TECHNICAL DOCUMENT · BFA-WP-01 · CREDIT RISK SCORING

The R-Score: Calculation Methodology

How Black Flag Alert scores the credit risk of companies registered in England & Wales. This document describes the inputs, the component scoring, the composite model, the overrides, and the validation evidence. It matches the production scoring pipeline.

Coverage
~3.4M scored companies
Output
Score 0–100 · Rating 1–10
Model validation
R² = 0.878
Data sources
Primary public records
Cost to user
£0

§0Abstract

The R-Score is a company credit-risk measure built entirely from public records. It reads the filed accounts of a company, converts them into thirteen financial ratios, scores three components (profitability, asset quality and funding) and combines them with a nonlinear composite model. The result is a 0–100 score, mapped to a 1–10 risk rating, and adjusted by hard overrides for insolvency events that outrank any accounting figure.

The model was validated against an established commercial H-Score credit-risk benchmark: R² = 0.878, with 94% of predictions within 20 points. Around 3.4 million England & Wales companies carry a score. The service is free, and this document is published because a risk score you cannot audit is marketing, not intelligence.

§1Scope & coverage

The score covers companies registered in England & Wales. Around 5.1 million are registered, of which approximately 3.4 million carry a score. Companies registered in Scotland (numbers beginning SC), Northern Ireland, or overseas are not covered.

A company is scored from its most recent filed accounts. A company without enough usable filed data shows as Unscored on the profile; its court and insolvency records are still displayed. Scoring is a data gate, not a paywall. There is no paid tier.

§2Inputs & sources

Four primary public sources feed the system. None of the inputs are resold bureau data.

SourceWhat it providesUsed for
Companies HouseFiled accounts (iXBRL), filing deadlines, company status, registered detailsThe financial core: all thirteen ratios; overdue-filing signals
HMCTS court recordsCounty Court JudgmentsThe CCJ register on each profile; risk-rating adjustments
The GazetteWinding-up petitions, insolvency notices, dissolution eventsThe hard override (§6) and event timeline
Charges registerMortgages and security granted over company assetsSecured-debt picture on each profile

§3Extraction & eligibility

Filed accounts arrive as iXBRL documents. The extraction pipeline reads each filing and pulls structured fields: turnover, operating profit, profit before tax, net assets, cash, borrowings, creditors, stock, debtors and the audit flags. From these raw fields the pipeline computes thirteen ratios: operating margin, gross margin, ROCE, return on net worth, return on assets, interest cover, current ratio, acid test, cash ratio, gearing, debt-to-equity, leverage, and debtor/creditor days.

Two eligibility gates decide whether a company gets a score:

Reduced-mode accounts
Filleted and micro-entity accounts omit the profit and loss statement. For these, the profitability component defaults to a neutral 50 and the score is driven by the balance-sheet components. The profile labels the account type so the reader knows which mode applied.

§4Component scoring

Each ratio is converted to a 0–100 sub-score by a piecewise curve fitted from observed company data: strong readings score high, weak readings score low, with thresholds where distress typically begins. The sub-scores are then combined into three components with fixed internal weights. Within a component, weights renormalise over whichever metrics are available, so a missing ratio does not distort the result.

Profitability — component weight 42%

Whether the trade itself makes money.

MetricWeightWhat it measures
Operating margin0.30Profit left from each £1 of sales after day-to-day costs
Pre-tax margin0.20Profit before tax relative to turnover
ROCE0.20Return on capital employed
Return on net worth0.15Return generated on shareholders' funds
Return on assets0.15Return generated on total assets

Asset quality — component weight 38%

Whether the balance sheet can meet its obligations.

MetricWeightWhat it measures
Current ratio0.35Short-term assets against short-term liabilities
Acid test0.25Current ratio excluding stock, the stricter liquidity test
Cash ratio0.20Cash alone against short-term liabilities
Debtor days0.10How long customers take to pay (lower is better)
Creditor days0.10How long the company takes to pay suppliers

Funding — component weight 20%

How much the company leans on borrowed money.

MetricWeightWhat it measures
Gearing0.40Borrowings relative to shareholders' funds
Debt-to-equity0.30Total debt against equity
Interest cover0.30How many times profit pays the interest bill (below 1x: profit does not service the debt)

§5Composite model

The three components combine in a nonlinear model. A simple weighted average would let one strong component hide two weak ones; a company with healthy profit but a collapsing balance sheet should not score like a healthy company. The composite therefore carries two corrective terms: a geometric mean that rewards balance across all three components, and a minimum-component penalty that drags the score down when any leg is weak.

