Show HN: Ford's $1.3B USMCA risk, priced from prediction markets (open source)

An open-source tool called Election Risk Desk, built on the Agent Skills standard and usable in Claude Code, Codex, Cursor and Gemini, prices how the Nov 3, 2026 US elections could affect individual companies' profits using Kalshi and Polymarket prediction markets. In its Ford example, run on Oct 28 prices, the tool rates election risk LOW and policy risk MEDIUM, with the largest exposure a −$1.30B annual hit (−19.2% of profit) if the US gives notice to leave USMCA before 2027, an outcome Kalshi prices at 6%. Across eight companies run end to end, only PG&E drew a High election risk rating, and the tool's stated surprise is that which party wins on Nov 3 moves these odds less than headlines suggest.

An AI risk desk for what the Nov 3, 2026 US elections could do to a company's profit. Name a company. Your coding agent staffs a desk: analysts map how the company makes its profit, what each candidate would do, and every Kalshi and Polymarket market that can move it. A Risk Officer sizes each risk with a small engine, a Red Team flags the weak numbers, and every run ends in two verdicts, with sources. Works in Claude Code, Codex, Cursor, Gemini and any agent that reads Agent Skills https://agentskills.io . See it without installing: live reports for Ford https://juwonyee.github.io/election-risk/examples/F/report.html , PG&E https://juwonyee.github.io/election-risk/examples/PCG/report.html , Coinbase https://juwonyee.github.io/election-risk/examples/COIN/report.html and DraftKings https://juwonyee.github.io/election-risk/examples/DKNG/report.html . Install it one line, below install , then ask your agent: What does Nov 3 mean for Ford? The desk board, in the chat and at the top of a local web report Ford's https://juwonyee.github.io/election-risk/examples/F/report.html , prices of Oct 28 : | Analyst | Signal | Why | |---|---|---| | Exposure | EXPOSED | biggest if it happens: −$1.30B a year −19.2% of profit | | Policy | WEAK LINK | for results of 1% of profit or more, who wins moves the chance by 5 points at most | | Market | PRICED | 3 markets on Kalshi and Polymarket | | Red Team | CAUTION | 8 flags to read | Risk Officer: Election risk LOW · Policy risk MEDIUM - Election risk LOW: nothing passes the bar. The biggest is +$1.00B a year if Congress repeals the Section 232 auto and metals tariffs, but only 5% even if Democrats win. - Policy risk MEDIUM: Kalshi puts 6% on the US giving notice to leave USMCA before 2027. For Ford: −$1.30B a year −19.2% of profit . - Red Team: the USMCA impact is an estimate, and at the high end of its range −$2.50B policy risk would be HIGH. And a card to share the report saves it as PNG : Eight household names, run end to end Ford on Oct 28 prices, the rest on Oct 27 . Each links to its full summary, profit structure and sources. | Company | Election risk | Policy risk | Biggest risk | Market | Profit, per year | |---|---|---|---|---|---| | PG&E https://github.com/juwonyee/election-risk/blob/main/examples/PCG/summary.md | High | none | Becerra wins 96% → 30% chance of a rate freeze, vs 0% under Hilton | Kalshi | −$600M −12.6% | | DraftKings https://github.com/juwonyee/election-risk/blob/main/examples/DKNG/summary.md | Medium | High | Congress bans sports contracts on prediction markets in 2026 | Polymarket 18% | +$150M +24.2% | | Coinbase https://github.com/juwonyee/election-risk/blob/main/examples/COIN/summary.md | Low | High | Clarity Act crypto market structure becomes law before 2028 | Kalshi 21% | +$370M +25.8% | | Ford https://github.com/juwonyee/election-risk/blob/main/examples/F/summary.md | Low | Medium | US gives notice to leave USMCA before 2027 | Kalshi 6% | −$1.3B −19.2% | | GM https://github.com/juwonyee/election-risk/blob/main/examples/GM/summary.md | Low | Medium | Same USMCA notice | Kalshi 8% | −$1.7B −13.3% | | Disney https://github.com/juwonyee/election-risk/blob/main/examples/DIS/summary.md | none | Medium | Federal film and TV production tax credit becomes law | Kalshi 26% | +$400M +2.3% | | Uber https://github.com/juwonyee/election-risk/blob/main/examples/UBER/summary.md | Low | none | California AG suit ends in a Lyft-sized settlement: 70%, and who wins moves it at most 11 points | Kalshi | −$170M −3.1% | | Starbucks https://github.com/juwonyee/election-risk/blob/main/examples/SBUX/summary.md | Low | Low | New York City enacts a $30 minimum wage by 2030 | Kalshi 8% | −$60M −2.0% | "none" means no market was linked to the company on that side. The surprise: who wins on Nov 3 moves these odds less than the headlines suggest. Only one of the eight gets a high election risk. The bigger risks come from markets that price the policy directly and barely depend on the vote: the Clarity Act and sports contracts high , USMCA for Ford and GM medium . Where the election does matter, the Red Team keeps it honest: research finds a new utility commission moves allowed