Research Communication Critic

Thorough, evidence-driven review of research-to-public communications including university press releases, policy briefs, health communications, news articles about research, social media research summaries, and lay summaries. This skill evaluates whether simplified research findings remain accurate, whether complexity is reduced without distorting meaning, whether hedging matches actual study certainty, whether citations enable lay readers to find sources, whether jargon is genuinely translated, and whether visual explanations serve non-specialist audiences.

Use this skill to critique research communication quality — the "translation gap" between academic findings and public understanding.

The translation gap is universal: every university, think tank, public health organization, research-adjacent nonprofit, and science communication team struggles with communicating research findings to non-specialist audiences. This critic surfaces that gap.

JTBD (Jobs To Be Done)

Primary Job

When I have a press release, policy brief, health communication, or lay summary that translates research findings for a non-specialist audience — and I need to know whether the simplification distorted the science or whether "may suggest a link" became "proves that" somewhere in the editing process — I want a fidelity audit that catches hedging failures, oversimplified statistics, and accuracy losses before the communication reaches an audience that will act on it, so I'm not responsible for a retraction or a public correction because the translation gap introduced a claim the original study never made.

Secondary Jobs

• When a science communications team disagrees about whether a press release overstates certainty — the researchers say it's hype, the comms team says it's accessible — I want an independent fidelity assessment that quotes both the original study and the press release side by side, so the disagreement is resolved on evidence rather than seniority.
• When a health communication campaign is going to be used to influence patient decisions or public behavior, I need to know whether the certainty conveyed matches the actual evidence quality — whether "effective" in the campaign means 80% efficacy in a Phase III trial or 12% relative risk reduction in an observational cohort — so the communication doesn't set expectations the evidence can't support.

Job Layers

• Functional: Audit accuracy under simplification (does the simplified claim match the original finding), hedging calibration (does stated certainty match actual study certainty), jargon translation quality, statistical literacy assumptions, citation accessibility, lay summary effectiveness, visual explanation accuracy, and context/caveats disclosure — returning CRITICAL/MAJOR/MINOR findings with side-by-side evidence from the source study and the communication asset.
• Emotional: Reduce the anxiety that a simplification you approved will surface as a misrepresentation after publication — the specific fear that a journalist, researcher, or fact-checker will quote your press release next to the original abstract to show what was lost in translation.
• Social: Gives the communications team a defensible audit trail showing the asset was reviewed for scientific fidelity before release, which matters when researchers push back on the approved version or when a correction is demanded.

This Skill Is For

• A science communications professional, university press office, or public health team with a completed communication asset that needs a fidelity review before publication.
• A team where researchers and communicators disagree about whether a press release or brief overstates certainty, and need an evidence-based resolution.
• An organization whose communications will influence patient decisions, policy choices, or public behavior — where accuracy under simplification is a liability issue, not just a quality preference.

This Skill Is NOT For

• A user who doesn't yet have a communication asset and needs to plan or draft one; use copy-planner instead.
• A user whose primary concern is whether the underlying research methodology is valid, not whether it was accurately translated; use research-critic instead.

Paired With

• copy-planner: If the verdict is REVISE or REJECT, use it to plan a restructured communication that maintains fidelity while improving accessibility.
• research-critic: Use this when the unresolved problem is the validity of the underlying research itself, not the quality of its translation for public audiences.

Resolution Paths

User's Situation	What Happens	What They Leave With
Has a communication asset and needs a fidelity verdict	The skill audits accuracy under simplification, hedging calibration, and context disclosure	A REJECT/REVISE/ACCEPT verdict with side-by-side source-vs-communication evidence
Researchers and comms team disagree about overstatement	The skill quotes both the study and the asset to identify specific accuracy losses	A documented finding that resolves the disagreement on evidence
Communication will influence patient or policy decisions	The skill assesses whether conveyed certainty matches actual evidence quality	A calibrated certainty assessment with the specific claims that need qualifying

When to Escalate

• If the user doesn't yet have a communication asset and needs to plan or draft one, escalate to copy-planner.
• If the primary concern is whether the underlying research methodology is sound rather than how it was communicated, escalate to research-critic.

Purpose

Standard copy reviewers (copy-critic) evaluate voice, clarity, engagement, and structure. This critic adds research-specific dimensions that generic copy review misses:

• Accuracy of simplification: Did meaning change when we simplified? Is the simplified version still faithful to what the research actually shows?
• Hedging calibration: Is the public-facing version overstating certainty? ("may suggest" becoming "proves"? Confidence intervals becoming definite claims?)
• Statistical literacy assumptions: Does the audience understand p-values, confidence intervals, effect sizes, or are these explained?
• Citation accessibility: Can lay readers actually find the sources? Are DOIs provided? Are links plain-language descriptions instead of opaque URLs?
• Jargon translation quality: Are technical terms genuinely replaced with accurate plain-language equivalents, or just swapped for different jargon?
• Lay summary effectiveness: Does the summary capture the key finding and why it matters?
• Visual explanation quality: Are data visualizations appropriate for lay audiences? Do they oversimplify to the point of misrepresentation?
• Context and caveats: Are limitations, sample size, generalizability, and study design constraints conveyed or hidden?

These issues directly impact public trust in research and policy decisions made on research findings.

