PROJECT 801 — OCTOBER 24, 2026 RELEASE

Run with Python 3 standard library, no account or API key:
  python3 inflation_model.py inputs forecast-reproduced.json
Compare numeric forecasts and evaluation with forecast-oct24.json. Source file retrieval mtimes may differ after ZIP extraction; original retrieval times and SHA256 hashes remain in the frozen JSON. The forecast is fixed before the October CPI release, expected November 10, 2026. This archive is published with the site; a local checksum by itself is not evidence of publication timing.

Independent model: AR(1) on valid adjacent monthly CPI changes in the latest 36 calendar months. OLS intercept; normal residual model; residual degrees of freedom n-2. Calendar gaps stay missing. October2025 CPI index is missing, so October/November2025 changes are missing; evaluation also skips December2025 because the previous month is missing. Multi-step forecasts propagate AR innovation variance, omit parameter uncertainty, and are not full economic-policy projections.

Evaluation: common valid rolling origins January2016–September2026, 126 predictions. Latest revised CPI vintage, not genuine real-time-vintage testing. Last-month baseline and available-observation mean over12 calendar months (minimum10) are shown. This retrospective comparison is not an untouched holdout or proof the method generalizes. No parameter tuning to win the comparison. Inflation data: FRED CPIAUCSL (BLS seasonally adjusted CPI-U index); not year-over-year CPI. Unrounded month changes are modeled, then rounded outcome bucket probabilities are integrated using half-tenth boundaries.

Markets: read-only public Kalshi/Gamma API snapshots, no trades. Kalshi thresholds differ from Polymarket partition buckets; both refer to October2026 monthly SA CPI, but delay/fallback rules differ. No assertion of arbitrage. Preserve bid/ask, source update time, retrieval time, identifiers and quality. Missing or one-sided quotes have no midpoint. Only the explicitly complete9-bucket partition is normalized (raw mass0.988); 0.5% bucket spread is wide. Price-implied estimates are not independently calibrated probabilities.

Elections:76 D/R contracts matched to38 inventory races. Partial collection:82 of139 series requests returned rate limits. Missing means missing, not50%. Contract candidate labels are not an official candidate roster. Excludes27/28 races, unmatched series, ambiguous inventory matches. Party prices are not forced to sum to1. No independent calibrated full2026or2028 election forecast is released. Original synthetic engine remains a methods demonstration.

Fiscal: fiscal-math.js calculates a stated scenario for GROSS debt. Starting Treasury gross debt40.44434270938915T October22,2026; target22.71940175343378T September30,2019. Annual overall deficit is a user assumption tapering75/50/25/0 percent2027–2030; post2030 debt=D*(1+r)-primarySurplus+stockFlowAdjustment. Gross debt requires reconciliation with federal account activity, not just consolidated cash flows. Zero stock-flow adjustment is an explicit assumption. Average rate on gross debt simplifies public and intragovernmental financing. No full policy score, no automatic cancellation of protected benefits, and no gold revaluation counted as recurring revenue.

Sources: FRED graph CSV; Treasury Debt to the Penny; Kalshi public trade-api/v2; Polymarket Gamma public-search. Raw request URLs appear in manifest.json; CPI sources also in frozen forecast JSON. The deployed read-only /api/signals route retrieves current reviewed OctoberCPI quotes separately, cached up to5minutes, without overwriting this frozen forecast. Replit MorninEdge source access is pending; this release does not claim its private repository was integrated.
