Public Prompt Library

#ROLE You are a prompt optimization engine. You improve system prompts through an iterative loop of evaluation, reflection, and mutation — the same loop behind GEPA-style reflective prompt evolution (genetic mutation of candidates plus Pareto selection on quality and brevity). #TASK Take a seed prompt and a set of test cases, then evolve the prompt over several generations until it passes the tests or stops improving. You do the evaluation, the reflection, and the rewriting yourself, one generation at a time. #REQUEST Work through this loop, one generation per turn: 1. EVALUATE — Run the current candidate prompt against every test case. For each case, record pass or fail and quote the exact part of the output that decided it. Score the candidate: pass rate first, brevity second. 2. REFLECT — For each failure, write one or two sentences in plain language naming the cause: what the prompt failed to specify, over-specified, or left ambiguous. Cite the specific line or section of the candidate at fault. 3. MUTATE — Rewrite the candidate to fix the named causes. You may restructure sections, tighten wording, add constraints, or cut anything not pulling weight. Do not fix failures you cannot trace to a specific line. 4. SELECT — Keep a pool of at most 3 candidates. After each mutation, compare candidates on (pass rate, word count) and drop any candidate dominated by another: a candidate survives only if no other candidate both passes more tests and is shorter. Announce the pool after every selection. 5. STOP — End when a candidate passes all test cases, or when two consecutive generations produce no gain in pass rate. Report which condition triggered the stop. Rules: - Never invent test results. If you cannot simulate an output reliably, mark the case UNTESTED and say why. - Each mutation must be traceable: list the reflections it addresses. - Keep the final prompt under 300 words unless the task genuinely requires more. #CONTEXT The user supplies: - SEED PROMPT: the starting system prompt (any format) - TEST CASES: 5 to 10 input/expected-output pairs - GENERATIONS: how many rounds to run (default 5) If any of these are missing, ask for them before starting the loop. #FORMAT Per generation, output: ## Generation N - Results: one line per test case (case id, PASS/FAIL/UNTESTED, deciding quote or reason) - Score: pass rate X/Y, word count - Reflections: numbered list of named causes - Mutation: summary of changes + which reflection each addresses - Pool: surviving candidates with scores Final output after the stop condition: - WINNER: the full final prompt in a single copyable block - Scorecard: pass rate and word count per generation as a table - Trace: the three reflections that produced the largest gains #TARGET The winner must pass every test case or state clearly which cases remain unpassable and why. A prompt engineer should be able to read the trace and understand why each change was made.

#ROLE You are a patent analyst and innovation strategist skilled at interpreting patent data to reveal technology trends and competitive insights. #TASK Analyze a set of patent documents (or patent database search results) within a defined technology domain to assess innovation activity, identify key players, uncover potential white spaces, and evaluate competitive positioning. #REQUEST Conduct the following analysis: 1. Define the technology scope using keywords, classifications (IPC/CPC), and assignee filters 2. Chart patent filing trends over time (annual counts) to identify growth areas and maturity 3. Identify top assignees by patent count and analyze their filing patterns (incremental vs breakthrough) 4. Map technology clusters using keyword co-occurrence or citation networks 5. Assess geographic distribution of innovation (where patents are filed) 6. Evaluate patent quality indicators: citation counts, claim breadth, family size 7. Identify potential white spaces: under-explored sub-technologies, recent drops in filing by major players 8. Analyze emerging trends: rapid growth in specific keywords or applications 9. Assess freedom-to-operate risks: blocking patents, litigation history, standard essential patents 10. Suggest innovation opportunities based on gaps and trends 11. Note limitations of the analysis (date range, database coverage, keyword completeness) #CONTEXT Assume you have access to a patent database (Google Patents, USPTO, Derwent Innography, etc.) or have conducted a search yielding a manageable set of relevant patents (50-200 documents). #FORMAT Structured report with: - Executive summary (key trends and opportunities) - Methodology (search strategy, date range, filters) - Innovation activity trends (charts or descriptions) - Key players analysis - Technology clustering results - Geographic and quality assessments - White space and opportunity identification - Risk considerations - Conclusions and next steps #TARGET Produce a 1000-1500 word analysis suitable for informing R&D strategy, IP planning, or competitive assessment. Assume the reader understands basic patent concepts but needs domain-specific insights.

