Apple 10-Q / 10-K PDF → Reducto Structured Financial Extraction

Purpose

Download a quarterly (10-Q) or annual (10-K) financial PDF from Apple's investor relations site (investor.apple.com), send it to Reducto's document-AI platform (platform.reducto.ai), and return structured JSON containing revenue, segment breakdown (product line + geographic), EPS (basic + diluted), and key balance-sheet items. Read-only — never submits forms or modifies Apple/Reducto state.

When to Use

• Quarterly financial-data refresh for an Apple watchlist (run after each 10-Q/10-K filing — Apple files ~Jan 30, ~May 1, ~Aug 1, ~Oct 31 each fiscal year).
• Backfilling historical fundamentals across multiple Apple fiscal quarters or years.
• Any pipeline that needs structured financials sourced from the original SEC PDF rather than a third-party API, with citations linking each field back to a page/bbox in the source document.
• Comparable extraction against other large-cap issuers — the same platform.reducto.ai/extract schema works on any 10-Q/10-K PDF; this skill encodes the Apple-specific PDF-discovery half.

Workflow

The fastest path skips the PDF download entirely: Reducto's /parse and /extract endpoints accept a public URL as input (URL passthrough), so the runtime simply hands Reducto the s2.q4cdn.com/... URL and lets Reducto fetch it server-side. No local file I/O, no multipart upload, no 24-hour file_id expiry to manage. The browser path exists as a fallback for when the canonical PDF index moves.

Prerequisites the caller must supply

• REDUCTO_API_KEY — Bearer token from studio.reducto.ai → API Keys. The skill cannot run without one; Reducto does not have an unauthenticated tier for /parse or /extract.

1. Resolve the target PDF URL

Apple publishes every 10-Q and 10-K as a PDF on its Q4Inc-powered CDN (s2.q4cdn.com/470004039/...) and surfaces the links directly on the investor.apple.com homepage in a section titled Financial Data → Quarterly Earnings Reports. The canonical resolver is to scrape that homepage — naming conventions drift year over year so hand-constructing the URL from a template is unreliable (see Gotchas).

# Fetch the IR homepage and pull every Q4 CDN PDF link
curl -fsSL 'https://investor.apple.com/' \
  | grep -oE 'https://s2\.q4cdn\.com/470004039/files/[^"]+\.pdf' \
  | sort -u

Observed shape of the result (2026-05-21 snapshot, fiscal-year sorted desc):

For lighter parsing, Apple also publishes a 3-page Consolidated Financial Statements PDF per quarter (~110 KB vs ~1 MB for the full 10-Q) at https://www.apple.com/newsroom/pdfs/fy{YYYY}{q}/FY{YY}_{Q}_Consolidated_Financial_Statements.pdf — same income statement, segment, EPS, and balance sheet, just stripped of MD&A. Use this when you don't need the narrative sections.

Alternate discovery — if investor.apple.com is unreachable, fall back to SEC EDGAR's JSON submissions API which is rock-solid and unauthenticated:

curl -fsSL 'https://data.sec.gov/submissions/CIK0000320193.json' \
  -H 'User-Agent: your-name your-email@example.com'  # SEC requires a UA

The JSON has filings.recent.{form, filingDate, accessionNumber, primaryDocument} parallel arrays; for entries where form == "10-K" or "10-Q", the primary document URL is https://www.sec.gov/Archives/edgar/data/320193/{accessionNumber-with-dashes-stripped}/{primaryDocument}. Note: the EDGAR primary document is the inline HTML, not a PDF. Reducto's Parse + Extract accept HTML input ("Supported file types: PDF, ... DOCX, ... + HTML via URL"), so this path still works — but it returns HTML-derived structure, not PDF-derived. Prefer the Q4 CDN PDF for Reducto-native PDF parsing.

2. Call Reducto Extract with a JSON schema (URL passthrough)

The most efficient call is a single POST to https://platform.reducto.ai/extract with the PDF URL as input. Reducto runs Parse server-side, then the LLM-backed extractor returns only the fields you defined.

