Apple Financial PDF → Reducto Structured Extract

Purpose

Download Apple's quarterly Condensed Consolidated Financial Statements PDF (the financial-data attachment that accompanies every "Apple reports {ordinal} quarter results" press release on apple.com/newsroom), POST it to Reducto's /extract endpoint with a financial JSON Schema, and return structured data: total net sales, segment-by-segment revenue (Americas / Europe / Greater China / Japan / Rest of Asia Pacific) and product-line revenue (iPhone / Mac / iPad / Wearables Home & Accessories / Services), basic & diluted EPS, and balance-sheet line items (cash, marketable securities, receivables, total assets, total liabilities, shareholders' equity, term debt, etc.). Read-only — fetches public PDF URLs and posts to Reducto's API; never logs in, never modifies anything.

When to Use

• Earnings analysis pipelines that need clean structured numbers within minutes of an Apple earnings release.
• Building a historical dataset of Apple revenue / segment / EPS across multiple quarters or years.
• Any flow where you'd otherwise hand-key Apple's income statement, segment breakdown, or balance sheet from a press release PDF.
• A reference template for "scrape financial PDF off a corporate IR site → Reducto" for other large-cap public companies (the URL-pattern + press-release discovery technique generalizes; the JSON schema does not).

Workflow

The entire skill runs over HTTP — no browser session needed in the happy path. Apple publishes the financial-statements PDF at a public, unauthenticated, predictable CDN URL on www.apple.com/newsroom/pdfs/..., and Reducto's /extract endpoint accepts a public URL as input (no upload step required). Total time per quarter: ~10–30 s wall, ~3–10 Reducto credits depending on settings.deep_extract and settings.citations.enabled.

Important framing: the prompt mentions "10-Q, 10-K" but Apple does not post the full 10-Q or 10-K as a PDF on apple.com — those are filed with the SEC as .htm documents (see Gotchas). What Apple does post is the Condensed Consolidated Financial Statements PDF — three pages containing the income statement (+ segment + product breakdowns), the balance sheet, and the cash-flow statement. That PDF contains every field this skill is asked to extract (revenue, segments, EPS, balance sheet) and is the right input.

1. Resolve the PDF URL for the target quarter

Apple's URL pattern is:

https://www.apple.com/newsroom/pdfs/{stem}/FY{YY}_Q{Q}_Consolidated_Financial_Statements.pdf

where the {stem} directory naming changed between FY25 and FY26:

The filename portion (FY25_Q2_Consolidated_Financial_Statements.pdf) has been stable across all observed quarters.

Recommended URL-resolution algorithm — robust against the directory-name change above and against any future renames:

1. Try both URL variants in parallel. Whichever returns 200 application/pdf is the canonical URL:

   https://www.apple.com/newsroom/pdfs/fy{YYYY}q{Q}/FY{YY}_Q{Q}_Consolidated_Financial_Statements.pdf
   https://www.apple.com/newsroom/pdfs/fy{YYYY}-q{Q}/FY{YY}_Q{Q}_Consolidated_Financial_Statements.pdf
   

2. If both 404 (Apple changed the pattern again), fall back to press-release discovery:
   1. Use Browserbase Search (or any web search engine): apple reports {ordinal} quarter results fy{YY} site:apple.com — returns the canonical press-release post on apple.com/newsroom/{YYYY}/{MM}/apple-reports-{ordinal}-quarter-results/.
   2. browse cloud fetch that press-release URL, regex href="(/newsroom/pdfs/[^"]+\.pdf)" out of the HTML — every Apple quarterly-results press release embeds exactly one PDF link to the Consolidated Financial Statements.

Fiscal-year ↔ calendar mapping (Apple's FY ends in late September):

Press-release months drift ±1 month year-over-year. Don't hardcode the month — search.