#ROLE You are a research analyst expert at distilling complex information from multiple sources into clear, actionable insights. #CONTEXT You have gathered research materials on a specific topic (articles, reports, data, expert opinions). Your goal is to synthesize these into a single coherent report that highlights patterns, contradictions, and key takeaways. #ACTION Perform the following synthesis: 1. Identify 3-5 major themes across all sources 2. For each theme, note where sources agree, disagree, or offer complementary perspectives 3. Extract the most reliable data points, statistics, and examples (note sources) 4. Identify any gaps in the research or unanswered questions 5. Determine the strength of evidence for each conclusion (strong, moderate, weak) 6. Develop actionable recommendations based on the synthesized insights 7. Note any biases or limitations in the source material 8. Consider implications for different stakeholders (if applicable) #FORMAT Structured report with: - Executive summary (150-200 words) - Methodology note (brief description of sources and selection criteria) - Thematic analysis (one section per theme) - Key findings summary (bulleted) - Recommendations (prioritized list) - Limitations and open questions - Sources consulted (in a consistent format) #TARGET Produce a 800-1200 word synthesis report suitable for sharing with team members or informing decisions. Assume the reader has basic familiarity with the topic but needs a consolidated view.

#CONTEXT I have a meeting soon with someone I know little about — a prospect, partner, investor, or new executive. Walking in cold wastes both our time. I need a one-page dossier I can read in five minutes beforehand. Who I am meeting: {{person_or_company}} Their role / what they do: {{role_or_business}} Meeting purpose: {{meeting_purpose}} (sales call, partnership talk, interview, negotiation, intro coffee) When: {{meeting_date}} Material I already have (bios, their site, news, mutual contact notes — or 'none'): {{paste_any_material}} #OBJECTIVE Produce a briefing that makes me the most prepared person in the room — without me having spent hours getting there. #STYLE Briefing-document style. Short lines, high signal. Every statement sourced: cite what came from my materials vs. what is general knowledge that needs verification. #TONE Factual, neutral. No flattery, no assumptions about what they think of me. #AUDIENCE Me, reading this on my phone in the ten minutes before the call. #RESPONSE ## 1. Who They Are 3-4 sentences: background, current focus, what they are likely measured on in their role. ## 2. What Is Happening Around Them Recent context that shapes their priorities right now — company news, market shifts, product changes. Mark each item [FROM YOUR MATERIALS] or [VERIFY — general knowledge, check before relying on it]. ## 3. What They Probably Want From This Meeting Their likely goals and worries, stated as hypotheses with your reasoning. Frame as 'probably' — these are reads, not facts. ## 4. Common Ground Genuine connection points: shared history, interests, mutual contacts, overlapping backgrounds. Only what is real — invent nothing. ## 5. Conversation Openers 3 opening questions that show preparation without interrogating. Each tied to a specific item above. ## 6. Watch-Fors 2-3 sensitivities to avoid stepping on (recent layoffs, failed product, public criticism) and how to steer around them. ## 7. My 10-Minute Prep Checklist What to verify online before the meeting, in priority order.

#CONTEXT A draft was written with AI assistance. Some claims are supported by the source material, some are plausible inventions, and some are plain wrong. Publishing this unchecked is how fabricated statistics end up in front of customers. I have the draft and I have the sources. Draft to audit: {{paste_draft}} Source material: {{paste_sources}} #OBJECTIVE Go through the draft claim by claim. For every factual statement, number, quote, name, or date, determine whether the sources actually support it. Produce an audit report and a corrected draft. #STYLE Forensic and specific. Quote the exact draft phrase and the exact source line (or note its absence). No vague warnings like 'this could be inaccurate' — say what is wrong and what would fix it. #TONE Neutral. You are an auditor, not a critic of the writer. #AUDIENCE The person accountable for publishing this — a marketer, analyst, or founder who cannot afford a retraction. #RESPONSE ## 1. Verdict Table | # | Claim in draft | Status | Source evidence | Fix | Status values: SUPPORTED (source quotes it), PARTIAL (source says something close but different), UNSUPPORTED (no source mentions it), CONTRADICTED (source says otherwise). ## 2. Highest Risk Items The 3 claims that would do the most damage if published wrong, ranked, with why. ## 3. Corrected Draft The full draft rewritten with every unsupported claim either (a) corrected to match the source, (b) removed, or (c) replaced with [VERIFY: what to check]. Keep the author's voice — do not rewrite style, only facts. ## 4. Gaps What the sources do NOT cover that the draft implies they do. What research is still needed before this is publishable.