PDF_URL='https://s2.q4cdn.com/470004039/files/doc_earnings/2026/q2/filing/10Q-Q2-2026-as-filed.pdf'

curl -fsSL -X POST 'https://platform.reducto.ai/extract' \
  -H "Authorization: Bearer $REDUCTO_API_KEY" \
  -H 'Content-Type: application/json' \
  -d "$(cat <<EOF
{
  "input": "$PDF_URL",
  "instructions": {
    "schema": {
      "type": "object",
      "properties": {
        "company_info": {
          "type": "object",
          "properties": {
            "name":             {"type": "string"},
            "ticker":           {"type": "string"},
            "cik":              {"type": "string"},
            "filing_type":      {"type": "string", "enum": ["10-K", "10-Q"]},
            "fiscal_period":    {"type": "string", "description": "e.g. 'Q2 FY26' or 'FY25'"},
            "period_end_date":  {"type": "string", "description": "Fiscal period end, YYYY-MM-DD"}
          }
        },
        "income_statement": {
          "type": "object",
          "description": "Consolidated Statements of Operations. All currency in USD millions unless noted.",
          "properties": {
            "total_net_sales":    {"type": "number"},
            "cost_of_sales":      {"type": "number"},
            "gross_margin":       {"type": "number"},
            "operating_expenses": {"type": "number"},
            "operating_income":   {"type": "number"},
            "net_income":         {"type": "number"},
            "eps_basic":          {"type": "number", "description": "Earnings per share, basic, in USD"},
            "eps_diluted":        {"type": "number", "description": "Earnings per share, diluted, in USD"},
            "shares_basic":       {"type": "number", "description": "Weighted-avg shares used in basic EPS (in thousands)"},
            "shares_diluted":     {"type": "number", "description": "Weighted-avg shares used in diluted EPS (in thousands)"}
          }
        },
        "segment_breakdown_products": {
          "type": "object",
          "description": "Net sales by product category from the segment information disclosure.",
          "properties": {
            "iphone":                          {"type": "number"},
            "mac":                             {"type": "number"},
            "ipad":                            {"type": "number"},
            "wearables_home_and_accessories":  {"type": "number"},
            "services":                        {"type": "number"}
          }
        },
        "segment_breakdown_geographic": {
          "type": "object",
          "description": "Net sales by reportable geographic segment.",
          "properties": {
            "americas":             {"type": "number"},
            "europe":               {"type": "number"},
            "greater_china":        {"type": "number"},
            "japan":                {"type": "number"},
            "rest_of_asia_pacific": {"type": "number"}
          }
        },
        "balance_sheet": {
          "type": "object",
          "description": "Selected items from the Consolidated Balance Sheets as of the period-end date.",
          "properties": {
            "cash_and_cash_equivalents":   {"type": "number"},
            "marketable_securities_current": {"type": "number"},
            "total_current_assets":        {"type": "number"},
            "total_assets":                {"type": "number"},
            "total_current_liabilities":   {"type": "number"},
            "long_term_debt":              {"type": "number"},
            "total_liabilities":           {"type": "number"},
            "total_shareholders_equity":   {"type": "number"}
          }
        }
      },
      "required": ["company_info", "income_statement", "segment_breakdown_products", "balance_sheet"]
    }
  },
  "settings": {
    "citations": {"enabled": true, "numerical_confidence": false}
  }
}
EOF
)"

The response is a V3ExtractResponse whose result[0] matches your schema. citations (when enabled) attaches per-field source bboxes so the caller can verify each number against the original PDF. Cost is roughly 5–7 Reducto credits per 10-Q (≈25–40 pages) and 15–25 credits per 10-K (≈80–100 pages); use the smaller Consolidated Financial Statements PDF (~3 pages, ~1 credit) when MD&A isn't needed.

SDK form (Python) — identical semantics, terser:

import os
from reducto import Reducto

client = Reducto()  # reads REDUCTO_API_KEY
result = client.extract.run(
    input="https://s2.q4cdn.com/470004039/files/doc_earnings/2026/q2/filing/10Q-Q2-2026-as-filed.pdf",
    instructions={"schema": SCHEMA_DICT},
    settings={"citations": {"enabled": True, "numerical_confidence": False}},
)
print(result.result[0])

3. (Optional) Two-step Upload + Extract when you must hold the PDF locally

Skip this unless URL passthrough is impossible (e.g., you have a PDF behind your VPN that Reducto's servers can't reach).

