Accountability Tracker

Attribution signal: Each article is scored by a large language model on a single question: is economic hardship being blamed on Ottawa, or on outside forces like US tariffs and global markets? Scoring runs from −1 (all external blame) to +1 (government held responsible). Articles that don't mention a federal political actor are filtered out by a relevance gate. This tracker uses national-circulation outlets only (see “Source coverage” below); regional and local articles are tagged at the row level and routed to per-province readings elsewhere.

Smoothing: The daily signal is smoothed with a 28-day exponentially weighted moving average (EWMA, half-life 14 days) and blended with public attribution surveys (direct polling questions asking who Canadians blame for economic conditions). The tracker then bins those daily values by ISO week and applies a 9-week sliding average, so each plotted point reflects roughly two months of underlying coverage rather than a single week.

Source coverage: The national outlets in the pool are CBC, the Globe and Mail, National Post, iPolitics, National Observer, CTV News, The Conversation, plus a MediaCloud backfill restricted (by URL domain) to national-circulation outlets such as Canadian Business Journal, Canadian Dimension, and Canada Standard. Regional outlets in the MediaCloud corpus (Toronto Star, Hamilton Spectator, Winnipeg Free Press, BC and AB local papers, etc.) and the provincial RSS feeds in the full 50-outlet pipeline are excluded from this national tracker.

Approval: Rolling aggregate from Abacus Data, Angus Reid, Léger, Ipsos, EKOS, and Research Co. Forward-filled between polls. The 40% midpoint reflects the long-run Canadian average across governments since 2000.

Confidence: Each weekly reading carries a confidence level (High / Medium / Low) based on article volume and signal consistency. Low-confidence periods are treated with more caution in the forecast model.

Rally-around-the-flag: When blame shifts outward, governments sometimes see a short-term approval boost as voters rally behind the incumbent. The ribbon around the trajectory shows the standard error of the mean framing signal: how confidently the daily average can be estimated given how many articles were scored and how much they disagreed. More articles and more consistent framing produce a narrower ribbon; fewer articles or split framing produce a wider one.

Forecast impact: This signal currently has no statistically significant effect on seat projections. The attribution modifier is set to zero pending further calibration against election outcomes. The tracker follows the signal in real time; the weight it receives in the forecast model will be updated as the relationship is validated.