#TASK Convert the following meeting transcript or rough notes into a clean action item list with owners, due dates, and enough context that anyone can execute without re-reading the transcript. #REQUEST Do not summarize the meeting. Extract only: decisions made, actions assigned, open questions, and blockers. Discard discussion, agreement sounds, and tangents. #ACTION ## 1. Decisions Made Each decision as: - **Decision:** What was decided - **Context:** One sentence on why - **Owner:** Who owns implementation - **Deadline:** When it ships ## 2. Action Items Each action as: - **Action:** Imperative verb + specific deliverable - **Owner:** Name (not "team" or "engineering") - **Due:** Date - **Context:** One sentence — why this matters, any dependencies - **Status:** Not started / In progress / Blocked (with blocker) ## 3. Open Questions Questions raised but not resolved: - **Question:** Exact wording - **Owner:** Who will resolve - **Needed by:** Date ## 4. Blockers Anything stopping progress: - **Blocker:** What is stuck - **Impact:** What it delays - **Owner:** Who unblocks - **Resolution path:** Next step ## 5. Next Meeting Prep - Agenda items carried forward - Metrics or data to bring - Decisions needed #FORMAT Clean markdown. Action items table at top for quick scanning. Full context below. #CONTEXT Meeting type: {{meeting_type}} (standup / planning / retro / 1:1 / board / client) Attendees: {{attendees}} Date: {{date}} Project/Team: {{project}} #TRANSCRIPT OR NOTES {{paste_transcript_or_notes}}

#SITUATION You need a defensible market size for a pitch deck, board meeting, or internal strategy. Guesswork destroys credibility. You need a calculation where every number traces to a source or a stated assumption. #PURPOSE Produce a TAM/SAM/SOM analysis with explicit logic, cited sources, and a sensitivity table so the reader sees how the numbers move when assumptions change. #EXPECTED OUTPUT ## 1. Definitions - TAM: Total addressable market — everyone who could theoretically buy - SAM: Serviceable available market — the segment you can actually reach - SOM: Serviceable obtainable market — what you can realistically capture in 3-5 years ## 2. Bottom-Up Calculation (Primary) - Unit economics: price × quantity logic - Customer segments with count estimates and sources - Penetration rate assumptions with justification ## 3. Top-Down Validation (Secondary) - Industry report totals (name the report, year, page) - Your segment as % of total - Reconcile with bottom-up — if they diverge >3x, explain why ## 4. Sensitivity Table | Scenario | TAM | SAM | SOM | Key Driver | | Conservative | $X | $Y | $Z | Low penetration | | Base | $X | $Y | $Z | Expected | | Aggressive | $X | $Y | $Z | High penetration | ## 5. Source Log Every number cited with: source name, publication date, URL if public, or "assumption" with rationale. #CONTEXT Product: {{product_desc}} Target customer: {{target_customer}} Pricing model: {{pricing}} (per seat, usage, tiered) Geography: {{geography}} Known data sources you have: {{sources_or_none}}

#TASK Tear down a competitor's product feature-by-feature and identify where they are strong, where they are weak, and where there is a gap we can win. #REQUEST Produce a structured competitive analysis, not a feature checklist. I want to know what to build, not just what they have. #ACTION Structure the output as: 1. **Competitor snapshot** — 2-3 sentences on who they are, their positioning, and their apparent target customer. 2. **Feature matrix** — list their core features. For each: what it does, how well it works (Strong/Adequate/Weak), and one specific observation (not 'good UX' but 'onboarding completed in 4 clicks with no email verification'). 3. **What they do better than us** — be honest. Name the specific advantage. 4. **Where they are vulnerable** — the gaps, rough edges, or underserved segments. 5. **Opportunities** — 3 concrete moves we could make that they are not covering. Each tied to a specific weakness above. #CONTEXT Our product: {{one-line description of what we build and who it's for}} Competitor: {{competitor name and URL}} What I know already: {{any prior research, customer feedback, or assumptions — or 'nothing yet'}} #EXAMPLE Good observation: 'Pricing page shows only annual plans on mobile — monthly is hidden behind a toggle. Likely losing price-sensitive mobile traffic.' Bad observation: 'Their pricing seems competitive.' Competitor details to analyze: {{paste competitor's feature list, pricing page content, or any notes you've gathered}}

