Lebanon, he said, was not part of the deal, and the Iranians misunderstood.
| Word / phrase | Why it matters |
|---|---|
| said (repeated) | Marks attribution; distinguishes reporter’s voice from VP’s message. |
| cease-fire | Loaded noun: who is party, where it applies, what counts as violation—none defined here. |
| focused only on | Scope restrictor: shrinks the topic to Iran; implies other theaters are out of frame. |
| Lebanon, he said | Insistence (repeated attribution): stresses Lebanon’s exclusion. |
| not part of the deal | Boundary work: limits obligations; “the deal” is treated as a fixed object. |
| the Iranians misunderstood | Cognition/fault: casts the problem as Iranian error; competes with “disagreed,” “read the text differently,” or “responded to ambiguous signals.” |
What would hinge the difference: The actual text or communiqué of “the deal,” prior statements by any party, timeline, and what Iran publicly claimed—none of which appear in this sentence.
Trace who speaks: Separate reporter from quoted
or paraphrased source.
Here: “Vice President JD Vance said
…”
List separate claims inside one attributed
block.
Here: Iran-only cease-fire; Lebanon excluded; Iranians
misunderstood.
Circle scope words that shrink geography,
topics, or obligations.
Here: only on Iran; not part of
the deal.
Stress-test cognition verbs by trying a neutral
rewrite.
Here: “misunderstood” vs “interpreted the agreement
differently.”
Ask who controls definitions of vague
nouns.
Here: the deal,
cease-fire—undefined in the snippet.
Ask who the implied audience is (allies,
adversaries, domestic public).
Here: reassurance that Lebanon is out of
scope.
List what is missing before you treat the line
as settled fact.
Here: no document, date, or quote from Iran.
Hold two forks until primary evidence
appears.
Here: honest mistake vs framing/blame vs decoupling.
Record provenance before you argue about
“bias.”
Here: not visible—demo clip has no outlet or
date.
Primary material that would narrow the forks: published terms of the cease-fire or related statements; verbatim remarks (transcript, official readout); dated prior statements. Corroboration: same line reported elsewhere, official White House/Iranian releases, full article context (what question prompted Vance). The lab does not supply those; it trains you to ask for them—and to separate “what the sentence does” from “whether outlet Z is trustworthy” (you need provenance for the second).
Source text (composite, for classroom use)
The minister said the reform applied only to large cities; rural municipalities were never included, and local leaders misunderstood the announcement.
Task: Supply mock provenance (e.g. “Agency X, 2026-04-01, wire”) or state stripped clip. On paper or with the lite prompt (§3), produce sections 1–4 of the output (parse—including provenance and sourcing type—key words, at least two forks + hinge, 8 habits with Here lines). Compare with a partner or an assistant; argue about whether a third fork is fair.
Provenance — Outlet: Date/time: Headline: Byline or wire: URL (optional):
Passage (paste or copy):
1. Parse — Provenance recorded? Sourcing type (on record / anonymous / doc)? Narrator? Asserted vs attributed? Atomic claims:
2. Key words (word → why it matters):
| Word / phrase | Why it matters |
|---|---|
3. Forks — Reading A:
Reading B:
Hinge:
4. Habits — Numbered habit + Here:
5. Evidence — What would I look up?
| Criterion | Not yet | OK | Strong |
|---|---|---|---|
| Attribution separated from reporter’s own claims | |||
| At least three atomic claims (or justified fewer) | |||
| Scope / fault language flagged | |||
| Two defensible forks + clear hinge | |||
| Provenance named or “missing—cannot judge chain” stated honestly | |||
| Habits are reusable (not only a summary of this text) | |||
| Each habit has a concrete Here cue |
These moves borrow interval, negative space, and encounter from a “relational publishing” mindset: rigor without pretending the reader is only a user retrieving an answer.
Interval before closure — The prelude and the evidence note model delay: confusion named, resolution deferred where the text does not justify it. In class, you can add one spoken line before the demo: what felt unstable about the passage the first time you read it?
Negative space in the argument — The lab requires at least two forks and a hinge; optionally forbid a “winner” fork in debrief. Pair a strong reading with one honest counter-reading students must record even if they dislike it.
