Credit limits, from appetite to enforcement

Credit appetite is recorded per legal entity, and every field is a decision somebody has to own. Nothing is pre-filled, because a defaulted credit tier looks identical on screen to a considered one. The calculator refuses to produce a limit until the inputs exist, and names the ones that are missing rather than quietly assuming them.

• Recorded per legal entity, not per trading account
• Each field labelled with the cap it feeds, so a limit can be explained
• Common market conventions offered as starting points, never as authority
• An unusual choice is visibly unusual rather than hidden in a number box

The workbench sizes a limit two ways over one engine: on what a prospect commits to post and intends to trade, or on what an onboarded entity actually holds. It returns four caps — credit, collateral, liquidity and concentration — and takes the smallest. The number is not the useful part: a limit of $31M means something very different when credit would have allowed $52M and liquidity would not.

• Declared mode for prospects with no book yet, book mode after onboarding
• Credit, collateral, liquidity and concentration caps computed side by side
• The binding cap named explicitly — that is the thing to negotiate
• Collateral valued at the same marks and haircuts the risk engine uses

Coverage requests, credit reviews, risk approves, operations confirms the platform actually took it. Every step is appended with its actor, and the approver may not be the requester. That single control is what makes the rest of the record worth keeping.

A free-text credit note throws away the three properties that make evidence reviewable — how strong it is, when it was established, and who attested it. The credit file records five pillars instead: four attested by a person, one measured from the platform’s own record of how the client has actually behaved. Every pillar is shown whether or not it has evidence, because a missing pillar is the most important thing on the page and nobody notices an absent row.

A limit approved in a calm market is a different quantity of risk once volatility and depth have moved, and most desks discover that at the worst possible moment. Every approved limit sits beside the limit today’s data would produce, sorted worst-divergence first, with a queue for the ones that are unenforced or overdue for review.

The workflow spans several screens, each a good view of its own step and silent about the others. The pipeline inverts that: it computes the stage for every entity — blocked, ready, in review, live but unenforced, live, or stalled — and gives you the single next action as a button that lands on the right page with the entity already selected.