2. Fetch the PDF (optional — only if uploading vs. passing the URL)

Reducto accepts a public URL directly as the input field on /extract, so step 2 is usually a no-op. Only download if you need a local copy for caching, or if you're embedding the PDF in a different downstream pipeline:

# Direct fetch (when network policy allows direct egress)
curl -sS -o apple-fy25-q2.pdf \
  "https://www.apple.com/newsroom/pdfs/fy2025-q2/FY25_Q2_Consolidated_Financial_Statements.pdf"

# Or via Browserbase Fetch (returns the binary as base64 in a JSON envelope)
browse cloud fetch \
  "https://www.apple.com/newsroom/pdfs/fy2025-q2/FY25_Q2_Consolidated_Financial_Statements.pdf" \
  > envelope.json
node -e "
  const fs=require('fs');
  const raw=fs.readFileSync('envelope.json','utf8');
  const env=JSON.parse(raw.slice(raw.indexOf('{')));
  fs.writeFileSync('apple-fy25-q2.pdf', Buffer.from(env.content,'base64'));
"

Observed PDF sizes (Apple's Q4 PDFs are ~60% larger than Q1–Q3 because Q4 also folds in full-year data):

3. POST to Reducto /extract with a financial JSON Schema

Reducto's /extract runs Parse internally and applies an LLM pass to populate a JSON Schema you supply. Pass the public URL directly as input:

curl -X POST https://platform.reducto.ai/extract \
  -H "Authorization: Bearer $REDUCTO_API_KEY" \
  -H "Content-Type: application/json" \
  -d @extract-request.json

extract-request.json:

{
  "input": "https://www.apple.com/newsroom/pdfs/fy2025-q2/FY25_Q2_Consolidated_Financial_Statements.pdf",
  "instructions": {
    "system_prompt": "This is Apple Inc.'s Condensed Consolidated Financial Statements PDF (3 pages: Statement of Operations with segment and product breakdowns; Balance Sheet; Statement of Cash Flows). All monetary values are in millions of USD unless the table says 'thousands' (shares). Extract the values for the most-recent (leftmost) Three-Months-Ended column as the primary report; include the prior-year comparison column and the six-months/nine-months columns when present.",
    "schema": {
      "type": "object",
      "properties": {
        "fiscal_period": { "type": "string", "description": "e.g. 'FY25 Q2' — read from the column headers / page footer" },
        "period_end_date": { "type": "string", "description": "ISO date of the most-recent quarter end (e.g. '2025-03-29')" },
        "currency": { "type": "string", "description": "ISO 4217 (always 'USD' for Apple)" },
        "units": { "type": "string", "description": "Should be 'millions' for all monetary values per the table headers" },
        "income_statement": {
          "type": "object",
          "properties": {
            "products_revenue": { "type": "number" },
            "services_revenue": { "type": "number" },
            "total_net_sales": { "type": "number" },
            "products_cost_of_sales": { "type": "number" },
            "services_cost_of_sales": { "type": "number" },
            "total_cost_of_sales": { "type": "number" },
            "gross_margin": { "type": "number" },
            "research_and_development": { "type": "number" },
            "selling_general_administrative": { "type": "number" },
            "total_operating_expenses": { "type": "number" },
            "operating_income": { "type": "number" },
            "other_income_expense_net": { "type": "number", "description": "May be negative — preserve sign" },
            "income_before_income_taxes": { "type": "number" },
            "provision_for_income_taxes": { "type": "number" },
            "net_income": { "type": "number" }
          }
        },
        "eps": {
          "type": "object",
          "properties": {
            "basic": { "type": "number", "description": "Earnings per share — Basic, in USD per share" },
            "diluted": { "type": "number", "description": "Earnings per share — Diluted, in USD per share" },
            "basic_shares_used": { "type": "number", "description": "In thousands of shares per the table header" },
            "diluted_shares_used": { "type": "number", "description": "In thousands of shares per the table header" }
          }
        },
        "segments": {
          "type": "array",
          "description": "Net sales by reportable geographic segment (footnote (1) on page 1).",
          "items": {
            "type": "object",
            "properties": {
              "segment": { "type": "string", "enum": ["Americas", "Europe", "Greater China", "Japan", "Rest of Asia Pacific"] },
              "net_sales": { "type": "number" }
            }
          }
        },
        "product_categories": {
          "type": "array",
          "description": "Net sales by product category (second footnote (1) on page 1).",
          "items": {
            "type": "object",
            "properties": {
              "category": { "type": "string", "enum": ["iPhone", "Mac", "iPad", "Wearables, Home and Accessories", "Services"] },
              "net_sales": { "type": "number" }
            }
          }
        },
        "balance_sheet": {
          "type": "object",
          "properties": {
            "cash_and_cash_equivalents": { "type": "number" },
            "marketable_securities_current": { "type": "number" },
            "accounts_receivable_net": { "type": "number" },
            "vendor_non_trade_receivables": { "type": "number" },
            "inventories": { "type": "number" },
            "other_current_assets": { "type": "number" },
            "total_current_assets": { "type": "number" },
            "marketable_securities_non_current": { "type": "number" },
            "property_plant_equipment_net": { "type": "number" },
            "other_non_current_assets": { "type": "number" },
            "total_non_current_assets": { "type": "number" },
            "total_assets": { "type": "number" },
            "accounts_payable": { "type": "number" },
            "other_current_liabilities": { "type": "number" },
            "deferred_revenue": { "type": "number" },
            "commercial_paper": { "type": "number" },
            "term_debt_current": { "type": "number" },
            "total_current_liabilities": { "type": "number" },
            "term_debt_non_current": { "type": "number" },
            "other_non_current_liabilities": { "type": "number" },
            "total_non_current_liabilities": { "type": "number" },
            "total_liabilities": { "type": "number" },
            "common_stock_and_additional_paid_in_capital": { "type": "number" },
            "accumulated_deficit": { "type": "number", "description": "Negative number — preserve sign" },
            "accumulated_other_comprehensive_loss": { "type": "number", "description": "Negative number — preserve sign" },
            "total_shareholders_equity": { "type": "number" }
          }
        }
      },
      "required": ["fiscal_period", "income_statement", "eps", "segments", "product_categories", "balance_sheet"]
    }
  },
  "settings": {
    "citations": { "enabled": true, "numerical_confidence": true }
  }
}