Process as content — Name a pedagogical dead end you have hit (e.g. face-value reading, or collapsing “bias” and “provenance”). Invite students to note one habit they previously skipped; optional short changelog line on their template (“what I used to do / what I’ll try”).
Encounter, not only explanation — Start once in a while with only the passage on screen—no learning-outcomes slide for two minutes—or with a single question (“Who is cast as mistaken, and who gets to define the noun?”) before any framework vocabulary.
Rhythm — Keep §3 lite and timed parses fast where the goal is pattern recognition. Protect slow time for forks, hinge, and debrief; treat the rubric’s fork row as non-negotiable depth, not skim.
Ethics of the quick answer — If a model produces ranked “takeaways,” have students move those to an appendix or strip them; the human core of the submission should be forks, hinge, and two annotations of agreement/disagreement with the tool.
| You want… | Use… |
|---|---|
| Fastest path: parse, forks, habits, evidence | §3 (lite) or Appendix A (same text) |
| Condensed analyst write-up + same learner checklist | Appendix B (standard) |
| One paragraph instruction to an assistant | Appendix C (compact) |
| Maximum depth: voice map through counter-voice, shared ground, long follow-ups | Appendix D (full analyst, sections 0–18) |
| Dense single pass without section labels in the prompt | Appendix E (compact analyst checklist) |
Learner checklist rule (shared by all deep tiers): Section 14 in Appendix D and section 12 in Appendix B require 8–12 numbered habits, each with a “Here: …” line tied to the passage, without repeating the whole prior analysis.
The full lite prompt text is in §3 above (single source of truth). If you print or split Part I from Part II, copy the fenced block from §3.
You are a senior analyst of dialogue and political/institutional discourse. Analyse the passage from the text alone unless the user adds context; do not invent facts, quotes, or events.
Use numbered markdown sections exactly as below.
0. **Genre, context, and provenance** — Kind of text; inferred setting. **Provenance:** If the user supplied outlet, date, headline, byline, wire/agency, language, or URL, summarize and note implications for timeliness and relay; if **not** supplied, list what should be recorded for real clips and state that the **publication chain** cannot be judged from the sentence alone.
1. **Voice map** — Narrator/reporter vs named speakers; reporting frame vs direct assertion; inferred audience if any. Note **journalistic sourcing** visible in the passage (named on-record, paraphrase, “sources said”, document—only what the wording supports).
2. **Paraphrase** — Gist in plain language; separate surrounding assertion from attributed claims.
3. **Structure** — How the passage moves; who controls definitions of key terms (e.g. “the deal,” “cease-fire”).
4. **Key terms and responsibility** — 3–8 items: implicature; scope restrictors; verbs of saying/believing/denying; fault or obligation.
5. **Named framing moves** — Up to five from: definitional narrowing; agentive shift; erasure; boundary work; strategic decoupling; semantic trap; modal hedging; false binary; etc. Tie each to **specific wording**.
6. **Narrative hierarchy** — Who is cast as clear vs confused, authoritative vs mistaken?
7. **Stakeholders** — At least three interested parties; for two divergent ones, brief “internal logic” (why this line might feel rational from their seat—hypotheses grounded in the text).
8. **Value collision & epistemic gap** — Conflicting principles; what each side’s framing foregrounds vs backgrounds (no invented events).
9. **Omissions** — What the text does not specify that a careful reader would want.
10. **Interpretive fork** — Reading A / Reading B (+ C if needed); **hinge** (definitions, values, missing facts).
11. **Synonym-swap / who benefits** — One or two loaded phrases; neutral paraphrase test; who gains if the audience accepts the speaker’s framing.
12. **How to read similar passages (learner checklist)** — **8–12 numbered habits** (one sentence each; cover **provenance** (outlet/date/wire); **journalistic sourcing** type if visible; attribution chain; atomic claims; scope restrictors; cognition/fault verbs vs neutral rewrites; who defines loaded nouns; audience; omissions). After each habit, **one** “*Here: …*” line citing **specific words** in *this* passage (*not visible in clip* if stripped). **No recap** of prior sections—**transfer** only.
13. **Counter-voice (optional if user asks)** — Core claim rewritten in an opposing stakeholder’s vocabulary; label as reconstructive.
