rotascale

Solutions — sector

The order goes out in eleven milliseconds. The authority was checked first.

A trade is the rare action that cannot be undone by apologising. Front-office automation has always been governed by pre-trade limits; what has changed is that the thing hitting the limit now decides for itself what to send, and can explain its reasoning only in retrospect.

EU AI Act · MiFID II · SEC and FINRA rules · SR 11-7 Securities regulators and exchange supervisors clause map
Status
Algorithmic trading obligations apply regardless of AI Act classification
Applies to
Execution, research, surveillance, portfolio construction, client reporting

Pre-trade, or it is not a control

Every firm here already has post-trade surveillance, and every firm here knows what it is worth on the day something goes wrong: it tells you precisely what happened, after it has happened. The control that matters is the one in front of the action.

That is the whole shape of an authorisation call. Nine gates in a fixed order, cheap structural checks first, and the answer returned before the agent acts — not a callback, not an async review, not a report the next morning. An action refused at gate one never reaches policy evaluation, which is why the check can sit on the path at all.

Latency is a design constraint, not a footnote
One call before a consequential action, with the gates ordered so that the common refusals are the cheap ones.
Enforcement moves one rung at a time
Observe and shadow record without refusing, so a desk can see what the control would have done to a live book before it does it. Canary and enforce refuse. Nobody is asked to switch a trading system from nothing to blocking in one step.
An exhausted limit is not a denial
Six distinct outcomes, because "not permitted" and "permitted but out of allowance" send a desk to different places.

MiFID II and the algorithmic trading obligations

RotaGrant ships no clause map for MiFID II. Its algorithmic trading requirements are largely organisational — testing regimes, kill functionality, records of the algorithms deployed — and much of what they ask for is a description of how a firm runs, not a property software can evidence. What the platform contributes is the part that is evidence: what each agent was authorised to do, what it did, what stopped it, and under which enforcement mode. Whether that discharges an obligation is your determination and your counsel's.

The agents your teams are about to ship

Every one of these exists in some form already. What is new is the agent deciding which action to take rather than executing one a human chose.

  1. 1 authority Is there any authority for this?
  2. 2 status Is the grant active?
  3. 3 window Is now inside the grant's window?
  4. 4 scope Does the grant cover this action?
  5. 5 clean_context Was the context clean, where that is required?
  6. 6 bounds Do the per-action limits hold?
  7. 7 policy Does the policy on the grant permit it?
  8. 8 budget Is there room under the ceiling?
  9. 9 review Does this need a person?
Nine gates, evaluated in this order before the action. The lit ones are the gates that hold the four agents below — which is what differs between one industry and the next. The order does not.
Execution agent bounds

Works a parent order across venues, choosing child order size and timing against live conditions rather than a fixed schedule.

Consequential action
Sends an order to a venue
What goes wrong
Each child order is inside the limit and the sequence is not. The loop that behaves perfectly in a calm book is the one that runs away in a dislocated one.
What is recorded
The bound, the scale attempted, and the refusal — pre-trade, before the message left, not in a post-trade exception report.
Research and idea generation agent clean_context

Reads filings, transcripts and news, and produces the notes a portfolio manager acts on.

Consequential action
Publishes a recommendation into the investment process
What goes wrong
It ingests a document an interested party wrote to be ingested. Market abuse via an agent's context window is a new route to an old offence.
What is recorded
Which retrieved content was untrusted, at which step it entered, and whether the recommendation downstream of it stands.
Trade surveillance agent scope

Triages surveillance alerts, closes the explainable ones, and escalates the rest to compliance.

Consequential action
Closes a surveillance alert without human review
What goes wrong
An agent granted read scope for triage that quietly reaches write scope for disposition. A closed alert is the one nobody opens again.
What is recorded
Every disposition with the scope that permitted it, and the refusal of every action that exceeded it.
Client reporting agent authority

Assembles and issues periodic client reporting and suitability documentation.

Consequential action
Sends a regulated communication to a client
What goes wrong
A regulated communication issued under nobody's authority is still a regulated communication, and the firm still owns it.
What is recorded
The named individual whose authority was used, and what the client was sent.

Ambition is the point of these: none of them is a chatbot. Each is an agent taking an action with a consequence somebody has to answer for — which is exactly the moment a bounded authority stops being paperwork and starts being the reason the project is allowed to ship.

What your regulator will ask

Not hypothetical questions. These are the ones that arrive in writing, with a deadline, and the honest answer to most of them is a query rather than a project.

Reconstruct the order lifecycle, including what you did not send.
Refusals are first-class records, not exception logs. An order the agent was stopped from sending is evidence the pre-trade control worked, and it is the half most systems discard.
What were the pre-trade limits, and were they live?
The bound and the enforcement mode are sealed into each decision at the moment it was made. A pack states what was enforcing then — not what is configured today, which is the distinction that makes reconstruction possible.
Which model version produced this recommendation?
The model is inventoried and the version is in the record. The platform does not score or grade models — that is a different product — but a regulator asking what was deployed gets an answer.
Does this touch our order flow?
No payload reaches Rotascale on any code path. The platform runs single-tenant in your environment and calls no LLM anywhere; a test fails the build if a model provider appears in the dependency graph.

Identifiers this deployment verifies here

Every one of these is a checksum, not a shape. The claim is “this is a valid IBAN and the check digits agree”, not “this looks like one” — which is why the count on a screen can be cited rather than caveated.

FamilyHow it is verified
Bank account (IBAN) — mod-97 over the rearranged string
US bank routing number — ABA weighted checksum
US Employer Identification Number — issued prefix set
EU VAT number — per-state check digit where the state publishes one
Singapore UEN — entity-type check letter
Australian Business Number — modulus 89 after the documented subtraction

And what it does not look for. Names, addresses and dates of birth in free text carry no checksum, and finding them needs a model this deployment does not run. Neither are card numbers, medical record numbers, or national schemes beyond the three above. A clean screen means no verified identifier was found — never that the payload holds no personal data.

See it on your own problem

Bring an execution or surveillance agent. We will show you the pre-trade check on the path — the refusal before the message leaves, not the exception report the next morning.

A walkthrough is a working deployment with your front office and compliance in the room, not a slide deck — bring the agent you are most nervous about and we will put it behind a grant while you watch. If you would rather look first, the demo needs a work address and one click.