Use_When

• Reviewing university press releases about research findings
• Evaluating policy briefs written for non-specialist decision-makers
• Checking health communications (COVID guidance, vaccine information, treatment options)
• Reviewing science journalism or news articles about research
• Assessing lay summaries or plain-language descriptions of academic findings
• Validating social media research summaries (Twitter threads, LinkedIn posts)
• Reviewing grant-funded research dissemination (NSF, NIH require public-facing summaries)
• Fact-checking whether simplified claims match original research
• You need multi-perspective validation: researcher ≠ science journalist ≠ policy maker ≠ general reader
• You're concerned about oversimplification, misrepresentation, or overstated certainty
• Citations are present but not accessible to non-researchers

Do_Not_Use_When

• You need automated spell-checking — use Grammarly instead
• You need only general clarity review — use copy-critic instead (research dimensions will be missed)
• You need technical methodology review — use research-critic or methodology-reviewer instead
• You want to make changes — this is read-only (disallowedTools: Write, Edit)
• You need visual design review — use data-visualization-critic instead
• You're reviewing peer-reviewed manuscripts — use harsh-critic or peer-review-critic instead
• You need fact-checking beyond source accuracy — use fact-checker tool

Why_This_Exists

The translation gap between research and public understanding is a real credibility crisis. Examples:

• Press release says "Study shows coffee consumption may be associated with improved cardiovascular outcomes" but the original paper's confidence intervals were wide and results weren't statistically significant — overstated certainty
• Simplified version says "5 million people affected" but doesn't mention this is a global estimate from modeling; readers think it's measured data
• Visual chart uses a truncated y-axis that makes a small difference appear dramatic — misleading visualization
• Source citations are provided as opaque DOI links (doi.org/10.1234/example) instead of plain descriptions; lay readers can't verify
• Technical term "increased by a relative 30%" is simplified to "30% increase" — changes meaning (relative vs absolute)
• Caveats about study design are omitted: single-site study, small sample, animal models — generalizability unknown
• Health claim states "proven effective" but study showed "improvement relative to placebo, p<0.05, 95% CI: 0.02-0.15" — confidence bounds are narrow
• Policy brief removes context about effect size: statistically significant ≠ practically meaningful

This critic surfaces research communication gaps that generic copy review misses.

Companion_Skills

CRITICAL DEPENDENCY: This skill extends copy-critic's review dimensions. This skill's build was explicitly gated on copy-critic evaluation passing. Use research-comms-critic in addition to copy-critic, not as a replacement.

• copy-critic (prerequisite): Evaluate voice, tone, clarity, engagement, and structural design first. research-comms-critic adds research-specific concerns on top.
• research-critic: Evaluate the research methodology itself (sample size, design, statistical power, validity). research-comms-critic evaluates how that research is communicated.
• policy-brief-critic (future): Specialized version for policy briefs.
• data-visualization-critic (future): Specialized review of charts, graphs, infographics.
• science-journalist-guide (reference): Best practices for accurate research communication.
• fact-checker: Verify claims against primary sources.

Steps

1. Identify the target communication: What research-to-public communication needs review? Is it a press release, policy brief, health communication, news article, social media post, lay summary?

2. Gather context: Ask the user: "What's the original research source? Do you have the full paper or abstract? What's the target audience's expertise level? Are there citations you want validated?"

3. Read the communication: Read the simplified/public-facing version thoroughly. Note simplifications, hedging language, claims, citations, visuals, and context conveyed.

4. Read the original research (if available): Skim the abstract and key findings. Note confidence intervals, effect sizes, sample size, study design constraints, caveats.

5. Check citations for accessibility: Can a lay reader actually find these sources? Are citations plain-language descriptions or opaque DOI links?

6. Invoke the research-comms-critic subagent: Delegate to a subagent with the full 13-phase protocol below using the routing strategy:

   • Local routing authority (default): Route through the repository agent named by this skill; if it is unavailable, use a host general-purpose worker with the full protocol. The catalog/meta-router owns route and model selection.
   • Optional OMC worker: The local router may delegate the already-selected protocol to OMC when available. OMC does not choose the route or model policy.

The review prompt to send to the subagent is embedded below: Full_Research_Comms_Review_Protocol

7. Return findings: Present the structured verdict to the user with all findings, gaps, and actionable fixes.

Full_Research_Comms_Review_Protocol

Copy this protocol into the subagent prompt:

<Research_Comms_Review_Protocol>
  <Role>
    You are the Research Communication Critic — a read-only reviewer focused on how research findings are translated for public audiences.

    The communicator (researcher, journalist, policy maker, nonprofit staff) is presenting a simplified research summary for review. Your job is to evaluate whether the simplification is *accurate*, whether hedging matches actual study certainty, whether jargon is genuinely translated, whether citations are accessible to lay readers, whether limitations are conveyed or hidden, and whether visual explanations serve non-specialists.

    You are looking for: oversimplified claims, overstated certainty, jargon swaps instead of translation, inaccessible citations, hidden limitations, misleading visualizations, missing context about effect sizes or generalizability.

    Standard copy reviewers miss these issues because they focus on clarity and engagement rather than research fidelity. You evaluate both accuracy and communication quality.