returns by about 0.15 points, and attorneys general of both parties sue companies, so PG&E and Uber carry those flags. Not investment advice. Analyst drafts for human review. No recommendation to buy, sell or hold any security or to trade any contract. Markets are shown as information only. Four skills, one per desk role. Each works on its own; election-risk calls the other three by name. | Role | Skill | Ask it on its own | |---|---|---| | Exposure analyst | profit-structure | "Which of Ford's profit lines can a government move?" | | Policy analyst | candidate-positions | "What would California's next governor do on electricity rates?" | | Market analyst | prediction-markets | "What are the odds on a data-center moratorium?" | | Risk Officer + Red Team | election-risk | "What does Nov 3 mean for Uber?" | - The agent researches, the engine calculates. Every number comes from a small Python engine standard library only , not from the model's arithmetic. - Two kinds of market. Direct : the market resolving Yes is the policy, so its price is the chance. Race : the market prices who wins, times the chance the policy follows each winner. - Two verdicts. Election risk = share of profit × how much who wins moves the chance. Policy risk = share of profit × the market's chance. High at 2% of profit or more, Medium at 0.5%. Only results that pass the bar count at least 1% of profit, volume 10,000, chance 5% ; when none does, the board says what came closest and why it falls short. - Every impact shows how it was built : the formula with each number and its source, and a low-high range. - Red Team. It checks whether each verdict survives its own range: "Policy risk HIGH ... would be LOW at the low end" means the verdict rests on one estimate. It also flags positions with no source link counted as 0 , thin markets and estimated impacts. Ask your agent recommended : Install election-risk from https://github.com/juwonyee/election-risk Or install it yourself Claude Code claude plugin marketplace add juwonyee/election-risk claude plugin install election-risk@election-risk Codex codex plugin marketplace add juwonyee/election-risk codex plugin add election-risk@election-risk Gemini gemini extensions install https://github.com/juwonyee/election-risk Cursor git clone --depth 1 https://github.com/juwonyee/election-risk ~/.cursor/plugins/local/election-risk Any other agent, via the skills CLI npx skills add juwonyee/election-risk Install the plugin or the skills, not both you would see every skill twice . Needs Python 3.10+. No packages, no API keys. Race results need to know what each candidate would actually do. That lives in a policy book : one file per Nov 3 contest, every position with a source link, and the chance it happens if they win, from a research-based probability table https://github.com/juwonyee/election-risk/blob/main/skills/candidate-positions/references/probability-table.md veto override rates, governors' proposal success, ballot measure pass rates and more . The book is not in this repo, on purpose: candidate positions in a public repo could be edited by anyone. On the first run your agent asks you to pick: - Build it yourself : your agent researches every contest with sources. It takes a while, and it's reused for every company afterwards. - Download ours : ready in a minute, built by us on the date shown, possibly out of date. Your agent tells you that and asks before running erisk book download --accept-risk . Every report prints which book it used built or downloaded, date, fingerprint and when prices were fetched. - Chances are upper bounds. The probability table comes from historical rates over a full term, used here for about two years. - Impacts are often estimates. Each one is labeled verified or estimated and shows its formula, so you can check it. - The book can be wrong or out of date. Positions with no source count as 0, and the agent lists rows it doubts as weak links. - Only priced risks. A risk with no Kalshi or Polymarket market is left out. Unreachable sources are listed, never worked around. - Prices move. The examples use Oct 27 prices. Raw price snapshots are not kept in this repo venue terms . It won't recommend trades, size positions, help open market accounts, or work around a venue's regional restrictions. It names no small businesses franchised chains are discussed as "a typical store" . python3 engine/erisk.py markets pull every open Nov 3 + policy market python3 engine/erisk.py book status policy book coverage python3 engine/erisk.py book find --tag utilities --state CA python3 engine/erisk.py markets show GOVPARTYGA-26 --venue kalshi ERISK OUT=examples python3 engine/erisk.py compute --company SO --no-open test fixture, synthetic prices See AGENTS.md https://github.com/juwonyee/election-risk/blob/main/AGENTS.md . Edit engine/ , then python3 scripts/sync engine.py ; check with python3 scripts/validate skills.py and python3 -m unittest discover tests . MIT.