Deal Book scoring methodology

Deal Book scoring on Stonewake ranks financeable opportunities with an explainable 0-100 relevance score built from six weighted components, with verbatim citations on customer-visible claims. Weights are tenant-tuned to a desk's mandate. The score is an origination relevance aid, not a credit grade and not a regulatory capital measure.

Purpose of Deal Book scoring

Export finance, project finance, and commercial real estate origination teams face more public signals than a desk can manually triage. The Deal Book organises candidate items into a ranked pipeline and attaches an auto-drafted origination memo labelled as AI-drafted. Scoring answers a narrow question: how relevant an item is to the tenant's desk filters right now.

Scoring does not answer whether the obligor will pay, whether an export credit agency will offer cover, or whether a project finance DSCR will hold through operations. Those judgments remain with underwriters, ECA partners, and credit committees.

Explainable components

Deal Book scoring is built from six weighted components: stage, ECA angle, ticket, country, sector, and recency.

Explainability means a desk can see which components drove the 0-100 figure. Weights are visible and adjustable at tenant level, so an EF desk, a PF desk, and a CRE desk on the same platform can apply different weight profiles to the same underlying signals. Exact feature engineering, model internals, and data-provider contracts are not part of this methodology.

Country context inside scoring sits beside the country risk composite and institutional country risk materials, including the OECD country risk classification used for Arrangement premium floors. ECA angle reflects whether public materials suggest an official-support pathway worth desk attention; it is not an ECA commitment.

Evidence attached to ranked items

Every customer-visible claim in the Deal Book carries a verbatim quote verified against the live source page, under the same citations and evidence standard used across the platform. Pagination and homepage URLs that lack the quoted text are not acceptable citation targets.

That evidence layer is what makes a ranked list usable inside a regulated bank. A bookrunner or originator preparing a syndication conversation still needs a source trail. An AI-drafted memo is a starting artefact, not a filing.

Reading the 0-100 score

Higher scores indicate a stronger match to tenant-weighted relevance components, not a lower probability of default. Two items with similar scores can differ sharply in credit quality. Recency can elevate a weak credit story that happens to be newly public, so novelty and bankability are separate questions. Ticket and stage components bias toward size and maturity signals present in sources; they do not substitute for term-sheet reading.

For EF desks evaluating buyer credit pathways, Deal Book items sit beside ECA product knowledge in the export credit agencies and export finance hubs. For PF desks, items sit beside security and cover-ratio concepts in the project finance and cover ratios and covenants hubs. For CRE desks, items sit beside property metrics in the commercial real estate lending hub and the glossary's property-metric definitions.

Human confirmation and commercial boundaries

Deal Book scores are drafts for analyst review. Stonewake operates with a human in the loop and is not a credit scoring institution under banking regulation; Deal Book scoring is not filed as one. Customer bank names are not part of methodology disclosure.

Commercial pages that describe desk outcomes without repeating scoring maths include export finance market intelligence, project finance origination intelligence, and CRE credit intelligence.

Relationship to monitoring and screening

Monitoring briefs and adverse media screening results can inform whether a Deal Book item still belongs in active coverage, but scoring and screening are separate methodologies. A high relevance score does not clear financial-crime concerns, and a clean screening disposition does not make an item relevant to a desk's ticket appetite.

Berne Union members and official ECAs publish market statistics and product information, including the Berne Union's own state-of-the-industry reporting, that desks already use for institutional context. Deal Book scoring does not replace those publications; it ranks public deal-shaped signals for a tenant's pipeline review.

Tenant weight design

Tenant-tuned weights exist because an EF desk that prioritises ECA-relevant mid-ticket opportunities differs from a PF desk hunting early-stage infrastructure or a CRE desk focused on refinance-stage signals. This methodology does not publish default weight tables or training data; it states that weights are visible and adjustable at tenant level, which is the explainability commitment.

Weight changes on a desk correspond to a documented mandate shift in the same way rating-model overlays correspond to a documented credit view; an unexplained weight change carries the same governance ambiguity as an unexplained rating notch.

Memo drafts and labelling

Auto-drafted origination memos are labelled as AI-drafted so readers do not mistake them for final credit papers. Citations inside memos obey the same evidence standard as the rest of the Deal Book: each material claim, including ticket size, parties, or location, remains traceable to a verbatim quote.

Excel export lets a desk take the ranked book offline for further review. Export is a distribution feature; it does not create a new scoring methodology.

Cross-desk calibration

A banking group running export finance, project finance, and CRE origination on one Stonewake tenant can still use separate weight profiles per desk. Deal Book scoring is relevance to a desk mandate, so a shared weight set can bury CRE refinance items under EF ECA-angle noise or the reverse. Weight profiles function as desk configuration artefacts with owners and review dates in the same way other underwriting parameters do.

Rank movement between weeks can follow a weight change, a spike in source volume, or a country-component move after a public indicator update. Rank movement without a traceable cause is the governance signal worth attention; the methodology's answer is explainability of components plus tenant-visible weights, not a promise of stable ordinals in a changing public-information environment.

Commercial solution pages translate these mechanics into desk language for export finance market intelligence, project finance origination intelligence, and CRE credit intelligence without exposing feature code.