# TASK Synthesize the provided sources into a structured literature review that a researcher or decision-maker can use to understand the current state of a topic. Topic: {{TOPIC}} Sources to synthesize: {{SOURCES}} # REQUEST Do not summarize each source one by one. Instead, identify the key themes, debates, and consensus points that emerge across sources. A good literature review tells a story about what the field collectively knows and where the gaps are. # ACTION ## Part 1: Thematic Synthesis Organize findings into 3-5 themes. For each theme: - Name the theme (specific, not generic — "Adoption barriers in enterprise vs. SMB" not "Challenges") - Which sources support this theme? (cite by [S1], [S2] etc.) - Where do sources agree? - Where do they disagree or contradict? (This is the most valuable part — surface tensions explicitly.) - What is the strength of evidence? (well-established, emerging, single-study, anecdotal) ## Part 2: Gap Analysis - What questions are sources NOT addressing that seem important? - Where is the evidence weakest? - What methodological limitations appear across the body of work? ## Part 3: Consensus Map One paragraph summarizing what the field collectively agrees on (if anything). Be honest — if there is no consensus, say so and explain what the main camps are. ## Part 4: Source Quality Notes For each source, one line on credibility: peer-reviewed? industry report? blog post? What's the potential bias? # CONTEXT - Number each source [S1], [S2], etc. as provided in the input - If sources conflict, do not resolve it for the reader — present both sides with the evidence each offers - Flag any claim that relies on a single source # EXAMPLE Input style: "[S1] Smith et al. (2024) found that 73% of enterprise teams adopted... [S2] However, a Gartner report (2025) showed..." Output theme example: **Theme: Adoption rates vary sharply by company size** - Enterprise adoption rates are consistently reported at 60-80% [S1, S3], while SMB rates lag at 20-30% [S2]. - Contradiction: [S1] cites cost as the primary barrier, while [S4] argues integration complexity matters more. - Evidence strength: Moderate. Multiple sources but reliance on survey data.

#ROLE You are an interview coach who has prepared executives, job candidates, and podcast guests for high-stakes conversations. #INSTRUCTIONS Build a complete interview preparation brief for the user. The brief must cover what they will likely be asked, what they should say, and what they should avoid saying. #STEPS ## Step 1: Anticipated Questions Generate 10-15 questions the interviewer is likely to ask, organized into three tiers: - **Tier 1 (Certain)**: Warm-up and background questions. Nearly guaranteed. - **Tier 2 (Likely)**: Role-specific, topic-specific, or behavioral questions. - **Tier 3 (Curveballs)**: Challenging, unexpected, or pressure questions designed to test composure. ## Step 2: Talking Points For each Tier 1 and Tier 2 question, provide: - A recommended answer structure (not a script — a framework like 'Problem → Action → Result' or 'Context → Choice → Outcome') - 2-3 key points to land - A specific example or story to use (from the user's background) - Target length (how long the answer should run) ## Step 3: Red-Flag Responses List 5 things to avoid saying: - Phrases that sound evasive or rehearsed - Topics that could derail the conversation - Over-sharing or under-sharing patterns - How to pivot if you start going down a wrong path ## Step 4: Questions to Ask Provide 5 questions the user should ask the interviewer, ranked by how much signal they provide about whether this is the right fit. ## Step 5: Logistics Checklist - What to prepare 24 hours before (research, materials, tech check) - First 60 seconds strategy (how to open strong) - Closing strategy (how to end on a memorable note) #END GOAL The user walks into the interview knowing what to expect, what to say, and how to handle surprises. #NARROWING - Keep answer frameworks flexible, not memorized scripts. The goal is preparedness, not robotic delivery. - Tailor everything to the specific interview type and context below. #CONTEXT - Interview type: {{interview_type}} (job interview, media appearance, podcast, panel, investor pitch) - Role or topic: {{role_or_topic}} - Company or show: {{company_or_show}} - User's background: {{user_background}} - Format: {{format}} (in-person, video, phone) and duration: {{duration}}

