Methodology, model, and publication changes. Prior records are not silently rewritten.
Added visible text-equivalent tables for every published flow, band, and company evidence-decomposition chart; published append-only corrections and scope-boundary policies; and added permanent build checks for those accessibility contracts without changing any estimate or feed.
Added per-lane run and release-SLA status, visible status and fallback notices, failure issues, versioned JSON/RSS events, deterministic social previews, exact dependency and vulnerability gates, and a conditional citation surface that remains absent until a real DOI is verified.
Status · Event feed · RSS
Added a reviewed TJX private-label-card lane using two dated Synchrony trust filings with program-level receivables; preserved methodology 1.2.0, left all 49 unaffected company result objects unchanged, and published the restatement under the versioned v2 assumption feed.
Added immutable, first-print-graded Monthly Treasury Statement predictions for federal deficit and total outlays; benchmark selection, empirical intervals, and live calibration remain separate for each series and never alter or blend with G.19 records.
Retired the NY Fed Household Debt and Credit source from the active auto-allocation assumption; reconciled FFIEC bank, NCUA credit-union, Federal Reserve Board G.19, and SEC Auto ABS evidence; separately versioned the resulting 50-company and headline-index restatement; and added computed source criticality with independent fallbacks and visible degraded-source bands.
Used the wide homepage's prediction rail for the live ledger count and existing historical backtest figures, while preserving explicit live-versus-backtest labels, mobile reading order, and the separate live-grade calibration record.
Separated the homepage claims into an ordered reading path, promoted the immutable prediction ledger, grouped 38 indistinguishable baseline-only Evidence Lift rows behind a native disclosure, and added a complete alphabetical company directory without changing any estimate or feed.
Compressed the heading scale so figures carry the hierarchy and added a server-rendered current-period flow view whose ribbon width represents estimated dollars and whose solid, dashed, and dotted styles expose evidence tiers without color dependence.
Published the first two live grades, prominently flagging that the revolving prediction missed direction and its 80% interval; added running live-only coverage and MAE comparisons without retraining or retuning the model.
Corrected the scoreboard compatibility status so every graded row reports GRADED, matching grade_status; the feed schema now states and tests that invariant.
Added Carvana's SEC-filed originated-auto-finance and ABS trust evidence, included CVNA in the covered-auto allocation, and published the derived Evidence Lift ranking under a non-breaking v2 company-feed schema.
Recomputed the 2026-Q1 50-company attribution from six Board first-print monthly flows and point-in-time SEC filing facts; 35 denominators advanced through comparable 10-Q facts and 15 remained explicitly on their annual baseline.
Expanded the evidence-first consumer-company universe from 10 to 50, published the dated 31-company exclusion ledger, and retained methodology 1.0.0 results for direct comparison.
Clarified seasonally adjusted monthly-flow units, added plain-language forecast orientation and primary-source links, and made scoreboard sorting a JavaScript-only enhancement.
Canonicalized attribution registry hashing and text assets so Windows and Linux cold clones publish byte-identical provenance metadata and site assets.
Published the first ten-company revenue-weighted DFRI estimate with 20,000-draw bands, tier shares, a public Assumption Registry, and company sensitivity rankings.
Published the immutable Federal Reserve credit-flow scoreboard and registered the scheduled prediction and first-print grading lanes.