30 Claude Prompts for a Discussion You Can Defend

<context> I need a limitations subsection that names weaknesses I already accept. Do not invent a new limit to look careful, and do not hide one I pasted. </context> <inputs> - Limits I already accept: [PASTE LIMITS] - Findings I can defend: [PASTE RESULTS] - Design: [STUDY DESIGN] - Sample n if known: [SAMPLE N OR UNKNOWN] - Research question: [RESEARCH QUESTION] </inputs> <task> Write a limitations block from [PASTE LIMITS] plus any limit that is already visible in [PASTE RESULTS] or [STUDY DESIGN]. For each: the limit, why it matters for [RESEARCH QUESTION], and a weaker claim I can still keep. Do not add a limit that is not in the paste just to look thorough. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent n, a missed wave, or a second coder I did not name. - If [SAMPLE N OR UNKNOWN] is UNKNOWN, refuse a small-n sermon that invents the n. - If a limit needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Numbered limits with weaker claims, then a not-in-paste refusal list, then UNVERIFIED rows. </format>

<context> I need an honest cannot-speak-to list in the discussion. Work from the sample and design I pasted. Do not invent a population I missed or an n I do not have. </context> <inputs> - Findings I can defend: [PASTE RESULTS] - Sample notes: [PASTE SAMPLE NOTES OR NONE] - Sample n if known: [SAMPLE N OR UNKNOWN] - Design: [STUDY DESIGN] - Who I could not reach: [MISSING POPULATION OR UNKNOWN] - Research question: [RESEARCH QUESTION] </inputs> <task> List what this set cannot speak to for [RESEARCH QUESTION]: other sites, other times, other groups, causal claims, prevalence. For each: why the paste blocks it, and a discussion sentence I can keep. If [SAMPLE N OR UNKNOWN] is UNKNOWN, keep every sentence n-free. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent n, a response rate, or a demographic split. - Do not invent that the sample is representative, or that it is fatally small, if n is UNKNOWN. - If [MISSING POPULATION OR UNKNOWN] is UNKNOWN, write UNKNOWN rather than guessing who is missing. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Cannot-speak-to table, n-free sentences, UNVERIFIED rows. </format>

<context> Design limits and result limits get mixed in my discussion. Split them. Do not invent a methods fact or a finding to fill a side. </context> <inputs> - Methods or protocol notes: [PASTE METHODS] - Findings I can defend: [PASTE RESULTS] - Limits I already accept: [PASTE LIMITS] - Research question: [RESEARCH QUESTION] </inputs> <task> Write two short subsections: (1) design limits from [PASTE METHODS] only (sample, timing, instruments, analysis you named), (2) result limits from [PASTE RESULTS] only (nulls, thin cells, mixed signs). Move any sentence that sits on the wrong side. Do not invent a limit to balance the lists. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent a procedure, a software version, or a missed wave. - Do not treat a null as a design error unless the paste says the measure failed. - If a sentence needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Design-limits subsection, result-limits subsection, moved-sentence list, UNVERIFIED rows. </format>

<context> I am tempted to close a limitation with a study I did not run. Refuse that close. Do not invent a follow-up design, a dataset, or a citation that would rescue the limit. </context> <inputs> - Limits I already accept: [PASTE LIMITS] - Findings I can defend: [PASTE RESULTS] - Follow-up I am tempted to claim: [TEMPTING FOLLOW-UP OR NONE] - What I actually can run later: [DATA ACCESS OR UNKNOWN] - Research question: [RESEARCH QUESTION] </inputs> <task> For each limit in [PASTE LIMITS]: write a discussion sentence that leaves the limit open, then refuse any close that is not already in the paste. If [TEMPTING FOLLOW-UP OR NONE] would close a limit I did not run, tag it REFUSED, NOT RUN. Do not invent a design that would save it. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent a registry, a second sample, or a longitudinal wave. - If [DATA ACCESS OR UNKNOWN] is UNKNOWN, do not promise a next study I can run. - Future work stays in a later prompt. Here, keep the limit open. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Open-limit sentences, REFUSED closes, UNVERIFIED rows. </format>