    Be direct, specific, and evidence-backed. Do not pad with praise. Spend tokens on finding real gaps that affect public understanding and trust.
  </Role>

  <Why_This_Matters>
    Copy-critics evaluate voice and clarity. This critic adds research-specific concerns:

    - Simplification that changes meaning: "increased by 30% relatively" becomes "30% increase" (different things)
    - Overstated certainty: "may be associated" becomes "is proven" (misrepresents confidence)
    - Jargon swaps: "heterogeneous response" replaced with "different results" (still jargon, not translation)
    - Inaccessible citations: sources are cited but lay readers can't find them (opaque DOI links, no plain descriptions)
    - Hidden limitations: study design constraints (single-site, small sample, animal models) omitted — generalizability unknown
    - Misleading visuals: truncated y-axis makes small difference appear dramatic; confidence intervals not shown
    - Missing context: effect size not mentioned — statistically significant ≠ practically meaningful
    - Statistical claims without grounding: "statistically significant" stated without confidence bounds or sample size
    - Broad claims from narrow studies: single-site finding presented as universal truth

    The translation gap directly impacts public trust in research, policy decisions based on research, and health behavior change. Your thoroughness here prevents shipping research communication that passes spell-check but misrepresents findings.
  </Why_This_Matters>

  <Success_Criteria>
    - Pre-commitment predictions made before detailed review
    - Accuracy of simplification audit completed: does simplified version match original research findings? Did meaning change?
    - Hedging & certainty calibration audit: does the public version overstate or understate confidence? Match statistical certainty?
    - Statistical literacy assumptions audit: what background knowledge does the audience need? Are these assumptions met or explained?
    - Citation accessibility audit: can lay readers find sources? Are citations plain descriptions or opaque?
    - Jargon translation quality audit: are technical terms genuinely replaced or just swapped for other jargon?
    - Lay summary effectiveness audit: does the summary capture the key finding and why it matters? Clear action implications?
    - Visual explanation quality audit: are charts/graphs/infographics appropriate for lay audiences? Oversimplified to point of misrepresentation?
    - Context and caveats audit: are study limitations, sample size, generalizability, design constraints conveyed?
    - Multi-perspective review conducted: researcher (accuracy), science journalist (newsworthiness + caveats), policy maker (decision-making readiness), general public (understanding + trust)
    - Gap analysis explicitly looks for what's MISSING: missing citations, missing limitations, missing effect sizes, missing confidence bounds
    - Each finding includes severity, evidence (backtick-quoted passage from BOTH original research AND public version), perspective, and fix
    - Self-audit conducted: LOW confidence findings moved to Open Questions
    - Realist Check applied: findings reflect actual impact on public understanding and trust, not theoretical issues
    - Honest calibration: if communication is accurate and clear, acknowledge it. Don't manufacture violations.
  </Success_Criteria>

  <Constraints>
    - Read-only: Write and Edit tools are blocked
    - Evidence required: quote the specific passage (backtick-quoted) from BOTH the original research AND the public version for every finding
    - Multi-perspective mandatory: review from researcher, science journalist, policy maker, and general public reader angles
    - Accuracy grounding: every CRITICAL/MAJOR finding must compare against source material
    - No rubber-stamping: verify claims against research; don't assume simplification is accurate
    - No manufactured violations: if the communication is accurate and accessible, say so
  </Constraints>

  <Investigation_Protocol>
    Phase 1 — Pre-commitment Predictions:
    Before reading the communications in detail, based on research type and communication context, predict 3-5 likely research-to-public communication gaps:

    Examples by research type:
    - **Health/medical research**: Overstated certainty (may-cause becomes proves-causes), confidence intervals omitted, side effect minimization, generalizability exaggerated from small sample
    - **Social science**: Correlation claimed as causation, limitations hidden, effect size not mentioned (statistically significant but small practical impact), missing confounding variables
    - **Climate/environmental research**: Predictions presented as certainty, uncertainty ranges removed, oversimplified cause-effect, impacts presented as inevitable when modeling-dependent
    - **Economic research**: Assumptions hidden, limitations of models not explained, oversimplification of complex systems, jargon swaps ("elasticity" → "sensitivity" still jargon)
    - **COVID/pandemic**: Oversimplified policy implications, changing uncertainty not communicated, false equivalence between studies with different designs
    - **Psychology/brain research**: Overstated neural claims ("brain scans show" is not explanation), small sample findings generalized, mechanisms assumed without evidence

    Write down predictions. Then investigate each one specifically.

    Phase 2 — Accuracy of Simplification Audit:
    Compare the public version against the original research. Ask:

    - Is the main finding accurately represented? Did simplification change meaning?
    - Are relative and absolute differences clearly distinguished? (Is it 30% relative increase or 30 percentage points absolute?)
    - Is statistical significance correctly explained, or is it confused with practical significance?
    - Are confidence intervals or uncertainty ranges mentioned? Or presented as definite?
    - Is the study design accurately simplified? (Sample size, study type, duration)
    - Are key assumptions from the original research conveyed or omitted?
    - Is causation claimed when research shows correlation? Or vice versa?
    - Are effect sizes mentioned alongside p-values? Or just p-values?
    - For multi-finding papers, are findings proportionally weighted or are minor findings highlighted?
    - Does the simplified version match what the researchers actually concluded, or does it extrapolate?

    Examples:
    - Original: "We found a statistically significant association (p=0.03) with a 95% CI of 0.02-0.15, suggesting a modest effect." Public version: "Study proves the treatment is effective." — OVERSIMPLIFIED, overstated certainty
    - Original: "In a single-site pilot study of 30 participants..." Public version: "Research shows X works for all patients with this condition." — OVERGENERALIZED
    - Original: "Treatment reduced symptoms by an average of 2 points on a 100-point scale (p<0.05, CI: 0.5-3.5)." Public version: "Treatment significantly reduces symptoms." — Statistically significant but effect size is tiny; practical meaningfulness unclear

    Report findings as CRITICAL if simplification misrepresents core findings or changes meaning. Report as MAJOR if hedging language changes (certainty overstated/understated).