# Step a — upload
UPLOAD=$(curl -fsSL -X POST 'https://platform.reducto.ai/upload' \
  -H "Authorization: Bearer $REDUCTO_API_KEY" \
  -F "file=@10Q-Q2-2026-as-filed.pdf")
FILE_ID=$(node -pe "JSON.parse(process.argv[1]).file_id" "$UPLOAD")
# FILE_ID is "reducto://<uuid>.pdf"; valid for 24 hours.

# Step b — extract using the file_id as input (same body as Section 2 with input=$FILE_ID)
curl -fsSL -X POST 'https://platform.reducto.ai/extract' \
  -H "Authorization: Bearer $REDUCTO_API_KEY" \
  -H 'Content-Type: application/json' \
  -d "{\"input\":\"$FILE_ID\",\"instructions\":{\"schema\":{...}}}"

Limits: 100 MB per direct upload (use presigned URLs for ≤5 GB); reducto:// file IDs expire after 24 h.

Browser fallback (only if both API paths fail)

Use this when Apple's homepage HTML changes, the Q4 CDN serves 403s, and SEC EDGAR is also unreachable — extremely unlikely but documented for completeness.

1. browse cloud sessions create --keep-alive — bare session is enough; Apple IR and SEC EDGAR are not anti-bot.
2. browse open https://investor.apple.com/ --remote
3. browse eval "[...document.querySelectorAll('a[href$=\".pdf\"]')].map(a=>({text:a.textContent.trim(), href:a.href})).filter(o=>o.href.includes('s2.q4cdn.com') || o.href.includes('newsroom/pdfs'))" — returns the full filings catalogue as JSON.
4. Pick the most recent 10-Q or 10-K href and feed it back into Section 2 as PDF_URL.

Never click into a "Download" or "Subscribe to alerts" button — those start an email-signup flow handled by Q4Inc's login.q4inc.com and are out of scope.

Site-Specific Gotchas

• investor.apple.com/investor-relations/financial-data/default.aspx is 404. The historical Financial Data subpage was decommissioned in favour of folding all earnings PDFs into the IR homepage's "Financial Data → Quarterly Earnings Reports" accordion. Direct fetch of the old URL returns Q4inc's branded 404. Always start from https://investor.apple.com/ and parse the homepage.
• PDF naming is inconsistent across fiscal years — do not templated-construct URLs. Observed in 2026-05 catalogue: FY26 Q1/Q2 use lowercase 10Q-Q{N}-{YYYY}-as-filed.pdf; FY25 Q1–Q3 also lowercase; FY24 10-K is 10-Q4-2024-As-Filed.pdf (note the typo-like Q4 in a 10-K filename, plus mixed-case As-Filed); FY23 Q1/Q2 live under doc_financials/ while FY23 Q3/Q4 and all of FY24/25/26 live under doc_earnings/; FY22 10-K uses parentheses: _10-K-2022-(As-Filed).pdf. The canonical resolver is always to scrape the homepage's anchor list rather than guess.
• Consolidated Financial Statements PDF path also drifts. fy2026q2/FY26_Q2_Consolidated_Financial_Statements.pdf has no hyphen; fy2025-q4/FY25_Q4_Consolidated_Financial_Statements.pdf has one. Scrape, don't template.
• Apple's fiscal year is not the calendar year. Apple's FY ends on the last Saturday of September. FY26 Q1 = Sep 28 2025 – Dec 27 2025; Q2 ends Mar 28 2026; Q3 ~Jun 27; Q4 (the 10-K) ~Sep 26. A "2026 10-K" doesn't exist until ~Oct 31 2026. When mapping calendar dates to filings, use SEC EDGAR's period_end_date (in the submissions JSON) as the canonical anchor.
• EDGAR primary document is HTML, not PDF. data.sec.gov/submissions/CIK0000320193.json lists primaryDocument: "aapl-20260328.htm" for 10-Q/10-K filings. EDGAR exposes a PDF rendering at https://www.sec.gov/cgi-bin/viewer?action=view&cik=320193&type=10-Q&dateb=&owner=include&count=40, but the canonical PDFs are Apple's own Q4-CDN copies. Reducto accepts both — pass whatever URL is closest at hand.
• SEC EDGAR requires a User-Agent header. curl https://data.sec.gov/submissions/CIK0000320193.json without -H 'User-Agent: name email@example.com' will succeed today from most IPs but the SEC's fair-use policy mandates it; failing to send a meaningful UA can get your IP rate-limited. The Q4-CDN endpoint has no such requirement.
• Reducto file_id (reducto://...) expires in 24 hours. If you upload-then-extract across a long delay, re-upload. URL passthrough sidesteps this entirely.
• Reducto Extract returns numbers in document units, not normalized to USD. Apple reports income-statement values in millions ("Total net sales $95,359" = $95.359B); EPS in dollars; share counts in thousands. The schema descriptions above match Apple's filings — do not try to coerce all numbers to a single unit in the schema.
• Segment table changed pre-FY18. Apple's "Wearables, Home and Accessories" line was reclassified in FY18; reading 10-K filings older than FY18 with the current product-segment schema will leave fields null. The Net Sales by Category reclassification PDF (Net-Sales-By-Category-Qtrly-FY18.pdf on the IR homepage) is the canonical bridge if you need to backfill.
• Q4Inc-powered IR site is Akamai-fronted but bare-friendly. investor.apple.com is served via Cloudflare + Q4inc's CDN, and a bare (un-stealthed) session fetches the homepage and Q4-CDN PDFs without issue — verified via browse cloud fetch with no proxies in 2026-05-21 run. The browser fallback does not need --verified or --proxies.
• Reducto cookbook for this exact task exists. https://docs.reducto.ai/cookbooks/web-browsing-browserbase and https://docs.reducto.ai/cookbooks/financial-analysis both use Apple 10-Q/10-K PDFs as their reference document. When in doubt, mirror the schema in those cookbooks — they're known-good against Apple's specific filing structure.
• citations: {enabled: true} doubles roundtrip latency but is recommended. Without citations, you get raw numbers with no provenance; with them, each field carries a bbox + page reference back into the PDF. For financial data where audit trail matters, the extra latency is worth it.