14. **Follow-ups** — 5–8 analyst questions; **evidence note** — what primary sources and **corroboration** would adjudicate forks.
Source text:
[PASTE PASSAGE HERE]Analyse this reported-speech or news snippet: (1) provenance if supplied else what is missing; journalistic sourcing type if visible; parse narrator vs attributed claims; (2) 4–6 key words (scope, fault, implication); (3) Reading A / Reading B + hinge; (4) **8–10 numbered reading habits**, each with *Here: …* tied to this text; (5) primary evidence and corroboration you would seek. Text only—no invented facts.
[PASTE HERE]You are a senior analyst of dialogue and political/institutional discourse. Analyse the passage or exchange the user provides. Work from the text alone unless the user supplies external context; do not invent facts, quotes, or events.
Use numbered markdown sections exactly as below. If saving to a file, start with a one-line title and put the raw passage under “Source text.”
0. **Genre, context, and provenance** — Kind of text (e.g. news report, transcript, op-ed, leak, communique); inferred setting. **Provenance:** If the user supplied outlet, date, headline, byline, wire/agency, language, or URL, summarize and note implications for timeliness and relay; if **not** supplied, list what should be recorded and state that the **publication chain** cannot be judged from the sentence alone.
1. **Voice map** — List each **distinct voice**: narrator/reporter, named speakers, institutions. For each: role in the exchange, **reporting frame** (“X said that…”) vs **direct assertion**, and **audience** if inferable (public, domestic, diplomatic, etc.). If only one surface voice, still note implied addressee(s). Note **journalistic sourcing** the passage displays (named on-record, paraphrase, anonymous sourcing, document—only what the wording supports).
2. **Paraphrase** — Gist in plain language (one short paragraph). Separate what the **surrounding text** asserts from what is **attributed** to each speaker.
3. **Structure of the exchange** — How the dialogue or passage **moves** (e.g. claim → qualification → rebuttal; question → non-answer). For multi-turn text, note **turn functions** (claim, challenge, repair, closure, topic shift). Flag **who controls the definition** of key terms (e.g. “the deal,” “cease-fire”).
4. **Key terms and carriers of responsibility** — 3–8 items: connotation and **implicature**, not only dictionary sense. Include **scope restrictors** (only, just, merely, purely), **verbs of saying/believing/denying**, and any terms that **assign fault** or **limit obligation**.
5. **Named linguistic / framing moves** — In one coherent paragraph, name **up to five** mechanisms that actually apply (draw from this list when relevant; skip what does not fit): definitional narrowing; agentive shift / blame placement; erasure; paternalism; boundary work; strategic decoupling; semantic trap; normalization of exception; hermetic logic; modal hedging; false binary; strategic omission. Tie each named move to **specific wording**.
6. **Narrative hierarchy** — Who is cast as **clear vs confused**, truthful vs mistaken, authoritative vs needing correction? What **moral or cognitive** grading does the wording imply?
7. **Stakeholder compression** — Identify **at least three** parties with a stake (they need not all speak). For each: **primary objective** in this situation as inferable from the text. Then, for **two** opposing or divergent stakes, answer briefly: “If I were in their position, why would this line or frame feel rational?” (assumptions and risks they may be managing—hypotheses grounded in the passage, not psychobiography).
8. **Value collision and epistemic gap** — Name **conflicting values or principles** both sides could appeal to (e.g. textual legalism vs holistic security). Then: what does side A’s framing **treat as central** that side B’s likely framing would **background or deny**, and vice versa—still without inventing off-text events.
9. **Tone and stance** — **Discourse stance** of the narrator/channel vs **stance of attributed content** per speaker where relevant.
10. **Hidden justification** — In three to five sentences: what **actions, policies, or moral conclusions** does this wording make easier to defend or harder to attack? If non-political, say “minimal” and justify briefly.
11. **What is left out** — Context, actors, evidence types, or counter-narratives the text avoids (e.g. dates, text of agreements, other parties, alternative explanations for “misunderstanding”). Be specific like a strong close reading, not a generic list.
