# Training Deck Company Research Prompt For the more detailed profile-page research workflow and machine-readable schema, use `company_profile_research_blueprint.md` and `company_profile_research_schema.csv` alongside this prompt. Use this prompt for every training deck/company. Replace bracketed fields before running. ```text You are building a company-specific research record for a pitch deck feedback and comparable-slide product. Inputs: - Company name: [COMPANY_NAME] - Deck file path: [DECK_FILE_PATH] - Known deck stage, if any: [PRE-SEED / SEED / SERIES A / UNKNOWN] - Known sector/category, if any: [CATEGORY] - Optional website: [WEBSITE] - Optional Crunchbase URL: [CRUNCHBASE_URL] - Output folder: [OUTPUT_FOLDER] Primary objective: Create a source-backed company intelligence record that connects the deck's slide claims to real company context, pre-seed/seed financing details, investors, later outcomes, and slide-level patterns that can be used for pitch deck feedback and comparable-slide recommendations. Important rules: - Do not invent data. If a field is unavailable, write "unknown" and explain what source would be needed. - Preserve source labels exactly, especially funding stage labels from Crunchbase/PitchBook/press. Also add our internal taxonomy label when the source label is misleading. - Separate deck-time facts from post-deck outcomes. Do not let later success rewrite what was actually shown in the deck. - Treat deck metrics as company-claimed unless independently verified. - For Crunchbase, use the user's logged-in browser session if available, but never ask for or handle passwords. If blocked, record the gap and use public sources. - Resolve company-name collisions by verifying founder, website, logo, headquarters, sector, and deck contents. - Use exact dates whenever available. - Every major claim must have a source URL or be marked as coming from the supplied deck. Research workflow: 1. Deck extraction - Render the deck to slide images. - Extract text from each slide. - Create a contact sheet image. - Count slides and identify deck stage, round ask, date clues, company positioning, and major metrics. 2. Company identity - Verify official company name, legal name if available, website, headquarters, founding year, founder(s), category, business model, customer type, and current operating status. - Record discrepancies across sources, for example formation date vs public launch date. 3. Funding history with emphasis on pre-seed and seed - Capture all funding rounds, but deeply research pre-seed, seed, and the first later institutional round. - For each round capture: - Source round label - Internal stage label - Announced date - Amount - Valuation, if available - Lead investor - Participating investors - Partner names - Investor count - Source URL - Confidence level - If a database labels multiple early rounds as "Seed," infer internal taxonomy using timing, amount, launch status, investor type, and later institutional seed, but preserve the original label. - Compare deck ask versus actual round closed. 4. Investor intelligence - Build an investor table for all pre-seed and seed investors. - Classify each investor as fund, angel, operator angel, strategic, accelerator, venture debt, syndicate, family office, or unknown. - Identify operator/founder angels and why they matter. - Mark repeat investors who participated in later rounds. - Note relevant consumer/category expertise and notable comparable portfolio companies where available. 5. Traction and business model - Extract deck-time traction: revenue, GMV, units sold, users, waitlist, downloads, subscribers, retention, reviews, NPS, CAC, ROAS, gross margin, contribution margin, payback, partnerships, pilots, retail doors, marketplace performance, press, community, and pipeline. - Verify post-deck traction separately from public sources. - For physical consumer products, prioritize supply chain, payment terms, inventory, retail/wholesale margin, and working capital. - For marketplaces, prioritize supply, demand, liquidity, repeat use, take rate, and geographic density. - For consumer apps/social, prioritize retention, DAU/MAU, virality, community loops, and creator/user acquisition. 6. Market and category context - Identify the category narrative at deck time: why now, incumbents, consumer behavior shift, platform shift, regulation, cultural shift, distribution shift, or cost shift. - Capture competitors named in the deck and competitors from current sources. - Note whether the company framed a wedge, a platform, a category creation, or a better-product substitution. 7. Later outcome signal - Determine whether the company raised after the deck, how much, when, from whom, and whether seed investors followed on. - Record acquisitions, IPO, shutdown, major retail expansion, major product expansion, lawsuits/regulatory issues, and other outcome signals. - Build a dated product, channel, financing, legal, and brand milestone timeline. Mark company-authored scale claims separately from independent reporting. - Mark whether the company is a strong, medium, or weak comparable for "went on to raise successfully." 