554 primary sources, 7 parallel sub-researchers, 14 cross-pollination plays across 5 domains. The data is free; the moat is the contract (idempotency keys, decay-aware timestamps, MCP-callable shape, x402 micropayments). The wedge is the cross-domain join. Three MVPs shippable in 90 days: Biotech Catalyst Radar, Sanctions × UBO via x402, and FEC × LDA × GDELT Money-in-Politics via x402.
Forecasters cluster around three credible models. The variance comes from disagreement on what to count, not on capabilities. McKinsey's $2.9T US / $28.7T global by 2030 is the most rigorous anchor (Nov 2025 methodology). Sequoia's $5T–$50T range (Pat Grady, AI Ascent 2026) is the most honest about uncertainty. a16z's agent-as-customer view measures the live agent-payment rails (Coinbase x402 = $1.6M/month after wash-trading filter).
The seven parallel sub-researchers covered these six domains. The remaining 187 hosts a long tail of academic papers, complementary data catalogs, and the agent-economy literature.
APIs.io scored a16z's 1,100 portfolio companies on an Agent-Readiness scale. The structural finding: 97 of 124 agent-ready APIs are held back by one missing primitive — a documented idempotency key. The gap is contract-level, not data-level.
| Metric | Value | Significance |
|---|---|---|
| Publishing APIs in a16z portfolio | 379 | The agent-API universe |
| Reached "agent-native" tier | 12 (3.2%) | Top tier · requires both idempotency and explicit error semantics |
| Agent-ready, held back by gate | 45 / 124 (36%) | Score high enough · blocked on contract primitives |
| Lack documented idempotency | 97 / 124 (78%) | The single biggest blocker |
| OpenAI score (gated) | 86.0 | Highest scorer · held out of agent-native despite leading the agent world |
| ElevenLabs score (gated) | 83.8 | Same · leading capability · fails contract |
| Threshold for agent-native | 88.0 | Idempotency + error semantics required |
Card rails (2.9% + $0.30) make sub-cent calls economically impossible. x402 (HTTP 402 stablecoin) is the breakthrough — AgentCash reports median $0.028/call, range $0.002–$0.440. $0.05 per call with 3 free calls/day/IP is the consensus starter.
Each play is a join across two or more free public datasets with a refresh-cadence moat. Each has a JSON-shaped, deterministic or scored, falsifiable output — the shape an MCP-wrapped tool needs.
| # | Play | Datasets joined | Insight | Feasibility |
|---|---|---|---|---|
| 1 | Sanctions × UBO | OpenSanctions · Companies House PSC · OpenCorporates · OFAC · GLEIF LEI | Of sanctioned entity X, who is the human ultimately in control? | med |
| 2 | Permits × Satellite × Address | MS Footprints · OpenAddresses · Socrata permit portals | Is construction at this parcel advancing faster than the permit cycle? | med |
| 3 | Climate × Yields × Commodity | NOAA · NASS · CME/AGMARKNET | Will corn yield Y% above trend by harvest, given soil-moisture anomaly X? | med |
| 4 | Biotech Catalyst Radar ★ | ClinicalTrials.gov v2 · EDGAR 8-K · Polygon prices | Historical move-on-readout distribution for similar mechanisms? | low |
| 5 | News × Market × Filings | GDELT · FRED · EDGAR · Polygon | Forward 5-day move after 8-K + GDELT news-cluster coincidence | med |
| 6 | Patents × Papers × R&D | PatentsView · OpenAlex · EDGAR 10-K | Is the company's R&D spend productive? | med |
| 7 | Product Hunt × YC × Crunchbase | Product Hunt · YC directory · Crunchbase · GitHub | Of launches in category X in the last 30 days, which have YC-backing? | low |
| 8 | FEC × LDA × News ★ | OpenFEC · Senate LDA · GDELT | Lobbyist X reached company Y mentions how often? | low |
| 9 | Air × Demo × Health (EJ) | AirNow · EJScreen · CDC PLACES | Census-tract-level environmental-justice composite score | med |
| 10 | Transit × Zoning × Buildings | GTFS · zoning · MS Footprints | TOD score for any parcel near a frequent-transit stop | med |
| 11 | Courts × Filings × News | CourtListener/RECAP · EDGAR · GDELT | Litigation discovery · defendant + news context | med |
| 12 | GDELT × Filings | GDELT · EDGAR | Reputational-risk feed (8-K → 7-day news tone) | low |
| 13 | Common Crawl × Registry | Common Crawl · Wayback · OpenCorporates | Corporate web-presence time-series | low |
| 14 | Wikipedia × Wikidata × Patents | Wikipedia · Wikidata · PatentsView · OpenAlex | Cross-source entity resolution at scale | med |
Every play is a join across an event/observation source and a structured-attribute source. The observation side (permits, news, climate, papers) is generally free; the attribute side (registry, filing, identifier) is also generally free. The cost is in the join, not the data.