H = 0.289·P + 0.068·A 0.055·F 0.685·min(P,A,F) + 1.445·∛(P·A·F) 0.574 where P = profitability, A = asset quality, F = funding (each 0–100) result clamped to 0–100

The linear weights come from the regression fit against the benchmark; the geometric-mean term carries the positive contribution of balanced funding. Two further behaviours are built in:

§6Overrides & guards

Four rules correct the pure-arithmetic result where arithmetic would mislead.

6.1 Negative equity → funding forced to zero

If net assets are zero or negative, liabilities exceed assets: the company is technically insolvent. The funding component is set to 0 regardless of what the ratios produced, because gearing ratios become meaningless once equity is wiped out.

6.2 The asset-cushion floor

The minimum-component penalty would otherwise push an asset-rich company with one catastrophic loss year to the bottom of the scale. That overstates the danger if the equity cushion can absorb years of losses at the current rate. When all of the following hold (net assets positive, current ratio at least 1.0, asset component at least 35, and net assets at least three times the annual loss), the score is floored between 15 and 22 depending on loss runway (three years of runway floors at 15, eight or more at 22). The company reads as High Risk (watch closely), not Distressed.

6.3 Winding-up petition → rating forced to 10

An active winding-up petition means a creditor has asked the court to liquidate the company. No accounting figure outranks that event: the risk rating is set to 10 (Distressed) whatever the accounts say. Other terminal insolvency events apply the same override at the API layer, alongside status, CCJ and overdue-filing adjustments.

6.4 Data-quality score

Every profile carries a separate data-quality score so the reader can see how much the filing supports the number. It starts from the share of the thirteen metrics that were extracted, then takes penalties: −10 for a going-concern uncertainty flagged by the auditor, −5 for a qualified audit opinion, and up to −15 where extraction was thin.

§7Risk rating scale

The 0–100 score maps to a 1–10 rating in ten-point bands, with the top band narrower to separate the strongest companies. A score of 25 or below also flags the profile as a warning area.

Score rangeRatingLabel
85–1001Very Low Risk
75–842Very Low Risk
65–743Low Risk
55–644Low Risk
45–545Moderate Risk
35–446Moderate Risk
25–347High Risk
15–248High Risk
8–149Critical Risk
0–710Severe Risk

§8Sector context

Raw scores are only half the picture, because normal financial structure differs by trade. A thin operating margin is routine in distribution and alarming in software. After scoring, each company's score is ranked against other companies in the same sector group and published as a sector percentile on the profile. The percentile answers the question the raw score cannot: is this company weak for its trade, or is the trade itself thin?

§9Validation

The composite model was validated against an established commercial H-Score credit-risk benchmark across a sample of companies with known outcomes:

MeasureResult
Coefficient of determination (R²)0.878
Predictions within ±20 points of benchmark94%
Simpler linear variant (for reference)R² = 0.828

The nonlinear composite outperforms the linear variant precisely because of the balance terms described in §5: real failures are rarely uniform across all three components.

§10Limitations

Transparency
The core of Black Flag Alert is free under a freemium pricing model. Every profile shows its component scores, the number of metrics used, the data-quality score and the account type, so any reader can see how much the filing supports the number.

§AQuestions

Is a higher R-Score better or worse?
Higher is better on the 0–100 gauge. The paired risk rating works the other way: 1 is the best (lowest risk) and 10 is the worst.
Why does a company show as Unscored?
The company has not filed enough usable accounts to pass the extraction and metric gates. This is common with very new companies and dormant ones. Court and insolvency records still display on the profile.
Why does the model use a geometric mean?
The geometric mean rewards balance. A company strong in all three components scores higher than one with the same average but a weak leg, which matches how companies actually fail.
How often is the score updated?
The financial component updates when new accounts are filed and processed. The live signals (CCJs, winding-up petitions and overdue filings) are monitored continuously, and the hard overrides apply as soon as those events are detected.
Can a company improve its R-Score?
Yes. Improve the underlying signals: file accounts on time, reduce gearing, pay or satisfy CCJs, strengthen profit and net assets. The score follows the public record, so changes appear as filings and records update.