<context> My discussion verbs may overclaim the results I pasted. I want hedges that match the evidence, not decorative may on every line. </context> <inputs> - Discussion draft or notes: [PASTE DRAFT] - Findings I can defend: [PASTE RESULTS] - Hedges I already use: [HEDGES I ALREADY USE] - Design: [STUDY DESIGN] </inputs> <task> Adjust claim strength in the discussion: prove / show / suggest / may / cannot say. Tie each choice to a line in [PASTE RESULTS]. Do not add evidence. Do not hedge a count I recorded. If [STUDY DESIGN] cannot carry a causal verb, refuse it. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent a statistic that would let me drop a hedge. - Over-hedging (it could perhaps be argued) is also a miss. Prefer one accurate verb. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Revised sentences with verbs marked, strength table (claim, verb, results line), UNVERIFIED rows. </format>

<context> I need theory implications for a paper discussion. Argue only from results and sources I pasted. Do not invent a theory, a landmark paper, or a finding that would make the theory fit. </context> <inputs> - Findings I can defend: [PASTE RESULTS] - Research question: [RESEARCH QUESTION] - Theory or framework I already named: [THEORY OR NONE] - Sources I may cite: [PASTE SOURCE LIST] - Citation style: [CITATION STYLE] </inputs> <task> Draft theory implications from [PASTE RESULTS] only. If [THEORY OR NONE] is NONE, write implications as claims about the constructs in the paste, not as a named school. Cite only [PASTE SOURCE LIST]. If a sentence needs a paper I do not have, mark UNVERIFIED. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent a theory name, a founder, or a year. - Do not claim the results confirm a theory I did not test. - If [PASTE SOURCE LIST] is empty, use zero citations. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Theory-implication paragraphs, claim-to-result map, UNVERIFIED rows. </format>

<context> I need practice implications a reader can act on only if my results support them. Do not invent a clinic win, a workflow, or a training claim I did not study. </context> <inputs> - Findings I can defend: [PASTE RESULTS] - Who is affected if I know: [AFFECTED GROUP OR UNKNOWN] - Setting I actually studied: [SETTING] - What I must not imply: [OFF LIMITS] - Research question: [RESEARCH QUESTION] </inputs> <task> Write practice implications that stay inside [PASTE RESULTS] and [SETTING]. If [AFFECTED GROUP OR UNKNOWN] is UNKNOWN, do not invent a population. Honor [OFF LIMITS]. For each implication: the result line that backs it, and a cannot-generalize note. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent a hospital, a classroom, a tool, or a dosage. - Do not turn a lab or interview pattern into a care or HR recommendation I did not paste. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Practice implications with backing lines, cannot-generalize notes, UNVERIFIED rows. </format>

<context> I need policy implications that do not open with an invented statistic, jobs-saved number, or law I did not paste. </context> <inputs> - Findings I can defend: [PASTE RESULTS] - Policy notes I can defend: [PASTE POLICY NOTES OR NONE] - Research question: [RESEARCH QUESTION] - Audience: [JOURNAL / FUNDER / DEPARTMENT] - Citation style: [CITATION STYLE OR NONE] </inputs> <task> Draft policy implications only if [PASTE RESULTS] or [PASTE POLICY NOTES OR NONE] support a policy-facing sentence. If the paste cannot carry policy, say this set cannot speak to policy and stop. Flag any impact number that is not in the paste as UNVERIFIED. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent prevalence, cost, votes, or a statute. - Do not promise an impact my design cannot produce. - If [CITATION STYLE OR NONE] is NONE, use zero citations. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Policy paragraph or a cannot-speak-to-policy note, claim-to-source checklist, UNVERIFIED rows. </format>

<context> I need the discussion to name who is affected only when my paste supports it. Do not invent a community, a patient group, or a workforce I did not study. </context> <inputs> - Findings I can defend: [PASTE RESULTS] - Who is affected, from my notes: [AFFECTED GROUP OR UNKNOWN] - Setting: [SETTING] - Missing population if I know it: [MISSING POPULATION OR UNKNOWN] - Research question: [RESEARCH QUESTION] </inputs> <task> Write who-is-affected sentences using only [AFFECTED GROUP OR UNKNOWN] and [PASTE RESULTS]. If UNKNOWN, keep every sentence group-free and list the cells I must open. If I named a missing population, say who the results cannot speak for. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent a city, clinic, school, or industry I did not name. - Do not upgrade a convenience sample into the public or patients. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Who-is-affected prose or UNKNOWN stub, cannot-speak-for list, UNVERIFIED rows. </format>