#SITUATION A stakeholder needs to understand what a dataset is telling them, but they do not have time to read raw numbers or build charts. They need the story the data tells, written in plain language. #PURPOSE Transform the provided data into a structured findings report that a non-technical decision-maker can act on in under five minutes. #EXPECTED OUTPUT Produce a report with these sections: ## Executive Summary (3-4 sentences) The single most important finding and why it matters. No jargon. ## Key Findings (5-7 bullet points) Each finding states the observation, the supporting number, and the implication. Example format: 'Revenue from product X dropped 23% month-over-month — investigate whether the price increase on June 1 is the cause.' ## Anomalies and Outliers List anything unusual: unexpected spikes, missing data, values that break a pattern. Flag whether each is worth investigating or likely a data quality issue. ## Trends Identify direction and velocity. Is the metric going up, down, or flat? How fast? Compare to the prior period if data allows. ## Recommendations (3-5 items) Specific, concrete next steps tied to the findings. Each recommendation references the finding it addresses. #CONTEXT - Data type: {{data_type}} (CSV / JSON / spreadsheet / pasted table) - Business context: {{business_context}} - Time period covered: {{time_period}} - Prior period for comparison: {{prior_period}} #STYLE - Plain English. Write for someone who skipped the meeting but needs to make the call. - Every claim must reference a number from the data. No unsupported assertions. - If the data is too sparse to support a finding, say so explicitly rather than guessing. #DATA {{raw_data}}

#ROLE: UX researcher who conducts discovery interviews. #TASK: Generate a user interview question guide for researching {{topic}}. #INTERVIEW STRUCTURE (45 minutes): ## 1. WARM-UP (5 min) - Background questions to build rapport - General context about their relationship to {{topic}} ## 2. CURRENT BEHAVIOR (10 min) - Walk me through how you currently {{activity}}... - What tools do you use? - What does a typical day look like regarding {{topic}}? ## 3. PAIN POINTS (15 min) - What's the hardest part about {{activity}}? - Can you tell me about a time when {{scenario}} went wrong? - What do you do AFTER {{activity}}? ## 4. WORKAROUNDS (10 min) - Have you tried to solve {{problem}} yourself? - What would your ideal solution look like? ## 5. WRAP-UP (5 min) - Is there anything I should have asked but didn't? - Who else should I talk to? #GUIDELINES: - Use open-ended questions (no yes/no) - Ask about PAST behavior, not future intentions - Follow-up probes for each main question

# ROLE You are a fact-checking auditor who specializes in finding hallucinations, unsupported claims, and grounding failures in LLM outputs. You treat every sentence as a claim that must be traced to a source. You do not rubber-stamp. You do not speculate. # CONTEXT LLMs hallucinate. They invent citations, fabricate statistics, misattribute quotes, and state assumptions as fact. In 2026 this problem compounds: agents chain multiple LLM calls, RAG systems retrieve imperfect context, and a single ungrounded claim can propagate through an entire workflow. A response that sounds confident is not a response that is correct. Grounding means every factual claim in the output can be traced to the provided source material. If the source does not support the claim, it fails — regardless of whether the claim is true in the real world. The auditor's job is narrow and strict: check the text against the sources, not against general knowledge. # OBJECTIVE Given an LLM-generated response and its source material, produce a structured audit that: 1. Extracts every factual claim from the response 2. Checks each claim against the source material 3. Flags hallucinations, unsupported claims, contradictions, and citation errors 4. Assigns severity (critical / major / minor) 5. Ranks fixes by impact # STYLE Output as a clean markdown report with tables and short verdicts. No filler. No hedging. Every flag must cite the specific claim and the specific source gap. # TONE Clinical and direct. Write like a QA reviewer grading a deliverable — specific, unsentimental, no hand-holding. A claim either has grounding or it does not. # AUDIENCE A developer, researcher, or content lead who needs to know whether an LLM output can be trusted, shipped, or published without manual verification of every line. # RESPONSE FORMAT ## Audit Summary - Total claims checked: [N] - Grounded: [N] - Unsupported: [N] - Contradicted: [N] - Citation errors: [N] - Overall verdict: [PASS / PASS WITH FIXES / FAIL] ## Claim-Level Audit | # | Claim (quoted from response) | Verdict | Source evidence or gap | Severity | |---|------|---------|------------------------|----------| | 1 | "[exact quote]" | Grounded | [source ref + quote] | — | | 2 | "[exact quote]" | Unsupported | [no matching source found] | Major | | 3 | "[exact quote]" | Contradicted | [source says X, claim says Y] | Critical | | 4 | "[exact quote]" | Citation error | [cited source does not contain claim] | Major | ## Hallucination Types Found List each type with the claim number: - Fabricated statistic: [claim #] - Invented citation: [claim #] - Misattributed quote: [claim #] - Contradicts source: [claim #] - Unsupported inference presented as fact: [claim #] ## Fixes Ranked by Severity 1. [Critical] [claim #]: [what to change or remove] 2. [Major] [claim #]: [what to change or add source for] 3. [Minor] [claim #]: [suggested rewording] ## Output Decision - Safe to ship as-is: [Yes / No] - Ships after critical fixes: [Yes / No] - Requires human review before any use: [Yes / No] --- # INPUT Paste the LLM response below the RESPONSE heading and the source material below the SOURCES heading. ## RESPONSE [ Paste the LLM-generated text to audit here ] ## SOURCES [ Paste the source documents, retrieved context, or reference material here. One source per block, labeled Source A, Source B, etc. ] ## AUDIT INSTRUCTIONS - Extract every factual claim. A claim is any statement that asserts a fact, statistic, name, date, quote, cause-effect relationship, or source attribution. - For each claim, search the provided sources for supporting evidence. - A claim is Grounded only if the source explicitly contains the same information. Paraphrase is acceptable if the meaning matches. Implication is not grounding. - A claim is Unsupported if no source addresses it — even if you believe it is true from general knowledge. - A claim is Contradicted if a source states the opposite. - A claim has a Citation Error if it references a source that does not contain the cited information. - Do not use your own training knowledge to validate claims. Only the provided sources count. - Severity guide: Critical = contradicts source or fabricates a citation. Major = unsupported factual claim presented as fact. Minor = unsupported minor detail or imprecise wording.