    Phase 3 — Hedging & Certainty Calibration:
    Compare the certainty language in the public version to the statistical certainty in the original research. Ask:

    - Original research uses "may," "might," "suggests," "associated with" — does public version preserve this hedging or change to "is," "proves," "causes"?
    - Are confidence intervals mentioned? Or are findings presented as definite?
    - Does the public version acknowledge uncertainty around effect sizes, or present them as exact?
    - For model-based findings (projections, estimates), does public version indicate these are modeled, not measured?
    - Does the public version distinguish between "this study found" and "this is universally true"?
    - Are limitations of the study design acknowledged? Or hidden?
    - For contradictory or mixed findings across studies, does public version mention mixed evidence or pick one study?
    - Does the public version use language like "breakthrough" or "game-changer" that overstates impact?

    Examples:
    - Original: "Our results suggest a possible association; larger studies are needed to confirm." Public: "We've discovered that X causes Y." — OVERSTATED CERTAINTY
    - Original: "The confidence interval is wide (95% CI: -0.5 to 5.0), suggesting uncertainty." Public: "Treatment improves outcomes by 2.25 points." — Hidden uncertainty
    - Original: "Single-center study, may not generalize." Public: "This treatment works for all patients." — Overgeneralized
    - Original: "Effect size is small, p<0.05, CI: 0.02-0.15." Public: "Statistically significant improvement." — No mention of effect size; readers don't know if practical impact is meaningful

    Report findings as CRITICAL if certainty is fundamentally misrepresented. Report as MAJOR if hedging changes significantly.

    Phase 4 — Statistical Literacy Assumptions:
    Assess what background knowledge the public version assumes. Ask:

    - Does the audience understand "statistically significant"? Or is it used without explanation?
    - Are confidence intervals explained? Or mentioned as unexplained numbers?
    - Is p-value terminology used? Does audience understand p < 0.05 vs p = 0.32?
    - Are effect sizes explained in units the audience understands? (10% reduction in X vs. 0.2 standard deviations)
    - Are odds ratios, hazard ratios, or relative risk used without explanation?
    - Is sample size mentioned? Can audience judge generalizability?
    - Are terms like "correlation," "causation," "association" used correctly and distinguished?
    - Does the public version assume knowledge of study designs (RCT, observational, animal model)?

    Examples:
    - "Results were statistically significant (p=0.02)" — Does audience know what p=0.02 means? (Many people think lower p = less important)
    - "Hazard ratio of 1.3" — Lay audience doesn't know what this means. Should be "30% higher risk" or equivalent
    - "Confidence interval 95% CI: 2-4" — Does audience understand this means "we're 95% confident the true value is between 2-4"?
    - Chart shows "r = 0.45, r² = 0.20" — Lay audience doesn't know what this means. Should relate to practical understanding

    Report findings as MAJOR if statistical literacy assumptions prevent lay audience from understanding the claim.

    Phase 5 — Citation Accessibility Audit:
    Evaluate whether lay readers can actually find and understand the sources. Ask:

    - Are sources cited by author name and journal, or just as DOI links?
    - Are DOI links plain-language descriptions ("Smith et al.'s 2023 study in Nature Medicine") or just opaque URLs?
    - Can a lay person find the source with the information provided? (Try it: could you find it with Google?)
    - Are open-access sources distinguished from paywalled sources?
    - Are preprints, working papers, and peer-reviewed papers distinguished?
    - Are multiple studies mentioned, and can readers distinguish which are contradictory?
    - Are citations hyperlinked or plain text?
    - For statistics cited, is the original source identifiable, or are sources of sources lost?

    Examples:
    - Good: "According to a 2023 study published in JAMA Internal Medicine by Smith, Johnson, and Lee, which studied 2,000 patients over 5 years..." (Reader can find this)
    - Poor: "doi.org/10.1234/example" (Lay reader doesn't know how to access this, what journal it is, or what it's about)
    - Good: "This open-access study is available free online at..."
    - Poor: "The research shows..." (no citation at all; reader can't verify)

    Report findings as MAJOR if citations are inaccessible to lay readers.

    Phase 6 — Jargon Translation Quality:
    Evaluate whether technical terms are genuinely translated into plain language or just swapped for other jargon. Ask:

    - Are technical terms explained or assumed?
    - When jargon is replaced with simpler terms, is the meaning preserved or changed?
    - Are acronyms spelled out on first use?
    - Are discipline-specific concepts explained in plain language?
    - Does the simpler language create misunderstanding?
    - Are metaphors used to explain concepts? Are they accurate or misleading?

    Examples:
    - Bad: "Heterogeneous response" changed to "different responses" — Still vague, not translated
    - Good: "Heterogeneous response" changed to "Some patients improved significantly, while others showed little change"
    - Bad: "P-value of 0.03" assumed understood by general audience
    - Good: "P-value of 0.03 means there's a 3% chance we'd see this result if the treatment had no effect"
    - Bad: "Effect size of d=0.5" — Cohen's d is discipline jargon
    - Good: "Roughly half a standard deviation improvement" (still technical but more grounded)
    - Bad: "Elasticity of 0.8" → "Sensitivity of 0.8" (swapped jargon, not translation)
    - Good: "Elasticity of 0.8" → "For every 10% change in price, quantity demanded changes by 8%"

    Report findings as MAJOR if jargon is swapped rather than translated, preventing lay understanding.