12. **Synonym-swap and benefit test** — Pick **one or two** loaded phrases. Would neutral paraphrase (e.g. “had a different interpretation” for “misunderstood”) change perceived fairness? **Who benefits** if the audience accepts the speaker’s definitions?
13. **Interpretive fork** — **Reading A** and **Reading B**, each defensible by a fair reader. End with **what hinges the difference** (definitions, values, missing facts, audience).
14. **How to read similar passages (learner checklist)** — **Teach** with **8–12 numbered habits** the reader can reuse on *other* news lines and reported speech (imperative or “Ask yourself…”; **one sentence per habit**). Habits should cover at least: **provenance** (record outlet, date, wire, language); **journalistic sourcing** (named on-record vs anonymous vs document—only if the passage shows it); **attribution chain** (narrator vs speaker vs who is named); **atomic claims** (split bundled sentences); **scope / quantity / geography** restrictors (only, merely, not part of, etc.); **cognition and fault verbs** (misunderstood, denied, claimed…) vs neutral rewrites; **who controls definitions** of loaded nouns (deal, cease-fire, attack); **audience** (who the line is meant to reassure or pressure); **omissions** (what evidence is missing). After the numbered list, add **one italicized application line per habit**: “*Here: …*” pointing to **specific words** in *this* passage (*not visible in clip* if stripped)—so the checklist is **transferable** but **grounded**. Do **not** paste long quotes from §§1–13; this section is for **reuse**, not recap.
15. **Counter-voice rehearsal** — Short **rewrite** of the passage’s core claim from an **opposing** stakeholder’s vocabulary and tone (clearly labeled as reconstructive, not a quote). Optional one sentence: **counter-passage** emphasis—which words would *they* stress?
16. **Shared ground (if any)** — One **outcome all major stakeholders might plausibly wish to avoid**, inferable without fabrication. If not inferable, write “not inferable from text.”
17. **Penetrative follow-ups** — List **5–8** concise questions an analyst could ask next (mix of: ambiguity/trap, audiences, linkage across actors or theaters, evidence needed to decide forks, stability of “decoupling,” internal domestic audience). Do not answer them unless the user asks.
18. **Evidence note** — What **primary** material would adjudicate the fork (documents, timestamps, prior statements, verbatim transcript)? What **corroboration** (other outlets, official releases) would you seek? Reminder: your output is **interpretive**, not verified reporting.
Source text:
[PASTE PASSAGE OR TRANSCRIPT HERE]Analyse this dialogue or reported-speech passage. Return: (1) genre + **provenance** (if supplied) or what is missing; **journalistic sourcing** if visible; voice map (narrator vs speakers); (2) paraphrase separating attribution from assertion; (3) structure / turn moves; (4) 4–6 key terms with implicature + any scope/fault words; (5) 2–4 **named** framing moves tied to wording; (6) narrative hierarchy (who sounds clear vs confused); (7) three stakeholders + brief opposing “internal logic”; (8) value collision + epistemic gap; (9) omissions; (10) Reading A / Reading B + hinge; (11) synonym-swap or “who benefits” on one loaded phrase; (12) **8–10 numbered transferable reading habits**, each with one “*Here: …*” tie to this text; (13) 4 follow-up questions; (14) primary evidence + corroboration that would decide. Text only—no invented facts.
[PASTE HERE]| § | Focus |
|---|---|
| 0 | Genre / setting / provenance (outlet, date, wire…) |
| 1 | Who speaks; reporting vs assertion; journalistic sourcing if visible |
| 2 | Paraphrase with attribution split |
| 3 | Move-by-move structure; who defines terms |
| 4 | Key words + responsibility |
| 5 | Named framing mechanisms |
| 6 | Who sounds clear vs confused |
| 7 | Stakeholders + “internal logic” |
| 8 | Values in tension; what each side foregrounds |
| 9 | Tone / stance |
| 10 | What the wording helps justify |
| 11 | Omissions |
| 12 | Synonym-swap; who benefits |
| 13 | Interpretive fork |
| 14 | Learner checklist + Here lines |
| 15 | Counter-voice rehearsal |
| 16 | Shared ground (if inferable) |
| 17 | Follow-up questions |
| 18 | Evidence that would adjudicate |
End of standalone lab. No external file required to run the unit or to use any prompt tier.