8. Competitive and product diligence - Build a current competitive-context table with direct peers, indirect alternatives, legacy incumbents, and litigation/regulatory competitors where relevant. - Separate deck-time competitor framing from current competitive reality. - Capture third-party product reviews, category critiques, customer experience issues, and performance/durability tradeoffs. - For consumer brands, distinguish brand heat, product quality, retention, and repeat-purchase evidence. 9. Slide-level intelligence - For every slide, produce a row with: - slide_number - slide_title - detected_slide_type - primary_claim_or_function - key_data_points - investor_question_answered - investor_objection_reduced - why_compelling_for_feedback_product - weakness_or_followup_question - comparable_tags - Identify the 5-12 most compelling slides to show future founders and explain exactly what pattern they should learn from. - Do not just label slides "traction" or "market." Capture the useful pattern, such as "named retail pipeline plus revenue math" or "pre-launch owned audience as acquisition moat." 10. Risk and claim quality - Capture risks visible in the deck and later public record: - unsubstantiated health/safety claims - regulatory risk - litigation - gross margin/inventory risk - paid CAC dependence - retention gaps - competitive response - retailer concentration - marketplace liquidity risk - scientific or product-performance disputes - Mark claim strength as proven, partially supported, deck-claimed, contradicted, or unknown. Required outputs: 1. [SLUG]_research.md - Company snapshot - Founder/team signals - Funding history - Pre-seed and seed investor details - Deck metadata - Deck narrative - Traction and economics - Market/category context - Later outcome signal - Most compelling slides - Risk/claim-quality notes - Comparable matching tags - Open data gaps - Source list 2. [SLUG]_slides.csv Required columns: slide_number, slide_title, slide_type, primary_claim_or_function, key_data_points, investor_question_answered, investor_objection_reduced, why_compelling_for_feedback_product, weakness_or_followup_question, tags 3. [SLUG]_investors.csv Required columns: company, round_internal_stage, round_source_label, round_date, round_amount, investor, investor_type, lead_status, partner, source, notes 4. [SLUG]_sources.csv Required columns: source_name, url, source_type, accessed_date, facts_used, reliability_notes 5. [SLUG]_timeline.csv Required columns: date, event_type, event, details, source, relevance_to_training_dataset 6. [SLUG]_competitive_context.csv Required columns: category, company_or_group, positioning_vs_[SLUG], source, training_dataset_note 7. [SLUG]_contact_sheet.jpg Contact sheet of rendered slides for quick visual review. 8. Brand profile page and graphics - Create `/output/profiles/[SLUG]/index.html` - Create `/output/profiles/[SLUG]/reference.html` - Create `/output/profiles/assets/[SLUG]-funding.svg` - The profile page should include: - company snapshot - funding graph with years on the x-axis and dollar amount on the y-axis - funding timeline - product/channel/outcome timeline - investor mix - outcome signals - competitive context - most useful comparable slide patterns - contact sheet - links to research memo, slides CSV, investors CSV, timeline CSV, competitive CSV, and sources CSV - The detailed reference page should include: - full company identity table - full funding-round table - complete pre-seed/seed investor table - deck metadata and deck-time traction claims - narrative synthesis - slide intelligence table - product/channel/outcome timeline - competitive and product-review context - later outcomes - risks and claim-quality notes - open data gaps - source audit with links - If a funding round amount is undisclosed, show it as a marker/note rather than converting it to $0. Quality bar: - The final research record should let the product answer: - What kind of company is this? - What round was the deck likely used for? - What did the company claim at deck time? - What actually happened afterward? - Which slides are worth recommending to other founders, and why? - Which investors participated in pre-seed/seed, and what signal did they add? - What data is missing or uncertain? ``` ## Caraway Calibration Example When applying this prompt, Caraway is the calibration case: - Preserve Crunchbase's "Seed" label for both March 2019 and May 2020 rounds, but internally classify March 2019 as pre-seed-equivalent and May 2020 as seed. - Store the deck's $2M seed ask separately from the announced $5.3M seed close. - Treat deck claims like $410K revenue, 4:1 ROAS, and 160K email subscribers as deck-claimed unless independently verified. - Mark slides 10, 18, and 29 as "omnichannel distribution proof," not generic GTM. - Mark slide 20 as "working-capital/supply-chain advantage," a high-value physical product pattern. - Record post-deck validation: $35M 2022 Series A/investment, broad retail rollout, and product expansion. - Record post-deck scale claims separately from independent reporting: 2.5M+ customers, 200K+ five-star reviews, and 400+ patents are company-claimed in 2026 PR. - Record post-deck risk: 2025 NAD advertising decision and 2026 PFAS advertising/legal dispute. - Record product-performance diligence: third-party reviews praise aesthetics but flag ceramic nonstick durability, care requirements, staining, weight, and hot lid handles.