Schema design notes that materially affect Reducto accuracy:

• Name every field after the literal table label Apple uses ("Vendor non-trade receivables", "Wearables, Home and Accessories"). Reducto's field-name parser is sensitive — generic names like wearables_revenue reduce hit-rate on the line-item match vs. the literal phrase.
• Use enum constraints on segment and category to force Reducto to canonicalize against Apple's published labels (which have been stable across many years). This catches OCR drift like "Greater China" → "China" or "Wearables, Home and Accessories" → "Wearables".
• Both segments and product_categories are nested arrays under a top-level array property — you must set settings.array_extract: true in settings if the schema's top-level is just array (it's not here; we use object with array properties, so array_extract is not required).
• Negative numbers as parenthesized: Apple prints (279) for negative operating-other-income, (15,552) for accumulated deficit. Add "description": "Negative number — preserve sign" on every field that can go negative; otherwise Reducto sometimes drops the sign.
• settings.citations.enabled: true is strongly recommended — it returns the source page + bbox + confidence per field. Use it in production to fact-check Reducto's output against the PDF before publishing. Note: citations + chunking are mutually exclusive (per Reducto docs).
• For Q4 PDFs that contain both quarterly and full-year columns, add a period: "quarterly" | "annual" field to each statement block (or run /extract twice with different system_prompt instructions) — otherwise Reducto picks one column arbitrarily and can mix quarterly + annual values in the same response.

4. Validate

Sanity-check Reducto's output before consuming downstream:

• Identity: total_net_sales should equal products_revenue + services_revenue (off-by-one is sometimes a rounding artifact in source — tolerate ±1).
• Identity: sum(segments[].net_sales) == total_net_sales and sum(product_categories[].net_sales) == total_net_sales.
• Identity: total_assets == total_liabilities + total_shareholders_equity (fundamental balance-sheet identity — if this fails by > ±1, Reducto mis-extracted at least one line).
• EPS sanity: net_income / diluted_shares_used (with shares scaled from thousands to actual: × 1000) should approximate eps.diluted to ~3 decimal places.
• Sign sanity: other_income_expense_net, accumulated_deficit, accumulated_other_comprehensive_loss, and parenthesized cash-flow line items should be negative.