#ROLE: Business strategy consultant. #TASK: Conduct a comprehensive SWOT analysis for {{subject}}. #FRAMEWORK: ## STRENGTHS (Internal, Positive) - What advantages do they have? - What do they do better than anyone? - What unique resources can they draw on? ## WEAKNESSES (Internal, Negative) - What could they improve? - What should they avoid? - What factors lose them opportunities? ## OPPORTUNITIES (External, Positive) - What trends could they capitalize on? - What market gaps exist? - What partnerships/alliances are possible? ## THREATS (External, Negative) - What obstacles do they face? - What are competitors doing? - What external factors could cause problems? #STRATEGIC RECOMMENDATIONS: 1. **S-O Strategy**: Use strengths to maximize opportunities 2. **W-O Strategy**: Improve weaknesses to capture opportunities 3. **S-T Strategy**: Use strengths to minimize threats 4. **W-T Strategy**: Minimize weaknesses to avoid threats #SUBJECT: {{subject}} #INDUSTRY: {{industry}}

#ROLE: Customer insights analyst. #TASK: Analyze the following {{num_reviews}} product reviews and extract actionable insights. #ANALYSIS: 1. **Overall Sentiment**: What % are positive, neutral, negative? 2. **Top 5 Praises**: What do happy customers love most? (rank by frequency) 3. **Top 5 Complaints**: What frustrates unhappy customers? (rank by severity) 4. **Feature Requests**: What do customers wish existed? 5. **Customer Segments**: Can you identify distinct user types from the reviews? 6. **Competitor Mentions**: Do any reviews mention alternatives? 7. **Action Items**: 5 specific recommendations for the product team #OUTPUT: Structured report with data tables and bullet points. #REVIEWS: {{reviews}}

#ROLE: Business strategy consultant specializing in competitive intelligence. #CONTEXT: Analyze the competitive landscape for: {{company_or_product}} #FRAMEWORK: Analyze each competitor across: Product/Service, Pricing, Target Market, Strengths, Weaknesses, Strategy. #OUTPUT: 1. **Competitor Matrix** (comparison table) 2. **SWOT** for top 3 competitors 3. **White Space Opportunities**: Gaps no one serves 4. **Strategic Recommendations**: 3 actionable moves #COMPETITORS: {{competitor_list}}

#ROLE: Research analyst skilled at synthesizing multiple sources. #CONTEXT: I will provide {{num_sources}} sources. Synthesize them into a coherent overview. #TASK: 1. **Key Findings**: 5-7 bullet points of collective insights 2. **Areas of Consensus**: Where sources agree 3. **Areas of Conflict**: Where they disagree 4. **Methodology Notes**: Strength of evidence 5. **Gaps**: Unanswered questions 6. **Implications**: Why this matters for {{field}} #OUTPUT: Structured Markdown. Cite source numbers. #SOURCES: {{sources}}