# What runs on 801

| Display | Code | Inputs/output | Status |
| --- | --- | --- | --- |
| Oct–Dec CPI baseline | inflation_model.py | inputs/oct24; published/forecast-oct24.json | Independent 36-month AR(1), frozen Oct 24, 2026 |
| Race and chamber-control prices | browser/signals.js, browser/app.js, browser/market-connector.ts | published/race-markets-oct24.json; forecast JSON controls | Market quotes; not 801 independent election predictions |
| 2026/2028 engine | election_model.py; browser/models.js | examples | Synthetic demonstrations; no calibrated election forecasts released |
| Human First / Grift | reproduce_scores.py; browser/accountability-math.js | published/officials.json | Two-vote policy pilot / 15 unverified editorial ratings; verified Grift null for all |
| Score/election link | browser/accountability-math.js | User inputs | Sensitivity scenario; coefficients default zero |
| Labor-to-capital components | policy_arithmetic.py; browser/policy-math.js | Explicit assumptions | Component arithmetic, not full fiscal score |
| Debt to 2040 | browser/fiscal-math.js | Treasury snapshots and assumptions | Recurrence scenario, not macroeconomic forecast |

CPI: x(t)=a+phi*x(t-1)+epsilon; OLS intercept on adjacent valid observations in the latest 36 calendar months. Missing months are not bridged. Gaussian intervals omit parameter uncertainty. Evaluation uses revised latest-vintage data, not real-time-vintage holdouts. A baseline is not proof of superior performance.

Score sensitivity: logit(p)=logit(p0)+betaH*(HA-HB)/100-betaG*(GA-GB)/100. Nonzero terms require both corresponding candidate scores. Missing inputs are not imputed. No fitted nonzero score coefficients are shipped. Policy votes do not establish personal financial misconduct.

Browser and Python election demos use different random generators; reproduce each implementation with its own seed. Their draws are not bit-identical. Python reproduces the published CPI output.

Preserve frozen releases after observing outcomes. Add separately versioned corrections and results audits. Local checksums establish byte integrity, not independently trusted publication time. Historical Senate joins require service at the vote date; see published/corrections.json.