If any identity fails, re-run with settings.deep_extract: true (higher cost / latency, agentic refinement).

Browser fallback

The browser path is only useful if (a) Apple changes the URL pattern and search-engine indexing hasn't caught up yet, or (b) Reducto's API is down and you need to hand-eyeball the numbers from a rendered PDF. To browse:

SID=$(browse cloud sessions create --keep-alive \
       | node -e "let s='';process.stdin.on('data',c=>s+=c).on('end',()=>process.stdout.write(JSON.parse(s).id))")
export BROWSE_SESSION="$SID"
browse open "https://www.apple.com/newsroom/" --remote --session "$SID"
# Click into the relevant quarterly-results press release card (year/quarter inferred from prompt),
# then on the post page locate the embedded /newsroom/pdfs/...pdf link.
browse get html body --remote --session "$SID" \
  | grep -oE 'href="/newsroom/pdfs/[^"]+\.pdf"' | head -1
browse cloud sessions update "$SID" --status REQUEST_RELEASE

Note: when you browse open the PDF directly, Browserbase's Chrome does not render the PDF inline in the viewport (the screenshot will still show the previously-loaded page even though browse get url reports the PDF URL). That's a sandbox artifact — the PDF is fetched correctly server-side; just don't try to OCR a screenshot. Use browse cloud fetch <pdf-url> to get the PDF bytes if you need them.

Site-Specific Gotchas

• Apple does NOT post the full 10-Q or 10-K as a PDF on apple.com. The actual SEC filings live at data.sec.gov/submissions/CIK0000320193.json (Apple's CIK is 0000320193) and the primary documents are .htm files (e.g. aapl-20260328.htm for FY26 Q2), not PDF. Reducto's supported input formats do not include .htm — see the API reference table (PDF / DOCX / XLSX / PPTX / images only). If a user asks for "the 10-Q PDF", they almost certainly want the Condensed Consolidated Financial Statements PDF this skill targets — it contains every line item asked for (revenue, segments, EPS, balance sheet). If they truly need the full 10-Q (MD&A, footnotes, risk factors), you must either (a) print the EDGAR HTM to PDF first via a headless browser, or (b) use the EDGAR FilingSummary.xml + iXBRL viewer print path — both out of scope for this skill.
• URL pattern changed between FY25 and FY26. FY25 and earlier (and FY26 Q1) use fy{YYYY}-q{Q} (hyphenated); FY26 Q2 onward dropped the hyphen and uses fy{YYYY}q{Q}. Always try both, or use press-release-page discovery rather than hardcoding the URL.
• apple.com/investor-relations/ is a 404. Apple's actual investor relations site is investor.apple.com, which is a Q4 Inc.–powered IR platform (widgets.q4app.com, identity.q4inc.com). The press-release PDFs are NOT served from investor.apple.com — they live under www.apple.com/newsroom/pdfs/. Do not waste time scraping investor.apple.com/sec-filings/default.aspx looking for a PDF link; the SEC-filings list on that page deep-links to SEC EDGAR HTM, not Apple-hosted PDFs.
• Press-release URL month drifts year-over-year. FY25 Q2 was /newsroom/2025/05/...; FY26 Q2 was /newsroom/2026/04/... (one month earlier). Don't hardcode the month — use search.
• All PDFs are 3 pages (Q1–Q3) or 4 pages (Q4), named CONDENSED CONSOLIDATED STATEMENTS OF OPERATIONS (page 1, with two embedded segment + product footnotes), CONDENSED CONSOLIDATED BALANCE SHEETS (page 2), CONDENSED CONSOLIDATED STATEMENTS OF CASH FLOWS (page 3). Q4 PDFs add a (In millions, except per-share amounts) annual summary page.
• Apple's fiscal year ends ~Sep 30, so "Q1 FY25" = Oct–Dec 2024 (NOT Jan–Mar 2025). Reducto extracts the literal column headers, which say "Three Months Ended March 29, 2025" — make sure your fiscal_period field is derived from the column header, not assumed from the URL. Many tools mis-label Apple quarters by 1.
• Shares are reported in thousands, dollars in millions. Every PDF says (In millions, except number of shares, which are reflected in thousands, and per-share amounts). If you compute net_income / diluted_shares × $/share, multiply the shares by 1000 first or you'll be off by 1000×.
• Q4 PDFs have both quarterly AND annual columns in the same statement. Reducto will pick one arbitrarily without a period_label field or system-prompt steer. For Q4, either (a) ask for both via separate quarterly_* and annual_* schema blocks, or (b) run /extract twice with different system_prompt overrides.
• Negative numbers are parenthesized, not minus-signed. Apple uses (279) for negative other_income/(expense), net. Reducto handles this correctly if the schema field's description explicitly says "negative number — preserve sign"; without that, it occasionally returns 279 (positive) for an obviously-negative line item. Mandatory on: other_income_expense_net, accumulated_deficit, accumulated_other_comprehensive_loss, and every parenthesized cash-flow line.
• PDF passes the public URL straight to Reducto — no upload step needed. Reducto accepts the apple.com newsroom PDF URL directly as the input field on /extract (and /parse). Don't waste a round-trip uploading via /upload unless you need a file_id to reuse across multiple endpoints (e.g. you want to classify, then parse, then extract — only then is /upload worth it).
• Q4 CDN can take ~30s on a cold fetch. First request to a brand-new quarter's PDF in a fresh CDN region can take 20–30s; subsequent requests are sub-second. Don't set Reducto's HTTP timeout too tight; default 60s is fine.
• Sandbox DNS is not always direct. From the browse.sh skill-generator sandbox, curl https://www.apple.com/... fails with Could not resolve host (no direct egress). Use browse cloud fetch instead, which routes through Browserbase's egress. End-user agents that DO have direct internet egress can curl directly — this is only a sandbox-specific gotcha.
• Reducto API key is the agent's, not embedded here. This skill assumes REDUCTO_API_KEY is in the calling agent's environment. Sign up at studio.reducto.ai (free tier ~100 pages/mo at time of writing); key format is a long opaque string. The skill does not ship a key.