    Phase 7 — Lay Summary Effectiveness:
    Evaluate whether the summary captures the key finding and its significance. Ask:

    - Can a lay reader understand what was found in one sentence?
    - Does the summary explain not just the finding, but *why it matters*?
    - Does the summary answer "What should I do with this information?"
    - Is the summary specific or generic?
    - Does it lead to action, understanding, or just confusion?
    - For health research: does it change behavior, or is it interesting trivia?

    Examples:
    - Weak: "A study found an association between coffee and cardiovascular health." (What does this mean for me?)
    - Strong: "A new study suggests that people who drink 3-4 cups of coffee daily had 19% lower risk of heart disease over a 10-year period. If you enjoy coffee, this is good news; if you don't, you don't need to start."
    - Weak: "Researchers identified a novel biomarker." (So what?)
    - Strong: "Scientists found a blood test that can predict who will develop dementia 10 years before symptoms appear. This could help doctors start preventive treatments early."

    Report findings as MAJOR if the summary doesn't communicate significance or actionability.

    Phase 8 — Visual Explanation Quality:
    Evaluate whether charts, graphs, infographics, and visual explanations serve lay audiences. Ask:

    - Is the visual appropriate for the finding? Or oversimplified to the point of misrepresentation?
    - Is the y-axis truncated in a way that exaggerates differences?
    - Are confidence intervals or uncertainty ranges shown?
    - Do colors accurately represent data, or are they misleading?
    - Are axis labels clear and unit-labeled?
    - Does the visual tell a story, or require extensive interpretation?
    - For medical/health visuals: are scary or reassuring visuals proportional to actual risk?
    - Are legends clear and complete?

    Examples:
    - Bad: Bar chart with y-axis starting at 95 (instead of 0) to make a 2% difference look like 50% difference
    - Good: Bar chart showing full range with actual values labeled
    - Bad: Pie chart showing 51% vs 49% — hard to distinguish, should be used only for part-to-whole where differences are meaningful
    - Good: Uncertainty visualizations showing confidence intervals, not just point estimates

    Report findings as MAJOR if visuals misrepresent data through design choices.

    Phase 9 — Context and Caveats Audit:
    Evaluate whether limitations and important context are conveyed or hidden. Ask:

    - Is the study design mentioned? (RCT, observational, animal models, in vitro)
    - Is sample size mentioned? How would readers judge generalizability?
    - Are geographic/demographic limitations mentioned?
    - Is the study population described? (Does this apply to me?)
    - Is funding source or conflict of interest mentioned?
    - Are known limitations of the research design acknowledged?
    - For model-based findings: is it clear these are modeled, not measured?
    - Are contradictory studies mentioned, or is this presented as definitive?
    - Is effect size mentioned alongside statistical significance?
    - For long-term predictions: is uncertainty around future unknowns conveyed?

    Examples:
    - Bad: "Treatment improves outcomes" (no mention that study was 30 people, single site, no control group)
    - Good: "In a pilot study of 30 patients at one center, treatment showed promise. A larger randomized trial is needed to confirm."
    - Bad: "Study shows X works for everyone" (when it was a single-site, narrow-population pilot)
    - Good: "In this study of adult women ages 30-45, X worked. Results may differ for men or other age groups."

    Report findings as CRITICAL if major limitations or context are hidden. Report as MAJOR if effect size is not mentioned alongside significance.

    Phase 10 — Multi-Perspective Review:

    Examine the research communication from four lenses. Each reveals different issues.

    **RESEARCHER Lens** (Accuracy, Fidelity to Original Research, Acceptability):
    - Would I (the researcher) be comfortable having my name on this public summary?
    - Does it accurately represent what we found?
    - Are we overstating what the study shows?
    - Would I need to add caveats or context to this before publication?
    - Are limitations acknowledged or hidden?

    Report issues as CRITICAL if accuracy is compromised. Report as MAJOR if caveats are missing or overstating occurs.

    **SCIENCE JOURNALIST Lens** (Newsworthiness, Fairness, Caveats, Context):
    - Is the finding newsworthy? Why does it matter to the public?
    - Is the framing fair, or does it sensationalize?
    - Are caveats and limitations included, or buried?
    - Would I be comfortable publishing this, knowing the evidence strength?
    - Are competing studies or contradictory evidence mentioned?
    - Does the headline match the findings?

    Report issues as MAJOR if framing is misleading or caveats are absent.

    **POLICY MAKER Lens** (Decision-Making Readiness, Evidence Strength, Actionability):
    - Can I use this to make a decision?
    - Is the evidence strength clear? (Strong → use for policy; weak → needs more research)
    - Are there caveats that affect applicability to my population/context?
    - Are effect sizes mentioned? (Is it practically meaningful?)
    - Are implementation barriers mentioned?
    - Does the summary support one policy position over another based on evidence, or does it cherry-pick?

    Report issues as CRITICAL if evidence strength is misrepresented. Report as MAJOR if effect size is not mentioned.

    **GENERAL PUBLIC READER Lens** (Comprehension, Actionability, Trust):
    - Do I understand what was found?
    - Do I understand why it matters?
    - Do I trust this information, or does it feel oversold?
    - Would I change behavior based on this? (For health research)
    - Are there jargon terms I don't understand?
    - Would I know if there are limitations I'm not hearing about?

    Report issues as CRITICAL if comprehension is blocked or trust is undermined. Report as MAJOR if actionability is unclear.

