# Firecrawl: verified indicators and explanatory demos

Verified live on 2026-10-20 at https://www.firecrawl.dev/ using Chromium Playwright 1440×1100; downloaded current client bundles. No Dealyo source files edited. Calibration files VOICE.md, BRAND.md, CLIENTS.md were not found by rg --files.

## Evidence and prioritized Dealyo proposals

| Priority | Actual Firecrawl pattern | Locator / evidence | Behavior verified | Proposal for Dealyo |
|---|---|---|---|---|
| P1 | Use-case selector + changing demo + matching customer story | section containing heading “Transform web data into AI-powered solutions”; buttons “Deep research”, “Smarter AI chats”, “AI agent tools”, “Onboarding”, “Lead enrichment”; IDs #ai-research, #ai-chats, #ai-skills, #ai-onboarding, #ai-leads | Clicking each button switches the visible demo, highlighted selection, and customer proof below (Aemon/Sierra/Lovable/Gamma/Cognism). | Select goal: more inquiries / faster replies / AI-ready website. Show a concrete corresponding workflow and deliverable with matching real client evidence when available. |
| P1 | Website input turns into extracted business assets | #ai-onboarding; screenshot tab-Onboarding.png; bundle 1-sr6ptlg1ni5.js | Animated typed stripe.com → “Analyzing website” → “Website analyzed” → rows with Theme, Summary, Logo, Assets. Source uses fixed data and setInterval; the website box is a div, not a real input. | Show “Twój adres strony → co sprawdzamy → co otrzymujesz”: offer clarity, contact path, information AI can read, prioritized recommendations. Mark sample explicitly. This is an explanatory mockup unless connected to an actual scan. |
| P1 | Before/after content reduction with live changing count | Performance section “Token-efficient”; screenshot performance.png; bundle 1-sr6ptlg1ni5.js module 620042 | Fixed 38,381 starting tokens; every 500ms removes nav/ads/footer from a list and updates counter to 2,788, then cycles page path. Fixed token removals 9,840 + 16,431 + 9,322. | Before/after: unstructured offer → clear offer usable by customers and AI; or manual repetitive tasks → automated handoff. Numeric savings only from a real baseline or a plainly labeled calculator assumption. |
| P1 | Multi-stage status label in a loading frame | .features-wait; screenshot capabilities.png; bundle 1-sr6ptlg1ni5.js | Fixed states “Request Sent” → “Page is loading...” → “Waits for key elements...” → “All content is present” → “Clean & complete data”, timers 2,500ms when visible; repeats through browser cards. | Explain an otherwise invisible service: receiving materials → diagnosis → priorities → implementation → checking. Clear current and completed states. Do not make fixed-time demo look like a real job in progress. |
| P2 | Category counters alongside incoming research cards | #ai-research; screenshots usecases.png and tab-Deep-research.png; bundle 2atuhb7om6qc7.js | Counters animate from zero, typed quantum-computing query, result cards appear. Exact source targets: academic papers 247; news 1,832; expert opinions 89; reports 156; industry data 423. | Audits: number of issues grouped by offer/contact/trust/technical readiness, alongside actual examples. Counts must come from the audit; sample mode labeled. |
| P2 | Enrichment summary + progressively filled contact cards | #ai-leads; screenshot tab-Lead-enrichment.png; bundle 2atuhb7om6qc7.js | Fixed summary array 1,243 tech startups; 892 with contact info; 3,421 decision makers; Series A+; 647 ready to engage. Contact cards resolve from scramble animation. These numbers are demo fixtures. | Show how a company record becomes usable: company → fit → decision-maker role → reason to contact → next action. This is not a valid funnel as-is: decision-makers exceed companies and units differ. |
| P2 | 3-node request/data path | “Live web data” card; .features-cached-inner-user, .features-cached-inner-firecrawl, .features-cached-inner-cache-or-web; screenshot capabilities.png | User / Firecrawl / index-or-web nodes connected by lines; source creates animated gradients along separators. | Inquiry → collection/qualification → owner and next step; or client materials → processing → ready output. Keep labels understandable for service customers. |
| P2 | Side-by-side reliability bars with benchmark link | Performance section “Reliable on any page”; [data-testid="reliability-bar-label"] (three instances); screenshot performance.png | 96% Firecrawl, 79% Puppeteer, 75% cURL; copy identifies 1,000-URL benchmark, date Jan 13 2026; links to comparisons. | Comparable before/after measures with definition, denominator, measurement period and source. Prefer audit finding categories over an invented universal “AI readiness score”. |
| P3 | Streaming latency table as activity visualization | Performance section “Speed that feels invisible”; screenshot performance.png; details.json performance.samples | Rows enter with URLs, Crawl/Scrape columns, changing timings. Initial rows ~49–52ms; incoming rows observed at 400–795ms. Caption separately claims benchmark P95 3,387ms. | Activity log can illustrate system work, but distinguish it from measured business gains. Do not copy the misleading impression that decorative timings prove the benchmark. |
| P3 | Capability grid inside browser frame | “Interact with pages” card in capabilities.png; items Navigate, Click, Type, Wait, Scroll, Press, Screenshot, Scrape | Actual capability list, not a demonstrated chronological business pipeline; source includes explanatory descriptions. Browser getByRole('button') did not expose these items. | Grid of concrete actions offered by an automation; use a true ordered pipeline only when there is a real sequence. |

## Controls also confirmed

Developer First has Search/Scrape/Interact buttons plus Python/Node.js/cURL/CLI tabs; switching product changes the headline, code, and matching JSON/Markdown output, while language selection changes input code. Good reusable input/output teaching pattern for an AI service, but technical code is inappropriate for Dealyo client-facing copy.

Agent Ready has Prompt/MCP/CLI tabs and copy controls; this is setup-mode selection, not a conversion/results indicator.

Alexandria section has four labeled counts (indexed sources/providers/capabilities/categories). These are homepage claims of inventory, not customer results or animated service execution.

## Source references

- Public page: https://www.firecrawl.dev/
- Main demo bundle: https://www.firecrawl.dev/_next/static/chunks/1-sr6ptlg1ni5.js?dpl=dpl_2B1YYmsKFkGJzkbk44qpNDUukqZr
- Research and lead fixtures: https://www.firecrawl.dev/_next/static/chunks/2atuhb7om6qc7.js?dpl=dpl_2B1YYmsKFkGJzkbk44qpNDUukqZr
- Captured DOM/text: body.html, body.txt, details.json in this folder.
- Screenshots: full.png, features.png, performance.png, capabilities.png, usecases.png, tab-Deep-research.png, tab-Smarter-AI-chats.png, tab-AI-agent-tools.png, tab-Onboarding.png, tab-Lead-enrichment.png.

## Constraints for a reusable checklist

- Separate illustration, calculator estimate, audited fact, and actual client outcome.
- Every indicator needs a label and units; percentages need a defined denominator; business outcomes need a period and baseline.
- Every stateful illustration needs a static final-state alternative and reduced-motion support (proposals; source accessibility behavior not fully audited).
- Use one clear explanatory change per card: input/output, stage, comparison, count, or path.
- Do not present randomized/scrambled/time-driven text as a live service run.
- Sharp boxes and #FE4A23 are Dealyo's established instructions; Firecrawl's rounded bars/cards are reference-specific and should not override them.