Expected Output

For a successful FY25 Q2 extraction (https://www.apple.com/newsroom/pdfs/fy2025-q2/FY25_Q2_Consolidated_Financial_Statements.pdf), Reducto's /extract response shape with citations.enabled: false would be:

{
  "result": [
    {
      "fiscal_period": "FY25 Q2",
      "period_end_date": "2025-03-29",
      "currency": "USD",
      "units": "millions",
      "income_statement": {
        "products_revenue": 68714,
        "services_revenue": 26645,
        "total_net_sales": 95359,
        "products_cost_of_sales": 44030,
        "services_cost_of_sales": 6462,
        "total_cost_of_sales": 50492,
        "gross_margin": 44867,
        "research_and_development": 8550,
        "selling_general_administrative": 6728,
        "total_operating_expenses": 15278,
        "operating_income": 29589,
        "other_income_expense_net": -279,
        "income_before_income_taxes": 29310,
        "provision_for_income_taxes": 4530,
        "net_income": 24780
      },
      "eps": {
        "basic": 1.65,
        "diluted": 1.65,
        "basic_shares_used": 14994082,
        "diluted_shares_used": 15056133
      },
      "segments": [
        { "segment": "Americas", "net_sales": 40315 },
        { "segment": "Europe", "net_sales": 24454 },
        { "segment": "Greater China", "net_sales": 16002 },
        { "segment": "Japan", "net_sales": 7298 },
        { "segment": "Rest of Asia Pacific", "net_sales": 7290 }
      ],
      "product_categories": [
        { "category": "iPhone", "net_sales": 46841 },
        { "category": "Mac", "net_sales": 7949 },
        { "category": "iPad", "net_sales": 6402 },
        { "category": "Wearables, Home and Accessories", "net_sales": 7522 },
        { "category": "Services", "net_sales": 26645 }
      ],
      "balance_sheet": {
        "cash_and_cash_equivalents": 28162,
        "marketable_securities_current