<context> I need a does-not-change list so implications stay honest. Do not invent a field consensus or a policy that my results leave untouched. </context> <inputs> - Findings I can defend: [PASTE RESULTS] - Research question: [RESEARCH QUESTION] - Tempting implications: [TEMPTING CLAIMS] - Limits I already accept: [PASTE LIMITS] - Design: [STUDY DESIGN] </inputs> <task> List what [PASTE RESULTS] do not change: theory I did not test, practice I did not observe, policy I did not measure, and each item in [TEMPTING CLAIMS] that the paste cannot carry. For each: why it stays unchanged, and a weaker implication I can keep. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent a consensus I would be overturning. - Do not invent a next study that would make a tempting claim true. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Does-not-change table (claim, why unchanged, weaker line), UNVERIFIED rows. </format>

<context> I need a future-work paragraph that follows results I already have. Do not invent a design, a dataset, or a citation that would make the next paper look funded. </context> <inputs> - Findings I can defend: [PASTE RESULTS] - Limits I already accept: [PASTE LIMITS] - Research question: [RESEARCH QUESTION] - What I can actually reach later: [DATA ACCESS OR UNKNOWN] - Months I might have: [MONTHS OR UNKNOWN] </inputs> <task> Propose 3 to 5 next studies that follow a hole in [PASTE RESULTS] or [PASTE LIMITS]. For each: the question, why this paste opens it, and FEASIBLE or NOT IN PASTE given [DATA ACCESS OR UNKNOWN]. Do not invent a sample, a grant, or a partner. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent a registry, an archive, or a lab I did not name. - If [DATA ACCESS OR UNKNOWN] or [MONTHS OR UNKNOWN] is UNKNOWN, mark feasibility UNVERIFIED. - Do not write we will as if the protocol exists. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Numbered next studies with FEASIBLE or NOT IN PASTE tags, UNVERIFIED rows. </format>

<context> I need future-work sentences about methods I would change, drawn from gaps I already see. Do not invent a procedure I never tried, and do not rewrite the methods section. </context> <inputs> - Methods or protocol notes: [PASTE METHODS] - Findings I can defend: [PASTE RESULTS] - What went wrong or stayed thin: [GAPS I SEE] - Research question: [RESEARCH QUESTION] </inputs> <task> List method changes that follow [GAPS I SEE] and clashes you can see between [PASTE METHODS] and [PASTE RESULTS]. For each: the current step, the change, and whether the change is a discussion future-work line or a methods correction I already owe. Do not add a tool I did not name. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent a second coder, a validated scale, or a software version. - Do not turn this into a methods rewrite. One or two sentences per change. - If a change needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Table: gap | current step | change | discussion line or methods debt | UNVERIFIED. </format>

<context> I need future questions that my results actually opened, not a wish list from another paper. Do not invent a finding that would make a trendier question look earned. </context> <inputs> - Findings I can defend: [PASTE RESULTS] - Research question: [RESEARCH QUESTION] - What surprised me, in my words: [SURPRISES OR NONE] - Limits I already accept: [PASTE LIMITS] </inputs> <task> Write 5 questions that [PASTE RESULTS] opened. Each question must point to a phrase in the paste (a null, a mixed cell, a limit, a leftover comparison). Drop any question that needs a literature I did not paste. Mark leftover ideas UNVERIFIED. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent a debate, a famous paper, or a dataset. - Prefer a narrower true question over a punchy field-level one. - If [SURPRISES OR NONE] is NONE, do not invent a surprise. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Numbered questions with the paste phrase that opened each, dropped ideas, UNVERIFIED rows. </format>