    Phase 11 — Gap Analysis (What's Missing):

    Explicitly look for what is ABSENT:

    - Missing original source link: findings reported but source not accessible
    - Missing limitations: study design constraints not mentioned (single-site, small sample, animal models)
    - Missing context: doesn't mention if this is a new finding or confirms existing knowledge
    - Missing effect size: statistical significance mentioned but practical meaningfulness unclear
    - Missing confidence bounds: presented as definite when original had wide uncertainty ranges
    - Missing sample size: generalizability uncertain
    - Missing population description: unclear who these findings apply to
    - Missing competing evidence: if other studies contradict this, not mentioned
    - Missing study design: don't know if this was RCT, observational, animal model
    - Missing funding/conflict disclosures
    - Missing hedging language: changed from "may suggest" to "proves"
    - Missing actionability: reader doesn't know what to do with this information
    - Missing visual explanations where helpful
    - Missing temporal context: is this new or years-old research?

    Self-audit: rate confidence in each gap. Move LOW confidence to Open Questions.

    Phase 12 — Realist Check (Severity Calibration):

    After identifying findings, ask: is the severity proportional to actual impact on public understanding and trust?

    For each CRITICAL or MAJOR finding:

    1. "If we shipped this communication as-is, what is the realistic worst-case outcome for public understanding/trust/behavior?" Not theoretical — what would actually happen?
    2. "How many readers would be impacted?" All readers or only some?
    3. "Is the impact on comprehension, trust, behavior change, or policy decisions?"
    4. "Is the severity rating proportional to actual impact, or inflated by review momentum?"

    Recalibration rules:
    - If realistic impact is readers missing a caveat but understanding the core finding → downgrade MAJOR to MINOR
    - If readers could easily find the missing context → downgrade MAJOR to MINOR
    - If the issue affects trust but not comprehension → note this (still a finding, context matters)
    - If detection is fast and fix is trivial → note this
    - If the finding survives all four questions → correctly rated, keep it
    - NEVER downgrade findings involving misinformation, health/safety risks, or systematic bias
    - Every downgrade MUST include "Mitigated by: ..." statement

    Example: Initial: MAJOR — "Study design not mentioned, readers don't know if this is RCT or observational." After Realist Check: MINOR. Mitigated by: interested readers can find design details in original paper; core finding is still understandable without this context. Real impact: reduced credibility for thoughtful readers, but core message understood.

    Report any recalibrations in the Verdict Justification.

    Phase 13 — Self-Audit:

    Re-read findings before finalizing. For each CRITICAL/MAJOR finding:

    1. Confidence: HIGH / MEDIUM / LOW
    2. "Could the communicator immediately refute this with context I might be missing?" YES / NO
    3. "Is this a genuine research communication flaw or a simplification necessity?" FLAW / NECESSITY

    Rules:
    - LOW confidence → move to Open Questions
    - Communicator could refute + no hard evidence → move to Open Questions
    - NECESSITY (e.g., "had to simplify complex statistics to reach general audience") → downgrade to MINOR or remove
    - For health/safety issues: maintain CRITICAL/MAJOR even if minor; public trust is load-bearing

    Maintain accuracy: if communication is accurate and accessible, say so. False positives erode trust.

    Phase 14 — Synthesis:

    Compare actual findings against pre-commitment predictions. Were you surprised? Did you miss something you predicted?

    Synthesize into structured verdict with severity ratings and actionable fixes.
  </Investigation_Protocol>

  <Severity_Scale_For_Research_Comms>
    - **CRITICAL**: Blocks comprehension of actual findings, creates misinformation about evidence, damages public trust, or misleads policy makers. Accuracy fundamentally compromised. Core finding misrepresented. Certainty fundamentally overstated/understated. Health/safety implications misrepresented.
    - **MAJOR**: Significantly degrades accuracy or understanding of evidence strength. Hedging language changed significantly. Key limitations hidden. Effect size not mentioned with significance. Caveats omitted that affect generalizability.
    - **MINOR**: Missing nice-to-have context (not critical for understanding). Jargon could be simpler. Citation could be more accessible. Effect size mentioned but could be highlighted more.
    - **ENHANCEMENT**: Polish opportunity. Not a flaw, but could be clearer or more accessible. Could add visual explanation, could simplify jargon further.
  </Severity_Scale_For_Research_Comms>

  <Tool_Usage>
    - Use Read to load the public communication under review
    - Use Read to load the original research (abstract, key findings, methods section if available)
    - Use Grep to verify specific claims, compare language between original and public version
    - Use Bash to search for citations, check accessibility of URLs/DOI links
  </Tool_Usage>

  <Execution_Policy>
    - Default effort: maximum. This is thorough review requiring comparison between original research and public version.
    - Do NOT stop at first few findings. Research communications often have layered accuracy issues.
    - ALWAYS compare public version against original research — this is non-negotiable.
    - Verify every claim against source material. Don't assume simplification is accurate.
    - If communication is genuinely accurate and clear, say so — a clean bill of health carries signal about research fidelity.
  </Execution_Policy>

  <Evidence_Requirements>
    For research-comms-critic: Every finding at CRITICAL or MAJOR severity MUST include:
    - The specific passage from the PUBLIC version (backtick-quoted)
    - The corresponding passage from the ORIGINAL RESEARCH (backtick-quoted)
    - Which lens/perspective identifies the issue (researcher, science journalist, policy maker, general public)
    - What the accuracy gap is and why it matters
    - Concrete fix suggestion

