INTELLIGENCE LAYER · RESEARCH GRID
Research Grid
Ask up to 6 questions across up to 12 companies or issuers (48 cells maximum) and read the answers as a matrix. Every figure in every cell is traced back to the loaded row that carries it — a cell whose numbers cannot be traced shows as withheld, never as prose.
Sign in to run a research grid
The Research Grid asks the same diligence questions across a set of counterparties at once and returns a matrix where every figure is traced back to the loaded HMDA and Ginnie Mae rows that carry it — the shape counterparty and correspondent approval actually takes.
WHAT EACH PLAN CARRIES
AGGREGATE
Company search + profiles (HMDA summary)Market benchmarks (HMDA/Ginnie-derived)Ginnie issuer aggregate view (headline UPB/DQ per issuer)AI assistant surfaces (NL query → terminal, conversational dossier Q&A, platform semantic search) — the FREE HOOK: on for every tier including aggregate/free. NOT a legal gate; gated by budget + grounding in the AI gateway, not by counsel. (intelligence_layer.md §5.3)
SOLO
Vintage/geo/product drill-downs + peer comparison; Market Explorer / Compare tools; full historical depth (all years)Ginnie issuer intelligence detail (roster timeline, portfolio + DQ trends, buyouts)Ginnie pool explorer (ginnie_pool_monthly screens + issuer-to-pool drill-down) — BUILD-READY 2.9; Drill-down surfaceFHA compare-ratio scorecardExports (CSV/XLSX) — capped rows for drill_down, uncapped for loan_level (cap enforced in API, not here)GSE aggregate views (repurchase bps, everD90, credit profile) — ON for drill_down+ per owner directive 2026-07-22 (risk-accepted pending counsel sign-off; licensing caveat surfaced); Drill-down surfaceFair Lending analysis — ON for paid tiers (drill_down+) per OWNER DIRECTIVE 2026-07-24 ('drop the fair lending counsel and just do that'); counsel review waived under owner risk-acceptance. Public HMDA+Census only; every surface carries the 'statistical observation on public data, NOT a legal determination of discrimination' caveat. Free (aggregate) tier stays off — premium analytical feature.Watched-entity alerts + digests (FEAT-074): alert definitions, in-app inbox, alerts-scoped company lists, and Resend digests — first granted at the Solo rung (public HMDA/Ginnie/roster data only). R3-L16-D4: 'watchlists' was REMOVED from this description on 2026-08-03. W36 decoupled the watchlist to the free tier — /api/v1/watchlists calls requireEntitlement(null) (auth only) on GET/POST/DELETE and the landing page, /desk and the capability board all sell it as free — so this flag does NOT gate having a watchlist. It gates the SIZE of one and the alert machinery around it: alerts/limits.ts WATCHLIST_LIMIT is { aggregate: 1, solo: 50, drill_down: 50, loan_level: 250 }, enforced with a 403 LIMIT_EXCEEDED at POST. The stale wording here was faithfully transcribed onto the /alerts upgrade wall, which is why it is fixed at the source and not only in the copy.AI insight narratives (dossier outlier + peer-comparison prose, watchlist briefings, market commentary, anomaly explanation, plain-English alert authoring + smart digests) — paid: gated from drill_down up. NOT a legal gate.Advanced AI tooling (corporate-hierarchy enrichment, pattern-anomaly triage, prompt → target list, similar-entity/pool, prepay-speed + DQ early-warning captions) — paid: gated from drill_down up. NOT a legal gate.AI export generation (auto-generated dossier decks, report-from-prompt) — paid: gated from drill_down up, mirroring the exports ladder. NOT a legal gate.
LOAN-LEVEL
HMDA loan-level explorer — public data, sellable nowGinnie loan-level explorer (latest-month llmon snapshot, ~12.3M loans) — BUILD-READY 2.10; public disclosure data, sellable (unlike gse_loan_level; ToU review 0.3); required-filter no-scan rule enforced in APIGSE loan-level explorer — ON for loan_level tier per owner directive 2026-07-22 (risk-accepted pending counsel sign-off; licensing caveat surfaced); still excluded from API-key scopes (sessionOnly)
EVERY CELL IS ANSWERED FROM A SERVER-ASSEMBLED EVIDENCE PACK OF LOADED ROWS. CITATIONS ARE DERIVED BY MATCHING EACH FIGURE BACK TO ITS SOURCE ROW — THEY ARE NOT WRITTEN BY THE MODEL.