<context> I need future work that fits time and access I actually have. Do not invent a grant, a partner, or a dataset that would make a larger follow-up easy. </context> <inputs> - Findings I can defend: [PASTE RESULTS] - Tempting follow-up: [TEMPTING FOLLOW-UP] - Months I might have: [MONTHS] - Data I can actually reach: [DATA ACCESS] - Research question: [RESEARCH QUESTION] </inputs> <task> Audit [TEMPTING FOLLOW-UP] against [MONTHS] and [DATA ACCESS]. Tag FEASIBLE, STRETCH, or REFUSED. If STRETCH or REFUSED, rewrite 3 smaller next steps that follow [PASTE RESULTS] and that I can name without a new archive or lab. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent access to a dataset, clinic, or panel I did not name. - Be blunt if [MONTHS] cannot support the tempting follow-up. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Verdict, reason, three smaller next steps, UNVERIFIED rows. </format>

<context> I have a future-work list I am tempted to stuff. Rank it by what my results support, not by how impressive the next paper would sound. </context> <inputs> - Future-work list: [FUTURE LIST] - Findings I can defend: [PASTE RESULTS] - Limits I already accept: [PASTE LIMITS] - Data I can reach: [DATA ACCESS OR UNKNOWN] - Research question: [RESEARCH QUESTION] </inputs> <task> For each item in [FUTURE LIST], tag KEEP (the paste opens it), STRETCH, or DROP. Rank KEEP items by how directly [PASTE RESULTS] open them. Propose replacements only from holes already in the paste. Do not fill the list with extras I did not study. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent a fashionable method or dataset as a replacement. - Empty slots are better than trendy next papers. - If [DATA ACCESS OR UNKNOWN] is UNKNOWN, mark feasibility UNVERIFIED. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Table: item, tag, paste hole, rank. Then a clean KEEP list. UNVERIFIED rows last. </format>

<context> I need a discussion-section outline a paper can follow, not a finished draft. Do not stock the outline with findings, citations, or implications I have not pasted. This is not an abstract outline and not a thesis map. </context> <inputs> - Research question: [RESEARCH QUESTION] - Findings I can defend: [PASTE RESULTS] - Limits I already accept: [PASTE LIMITS] - Venue headings if I have them: [DISCUSSION HEADINGS OR NONE] - Word count target: [WORD COUNT] </inputs> <task> Build an outline: restated findings, meaning, comparison only if I later paste sources, limits, implications, future work, close. Each heading needs a job, a word budget that sums to [WORD COUNT], and a blank where a finding, cite, or group is missing. If I pasted [DISCUSSION HEADINGS OR NONE], use those headings. </task> <constraints> - Do not invent findings, citations, or data. - If a heading has no pasted support, leave it as a heading plus NEED NOTES. - Do not invent a journal heading I did not paste. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Nested outline with word counts and NEED NOTES tags. UNVERIFIED list at the end. </format>

<context> I will paste the results I actually have. Draft a complete discussion section from that paste only. Do not invent findings, citations, data, or a literature I have not supplied. This is not an abstract and not a methods rewrite. </context> <inputs> - Research question: [RESEARCH QUESTION] - Findings I can defend: [PASTE RESULTS] - Limits I already accept: [PASTE LIMITS] - Methods notes if I have them: [PASTE METHODS OR NONE] - Sources I may cite: [PASTE SOURCE LIST OR NONE] - Word count: [WORD COUNT] - Venue: [VENUE] - Citation style: [CITATION STYLE] </inputs> <task> Draft the full discussion under [WORD COUNT]: restated findings, meaning, comparison only to [PASTE SOURCE LIST OR NONE], limits, implications, future work, close. Cite only sources in the paste. Tag any sentence that still needs a finding, count, or source as UNVERIFIED. Leave counts as UNKNOWN unless I pasted them. </task> <constraints> - Do not invent findings, citations, or data. - Do not write a cleaner result than [PASTE RESULTS]. - If a heading has no pasted support, write NEED NOTES under it instead of filling it. - Prefer a thinner honest draft over a complete-looking section I cannot evidence. - If [PASTE SOURCE LIST OR NONE] is NONE, use zero citations. </constraints> <format> Full discussion draft, then a master UNVERIFIED and NEED NOTES list. </format>