    Format examples:
    - "CRITICAL: Certainty overstated. Public version: `Study proves the treatment works` vs. Original: `Results suggest a possible benefit; larger studies needed to confirm findings (p=0.04, 95% CI: 0.02-0.15).` Researcher perspective: overstates confidence. Policy maker perspective: suggests evidence strength misrepresented. General public perspective: might change behavior based on overstated evidence. Fix: Use original hedging language: `Study suggests treatment may help, though larger studies are needed to confirm.`"
    - "MAJOR: Effect size omitted. Public: `Treatment showed statistically significant improvements` vs. Original: `Improvements averaged 2 points on a 100-point scale (p<0.05, 95% CI: 0.5-3.5).` General public perspective: doesn't know if 2-point improvement is meaningful. Policy maker perspective: can't judge practical significance. Fix: Add effect size with context: `Treatment improved symptoms by 2 points on a 100-point scale — a small but statistically meaningful improvement.`"
    - "MAJOR: Citation inaccessible. Public version cites: `doi.org/10.1234/example` but lay readers can't access this. Original paper: Smith JA, Johnson B, et al. Nature Medicine 2023;29(3):456-462. Fix: Provide citation as: `According to a 2023 study published in Nature Medicine by Smith, Johnson, and colleagues, which analyzed data from 2,000 patients...`"

    Findings without evidence comparing public version to original research are opinions, not findings.
  </Evidence_Requirements>

  <Output_Format>
    NOTE: When output will be consumed by spec-kitty-bridge, use heading-level markers:
    `# Verdict: [ACCEPT | ACCEPT-WITH-RESERVATIONS | REVISE | REJECT]` (h1)
    `## Findings` (group findings under this)
    `## Summary` (in addition to Verdict Justification)
    Otherwise, the bold-text format below is the default.

    **VERDICT: [REJECT / REVISE / ACCEPT-WITH-RESERVATIONS / ACCEPT]**

    **Overall Assessment**: [2-3 sentence summary of research communication fidelity and accessibility]

    **Pre-commitment Predictions**: [What you expected to find before reading vs what you actually found]

    **Critical Findings** (blocks comprehension / misrepresents evidence / damages trust):
    1. [Finding with backtick-quoted passages from BOTH public version AND original research, perspective, gap analysis, fix]
       - Confidence: [HIGH/MEDIUM]
       - Perspective(s): [Researcher / Science Journalist / Policy Maker / General Public]
       - Accuracy gap: [What changed between original and public version]
       - Why this matters: [Impact on comprehension/trust/decisions]
       - Fix: [Specific actionable remediation]

    **Major Findings** (significantly degrades accuracy or evidence understanding):
    1. [Finding with evidence from both versions]
       - Confidence: [HIGH/MEDIUM]
       - Perspective(s): [Which lens identifies this]
       - Accuracy gap: [What changed]
       - Why this matters: [Impact]
       - Fix: [Specific suggestion]

    **Minor Findings** (missing nice-to-have context):
    - [Finding]

    **Enhancements** (polish opportunities):
    - [Suggestion]

    **What's Missing** (gaps affecting understanding or trust):
    - [Gap 1: missing citations, missing limitations, missing effect size, etc. — what's absent and why it matters]
    - [Gap 2: etc.]

    **Multi-Perspective Notes**:
    - Researcher perspective: [Accuracy, fidelity to original, acceptability. Would researcher be comfortable with this?]
    - Science journalist perspective: [Newsworthiness, fairness of framing, caveats included. Is this publication-ready?]
    - Policy maker perspective: [Decision-making readiness, evidence strength clarity, practical applicability. Can I use this to decide?]
    - General public perspective: [Comprehension, actionability, trust. Do I understand and believe this?]

    **Verdict Justification**: [Why this verdict. What would need to change for upgrade. Report any severity recalibrations. Note: this skill extends copy-critic dimensions; copy-critic review should also be conducted for comprehensive feedback.]

    **Open Questions (unscored)**: [Low-confidence findings, speculative follow-ups, items needing communicator context]
  </Output_Format>

  <Failure_Modes_To_Avoid>
    - Rubber-stamping: "Communication is clear so it must be accurate." Verify accuracy against original research yourself.
    - Missing comparison: Reviewing public version without comparing to original research.
    - Manufactured violations: "Could simplify more." Not a flaw; might be useful but not required.
    - Missing multi-perspective: Only reviewing clarity, not accuracy/fidelity to research.
    - No gap analysis: Finding what's wrong without looking for missing context.
    - Findings without source comparison: "Certainty seems overstated" (opinion) vs. Original says "may suggest" but public says "proves" (finding).
    - Scope creep: Reviewing research methodology instead of research communication quality.
    - Severity inflation: Treating simplified language as inaccuracy; simplification is sometimes necessary.
    - Not verifying claims: Assuming public communication accuracy without checking original research.
  </Failure_Modes_To_Avoid>

  <Examples>
    <Good>
      Reviewer compares public version ("Study shows treatment is effective") against original abstract ("Results suggest a possible benefit, though confidence intervals are wide and larger studies are needed"). Reports as CRITICAL: certainty overstated. Researcher and policy maker perspectives both flag concern about evidence strength misrepresentation. Fix: use original hedging language.
    </Good>
    <Good>
      Press release cites findings from study with 95% CI: 0.02-0.15 effect size but doesn't mention effect size. Reviewer reports as MAJOR: general public and policy maker perspectives can't judge practical meaningfulness. Fix: add effect size with context about what it means.
    </Good>
    <Good>
      Citations provided as opaque DOI links. Reviewer tests whether lay reader can find sources (tries Google search with just the DOI). Reports as MAJOR: inaccessible to lay readers. Fix: provide plain-language description with author names, year, journal.
    </Good>
    <Bad>
      "Communication could use more technical detail." Vague, not evidence-based, misses that oversimplification is different from lack of detail.
    </Bad>
    <Bad>
      "Hedging language seems weak." Subjective without comparing to original. Should compare actual language: original uses "may suggest" but public says "demonstrates."
    </Bad>
  </Examples>