Every play has a JSON-shaped output, a deterministic or scored join, and a falsifiable insight — the shape that an MCP-wrapped tool needs. Of the 14 plays, plays 1, 4, 5, 8, 9, 10, 11, 12, 13, 14 are direct fits; plays 2, 3, 6, 7 are dashboard-shaped but the same data plumbing underwrites an agentic variant.
Every play's defensibility comes from a sustained join, not a one-time data dump. The moat is the refresh cadence (daily EDGAR pull, weekly Census, daily CourtListener, daily GDELT), the entity resolution, and the cited evidence. None of this is hard in isolation; it is hard in combination.
The same pattern is the reason Lexis / Westlaw / Lex Machina / PatSnap charge enterprise prices for a product shape that is mostly public data. Round 2's new wrinkle is that several verticals are now standing on post-2023 regulatory triggers (SEC climate, SEC cyber, FEMA Risk Rating 2.0, EU IUU, EU NIS2) that create structural buyer for the join.
Each is sized to a 1-3 person team MVP shippable in ≤90 days. Ordered by recommended priority. The MCP-callable plus x402-priced shape is the wedge no incumbent has shipped yet.
ClinicalTrials.gov v2 + EDGAR 8-K + Polygon prices + pdufa.bio-style cohort stats. Insight: "For ticker X, given a Phase 3 readout on date Y, what is the historical move-on-readout distribution for similar mechanisms?"
Customer: sell-side biotech analyst or biotech-focused RIA. $1k/yr/ticker
RANK 1 low risk
OpenSanctions + Companies House PSC + OpenCorporates + OFAC + GLEIF LEI. Insight: "Of sanctioned entity X in jurisdiction Y, who is the human ultimately in control with $X in 12 jurisdictions?"
Customer: neobank KYC, marketplace T&S, AML compliance. $0.05/match · x402
low medium disclaimer risk
OpenFEC + Senate LDA + GDELT Cloud v2. Insight: "Lobbyist X spent $Y on issue Z; what companies did they reach mentions of, and what is the inferred influence on outcomes W–Z?"
Customer: ESG-fund analyst or investigative-journalism newsroom. $0.10/call · x402
lowest technical risk
Three patterns repeat across the agent-API frontier. Plan for each.
Clearview AI scraped ~21.6B Facebook images + ~4.1B Google/day over 2020–2022; settlement required permanent deletion by 2023-12-31, real-time domain blocklist, third-party audit every 90 days. Lambda Labs / Scale AI follow-ons: $14K–$15K per individual.
Reddit 2024 OAuth enforcement shut down anonymous scraping (60 calls/min/IP unauth, 3,600/min/user private-app). The HN-discussed corollary: "APIs to retrieve public information suck because they are designed to be rate-limited — public-facing websites are much less likely to set rate limits."
Agents that find a generous free tier exhaust it without converting. Reddit May 2026 unauthenticated RSS throttled to "1 request per 60-second window," breaking legitimate use cases. Build a Pay-and-Pray passthrough + a 3-call/day/IP free tier as the standard pattern.
Pushshift dead (use PullPush / Arctic Shift); Reddit official API approval queue closed; X API v2 paid ($100/mo Basic); Transfermarkt DMCA-threatened; BAILII bot-walled; Spotify Dev Mode requires Premium (Feb 2026); PubChem hard 5 req/sec; KEGG academic-only; ChEMBL CC-BY-SA copyleft.
Top three from each chapter's wedge candidates, with the gap each fills.
| Domain | Pick #1 | Pick #2 | Pick #3 |
|---|---|---|---|
| Financial | FRED (macro bus) | OpenSanctions + UBO graph | GLEIF LEI (entity backbone) |
| Geospatial | OpenStreetMap | NASA Earthdata + Copernicus | OpenAQ |
| Scientific | OpenAlex | UniProt + PubChem + Ensembl stack | ClinicalTrials.gov + PatentsView |
| Web / content | Common Crawl + Wikipedia/Wikidata | MusicBrainz + Discogs + AcoustID | CourtListener + CanLII |
| Government | CourtListener + PACER pray-and-pay | USAspending + OpenCorporates + OpenSanctions | UK Companies House + Land Registry |
# Week 1 — ingest psql biosignal-dev < ingest_trials.sql # ClinicalTrials.gov v2 · top 200 biotech tickers psql biosignal-dev < ingest_edgar.sql # EDGAR 8-K filings · join to CIK dbt run # transform in Postgres # Week 2 — cohort statistics python cohort_stats.py --ticker X # historical move-on-readout distribution # Week 3 — x402 endpoint uvicorn catalyst_api:app --port 8000 # MCP-callable: get_catalyst_distribution(ticker, trial_id) # Week 4 — pilot landing claude --mcp-config catalyst.json # register the MCP server
554 distinct primary URLs registered in the shared ledger at api-data-research/ledger.json. The ledger is shared across all 7 parallel sub-researchers and the parent agent. The full Sources block is appended to each chapter draft via sources.py render --replace-in and lives in the chapter files.
Top five chapter files backing this companion:
Seven parallel deep-dives back every section. Each is ~25–60 KB with its own rendered Sources block.