<context> I need a discussion that follows rules I actually pasted, not a generic IMRaD costume. Do not invent a heading, word limit, or citation rule I did not supply. </context> <inputs> - Findings I can defend: [PASTE RESULTS] - Venue: [VENUE] - Venue rules I copied: [VENUE RULES] - Required headings or none: [DISCUSSION HEADINGS OR NONE] - Word count: [WORD COUNT] - Limits I already accept: [PASTE LIMITS] </inputs> <task> Draft the discussion to [VENUE RULES] and [WORD COUNT]. If I pasted headings, use only those headings. If I said NONE, write a narrative discussion in the usual order (meaning, limits, implications, future work). Flag any rule in my paste you cannot meet from [PASTE RESULTS] as NOT IN PASTE. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent a journal rule, heading, or example paper from that journal. - Do not invent data or results to satisfy a heading. - If [VENUE RULES] is thin, keep items as blank checks I must verify. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Discussion in the required shape, word count, rules you could not meet, UNVERIFIED rows. </format>

<context> I must cut a discussion draft to [WORD COUNT] without losing the argument or inventing a tighter finding to save words. </context> <inputs> - Discussion draft: [PASTE DRAFT] - Target word count: [WORD COUNT] - Must keep: [MUST KEEP] - Findings I can defend: [PASTE RESULTS] - Citation rule: [NO CITES / ONLY PASTED CITES] </inputs> <task> Cut to [WORD COUNT]. Protect [MUST KEEP]. Prefer deleting repetition and throat-clearing. Do not delete a hedge and keep the claim as if it were proven. Do not invent a tighter statistic to save space. Show a cut log. </task> <constraints> - Do not invent findings, citations, or data. - Do not add new citations while cutting. - Do not merge two sources into one name. - If the claim cannot survive the cut, say so and stop before you gut the evidence. - If a claim needs a source I did not give, mark it UNVERIFIED. - Report old and new word counts. </constraints> <format> Cut draft, word counts, cut log (what left and why), UNVERIFIED rows. </format>

<context> I need to know whether the discussion promises a cleaner paper than I wrote. Compare the texts I paste. Do not invent the missing paragraph. This is not an abstract check yet. </context> <inputs> - Discussion draft: [PASTE DRAFT] - Findings I can defend: [PASTE RESULTS] - Methods or protocol notes: [PASTE METHODS] - Research question: [RESEARCH QUESTION] - Word count: [WORD COUNT] </inputs> <task> Compare [PASTE DRAFT] to [PASTE RESULTS] and [PASTE METHODS]. Flag: a result that is cleaner than the table, a method the methods never state, a limit the discussion dropped, and a hedge the discussion lost. Then rewrite the discussion within [WORD COUNT] so it matches the weaker, true paper. Do not fill holes with new findings. </task> <constraints> - Do not invent findings, citations, or data. - If results or methods are missing a cell, write NOT IN PASTE and shrink the discussion. - Do not make the rewrite more positive than the results. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Mismatch list, rewritten discussion, word count, NOT IN PASTE and UNVERIFIED rows. </format>

<context> I need a revision pass that cuts overclaim verbs in a discussion I already drafted. Do not add evidence that would let a loud verb stay. </context> <inputs> - Discussion draft: [PASTE DRAFT] - Findings I can defend: [PASTE RESULTS] - Hedges I already use: [HEDGES I ALREADY USE] - Design: [STUDY DESIGN] </inputs> <task> Search [PASTE DRAFT] for prove, show, demonstrate, confirm, effect, and representative. For each hit: quote the line, tag MATCH or OVERCLAIM against [PASTE RESULTS], and rewrite with a verb from [HEDGES I ALREADY USE] or cannot say. Do not invent a number that would save the loud verb. </task> <constraints> - Do not invent findings, citations, or data. - Do not add citations while revising verbs. - If [STUDY DESIGN] cannot carry causal language, refuse effect and impact as verbs. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Table: quoted line | tag | rewritten line | UNVERIFIED. Then the revised discussion. </format>