Expected Output

A single JSON object matching the schema declared in step 2. Real shape from a Q2 FY26 10-Q (filed 2026-05-01) — values are illustrative of the structure, not a guaranteed snapshot:

{
  "company_info": {
    "name": "Apple Inc.",
    "ticker": "AAPL",
    "cik": "0000320193",
    "filing_type": "10-Q",
    "fiscal_period": "Q2 FY26",
    "period_end_date": "2026-03-28"
  },
  "income_statement": {
    "total_net_sales": 95359,
    "cost_of_sales": 53814,
    "gross_margin": 41545,
    "operating_expenses": 15275,
    "operating_income": 26270,
    "net_income": 21882,
    "eps_basic": 1.46,
    "eps_diluted": 1.45,
    "shares_basic": 14994000,
    "shares_diluted": 15083000
  },
  "segment_breakdown_products": {
    "iphone": 46841,
    "mac": 7949,
    "ipad": 6402,
    "wearables_home_and_accessories": 7521,
    "services": 26645
  },
  "segment_breakdown_geographic": {
    "americas": 40315,
    "europe": 24454,
    "greater_china": 16002,
    "japan": 7298,
    "rest_of_asia_pacific": 7290
  },
  "balance_sheet": {
    "cash_and_cash_equivalents": 28162,
    "marketable_securities_current": 35257,
    "total_current_assets": 131697,
    "total_assets": 331233,
    "total_current_liabilities": 142500,
    "long_term_debt": 79100,
    "total_liabilities": 264101,
    "total_shareholders_equity": 67132
  },
  "_meta": {
    "source_pdf": "https://s2.q4cdn.com/470004039/files/doc_earnings/2026/q2/filing/10Q-Q2-2026-as-filed.pdf",
    "reducto_job_id": "<uuid>",
    "credits_used": 7
  }
}

When the requested filing does not yet exist (e.g., asking for FY26 10-K in May 2026), the resolver step returns no matching anchor on the homepage; emit {"success": false, "reason": "filing_not_published", "latest_available": {"filing_type": "10-Q", "fiscal_period": "Q2 FY26", "period_end_date": "2026-03-28"}} and stop before calling Reducto.

When Reducto returns a field as null (e.g., a 10-Q omits the full balance sheet detail), keep the key in the output with a null value rather than dropping it — downstream callers depend on the schema shape being stable.