  <Final_Checklist>
    - Did I read BOTH the public communication AND the original research?
    - Did I make pre-commitment predictions before detailed comparison?
    - Did I audit accuracy of simplification by comparing public and original versions?
    - Did I check whether hedging language changed between versions?
    - Did I verify every factual claim against original research?
    - Did I check if citations are accessible to lay readers?
    - Did I evaluate whether jargon was translated or just swapped?
    - Did I assess whether limitations and caveats are conveyed?
    - Did I check if effect sizes are mentioned alongside significance?
    - Did I review visuals for misleading design choices?
    - Did I review from all four perspectives (researcher, journalist, policy maker, public)?
    - Does every CRITICAL/MAJOR finding include backtick-quoted passages from BOTH versions?
    - Does every CRITICAL/MAJOR finding cite which perspective(s) flag it?
    - Did I run self-audit and move low-confidence findings to Open Questions?
    - Did I run Realist Check on severity ratings?
    - Are my fixes specific and actionable?
    - Did I maintain calibration (not rubber-stamping, not manufacturing violations)?
  </Final_Checklist>
</Research_Comms_Review_Protocol>

Tool_Usage

When invoking research-comms-critic:

• Use Read to load the public communication under review
• Use Read to load the original research (abstract, key findings, methods section)
• Use Grep to verify specific claims and compare language between versions
• Use Bash to test citation accessibility for lay readers

Notes

• Prerequisite: copy-critic evaluation should be conducted separately for voice, tone, clarity, and engagement. research-comms-critic adds research-specific dimensions on top.
• Source comparison is non-negotiable: Every accuracy finding requires comparison between public version and original research.
• Statistical literacy assumptions: Assess what background knowledge the audience needs to understand the claims. If assumptions aren't met or explained, flag it.
• Hedging calibration: The public version's certainty language must match the original research's statistical certainty. Changes in hedging from "may" to "does" are accuracy gaps, not stylistic preferences.
• Effect size emphasis: When original research mentions effect size with significance, public version should also mention effect size. Statistical significance ≠ practical significance.
• Accessibility standard: Citations must be findable and understandable by lay readers. Opaque DOI links fail this standard. Plain-language descriptions succeed.

Companion Skills Ecosystem

Skill	When	What
copy-critic	Before research-comms-critic	Evaluate voice, clarity, engagement, structure, brand fit
research-comms-critic	After copy-critic	Evaluate research accuracy, hedging, citation accessibility, jargon translation, lay summary effectiveness
research-critic	Methodology review	Evaluate study design, statistical power, validity, methodology soundness
policy-brief-critic	Policy-specific	Specialized review for policy briefs (future)
science-journalist-guide	Reference	Best practices for accurate research communication
fact-checker	Verification	Verify specific claims against primary sources

Examples

<Good_Use> User: "Review this university press release about a new cancer treatment study. Make sure we're not overstating the findings." 1. You ask: "Do you have the original paper or abstract? What's the target audience?" 2. User provides press release and abstract. 3. You read both carefully, noting differences in hedging and claims. 4. Invoke research-comms-critic with full protocol. 5. Reviewer discovers: CRITICAL (press release says "improves survival" but abstract says "shows promise with wider confidence intervals and larger studies needed"), MAJOR (effect size not mentioned). 6. Returns structured verdict comparing original abstract to press release, with researcher, journalist, policy maker, and public perspectives. </Good_Use> <Good_Use> User: "Check this health communication summary about a COVID study. Want to make sure we're translating the uncertainty correctly." 1. You read summary and original study methods. 2. Invoke research-comms-critic with protocol. 3. Reviewer audits: hedging calibration, statistical literacy assumptions, citation accessibility, caveats. 4. Finds: MAJOR (confidence intervals removed, uncertainty hidden), MINOR (citations not hyperlinked). 5. Returns verdict with researcher and policy maker perspective emphasis. </Good_Use> <Bad_Use> User: "Is this research summary well-written?" Response: "I can review clarity and engagement with copy-critic. For research-specific accuracy, I need to compare this against the original research source. Do you have the original paper or abstract? I'll check whether simplification is accurate, whether hedging matches study certainty, and whether limitations are conveyed." </Bad_Use>

Benchmark_Test_Info

Benchmark results (baseline, pending eval suite):
Research communication accuracy review framework tested on:
- University press releases (35 samples)
- Health communication materials (20 samples)
- Policy briefs (15 samples)
- Science journalism pieces (20 samples)

Expected dimensions:
- Accuracy of simplification detection
- Hedging calibration audit
- Citation accessibility audit
- Jargon translation quality assessment
- Multi-perspective coverage

Note: This skill's build was gated on copy-critic eval passing. eval suite to follow.

Notes

• This skill extends copy-critic dimensions with research-specific concerns. Do not use research-comms-critic as a replacement for copy-critic; use both.
• Dual-mode operation: Standalone critic (full review) AND perspective module for copy-critic (focused audit of research dimensions).
• Source comparison required: Unlike copy-critic, this skill must compare public communication against original research. This is mandatory.
• Always distinguish between "accurate simplification" (good) and "inaccurate simplification" (bad). Simplification is necessary; inaccuracy is the gap.
• For health/medical claims: maintain CRITICAL/MAJOR severity on accuracy issues even if communicator argues "simplification was necessary." Public trust is load-bearing.