<context> I have a discussion draft that may contain names I did not paste. Strip those. Do not complete a missing author-year from memory. </context> <inputs> - Discussion draft: [PASTE DRAFT] - Sources I may cite: [PASTE SOURCE LIST] - Findings I can defend: [PASTE RESULTS] - Citation style: [CITATION STYLE] </inputs> <task> For each in-text citation in [PASTE DRAFT], tag KEEP (it is on [PASTE SOURCE LIST] and supports the sentence), STRETCH, or DROP. Remove DROP citations and either weaken the sentence or mark UNVERIFIED. Do not replace a dropped cite with a paper from memory. </task> <constraints> - Do not invent findings, citations, or data. - Do not invent years, pages, or DOIs for a KEEP row. - Do not attach a pasted source to a sentence it does not support. - If [PASTE SOURCE LIST] is empty, strip every citation and mark those sentences UNVERIFIED. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Cite table (cite, tag, action), cleaned discussion, UNVERIFIED rows. </format>

<context> I have reviewer, editor, or supervisor comments on a discussion. I will revise the draft I paste. I will not invent sources or findings to look responsive. </context> <inputs> - Discussion draft: [PASTE DRAFT] - Comments: [PASTE COMMENTS] - Findings I can defend: [PASTE RESULTS] - Sources I may add if they are already mine: [PASTE SOURCE LIST OR NONE] - Word count: [WORD COUNT] </inputs> <task> Map each comment in [PASTE COMMENTS] to: agree and change, ask for clarification, or push back with evidence already in [PASTE RESULTS]. Then draft the revised discussion within [WORD COUNT]. If a comment demands a source or finding I do not have, write FIND THIS SOURCE or NOT IN PASTE. Do not invent it. </task> <constraints> - Do not invent findings, citations, or data. - Do not add papers the comments reminded you of unless they are on [PASTE SOURCE LIST OR NONE]. - Do not treat every comment as law if it contradicts the pasted results. Say so. - Do not claim a change you did not draft. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Comment table (comment, decision, draft change), revised discussion, FIND THIS SOURCE list, UNVERIFIED rows. </format>

<context> I need an honesty pass: does this discussion say only what my results support? </context> <inputs> - Discussion draft: [PASTE DRAFT] - Findings I can defend: [PASTE RESULTS] - Limits I already accept: [PASTE LIMITS] - Work status: [COMPLETED / IN PROGRESS] </inputs> <task> For each claim in the discussion, tag MATCH (results support it), STRETCH (stronger than the results), or NOT IN PASTE. Quote a short supporting phrase for MATCH. Do not fix STRETCH by inventing evidence. Propose a weaker wording I can use until I find a source. </task> <constraints> - Do not invent findings, citations, or data. - Do not treat your own memory of a famous paper as evidence. - If the discussion cites a work that is not in the paste, flag CITE NOT IN PASTE. - If status is IN PROGRESS, tag past-tense findings as STRETCH unless the paste shows they are done. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Claim table: tag, quoted support or none, weaker wording. Then CITE NOT IN PASTE and UNVERIFIED lists. </format>

<context> I need to know whether the discussion and the abstract tell the same study. Compare the two texts I paste. Do not invent the missing sentence. This page still writes discussion, not a new abstract. </context> <inputs> - Discussion draft: [PASTE DRAFT] - Abstract I already have: [PASTE ABSTRACT] - Findings I can defend: [PASTE RESULTS] - Word count for the discussion only: [WORD COUNT] </inputs> <task> Compare [PASTE DRAFT] to [PASTE ABSTRACT]. Flag: a finding that appears in only one text, a limit the abstract dropped, a hedge one text lost, and a claim neither text can back with [PASTE RESULTS]. Rewrite only the discussion within [WORD COUNT] so it matches the weaker, true results. Do not rewrite the abstract here. </task> <constraints> - Do not invent findings, citations, or data. - Do not add a finding to the discussion because the abstract sounds bigger. Shrink the discussion, or mark the abstract line UNVERIFIED for me to fix on the abstracts page. - Do not produce a new abstract. - If a claim needs a source I did not give, mark it UNVERIFIED. </constraints> <format> Mismatch list, rewritten discussion only, word count, UNVERIFIED rows (including abstract-only claims I must fix elsewhere). </format>