Markets · Agent pack
New packTrade surveillance monitoring
Cadre pack · Lookout
Applies named surveillance rules and raises episodes that warrant deeper review.
Demonstrated on representative cases. Source coverage, thresholds, policy interpretation, escalation design, and performance require institution-specific validation.
Operating dossier
The job, permitted decisions, and owner
Mandate
Examines one supplied account-and-symbol window with deterministic detectors for spoofing, layering, wash trading, marking the close, and insider timing, then records whether the evidence warrants an alert.
- Case arrives with
- Window · Account · Window · Symbol · Window · From · Window · To
- Allowed outcomes
- Monitor · Open Case · Refer To Tribunal
- Where it fits
- Trade Surveillance
Operating logic
- 01Examines a supplied window of an account's orders and trades with defined detection rules, then records whether the evidence warrants an alert.
- 02Assesses the named manipulation typologies — spoofing, layering, wash trades, marking the close, trading ahead of news — using recorded measures rather than an unsupported impression.
- 03Treats alert quality as part of the job: representative clean and legitimate-but-noisy windows should remain unalerted, with the reason recorded.
- 04Evaluation cases include legitimate look-alikes — a parent order worked in mostly filled slices, two-way volatility, and trading before immaterial news — so those facts alone do not establish misconduct.
- 05Evidence is id-level or it is not evidence; patterns that look real but are contestable go to adversarial adjudication rather than a unilateral verdict.
Decision logic
- 01Examines a supplied window of an account's orders and trades with defined detection rules, then records whether the evidence warrants an alert.
- 02Assesses the named manipulation typologies — spoofing, layering, wash trades, marking the close, trading ahead of news — using recorded measures rather than an unsupported impression.
- 03Treats alert quality as part of the job: representative clean and legitimate-but-noisy windows should remain unalerted, with the reason recorded.
- 04Evaluation cases include legitimate look-alikes — a parent order worked in mostly filled slices, two-way volatility, and trading before immaterial news — so those facts alone do not establish misconduct.
- 05Evidence is id-level or it is not evidence; patterns that look real but are contestable go to adversarial adjudication rather than a unilateral verdict.
- Must not
- Invent missing evidence or act outside the permitted outcomes.
- Human owner
- Ambiguous episodes and investigation ownership
- Evidence retained
- Structured outcome, observed path, and the source evidence used.
Case and record
One case, from supplied evidence to recorded decision
Supplied case
- Window · Account
- ACC-CLEAN
- Window · Symbol
- FLUX
- Window · From
- 1,782,147,540,000
- Window · To
- 1,782,148,500,000
View the case data
Case key clean-volatility
{
"window": {
"account": "ACC-CLEAN",
"symbol": "FLUX",
"from": 1782147540000,
"to": 1782148500000
}
}Evaluation and limits
Evaluation coverage and stopping points
Published evaluation map
Expected decisions, boundaries, and costly failure modes.
These cases show what the pack is asked to decide, when it should stop, and which plausible errors the evaluation is meant to expose. They are not a reliability score or independent validation.
Expected and borderline behavior
Routine judgments, close calls, and named stopping points.
FLUX two-way volatility, both sides fill, no self-match, no close burst, no news edge
no alert.
Case key
clean-volatilityACME layered non-bona-fide sells cancelled near the touch, high order-to-trade ratio, small genuine buy fills
spoofing.
Case key
spoofingSame ACME footprint assessed as layering
cancelled sells stacked across multiple price levels on one side.
Case key
layeringORBT concentrated buy ~6 minutes before a material FDA-approval headline, unwound after the pop
insider timing.
Case key
insider-timing
Boundary and misuse cases
Plausible inputs that could produce a costly or overconfident decision.
VOLT genuine large parent order worked in child slices that mostly fill, one repriced cancel away from the touch, low order-to-trade ratio. Superficially resembles layering; must NOT false-alert.
Case key
noisy-but-legit
Operating fit
Workflow placement and validation
Operating pattern
Monitor broadly, investigate selectively
Use broad monitoring to find episodes that merit attention, then reserve deep investigation for the ones that meet clear criteria.
Decision sequence and workflow placement
- 01Observe the stream
- 02Run named checks
- 03Assemble an episode
- 04Apply escalation criteria
- 05Hand off supporting evidence
Wider workflows
Institution-specific validation
These questions remain open until the pack is fitted to an institution.
Validation questions
- Policy and legal interpretation
- Source coverage, quality, and freshness
- Thresholds and exception calibration
- Integration and degraded-mode behavior
- Human approval and escalation design
- Performance on the institution’s own case mix
Inputs, actions, and outcomes
- Window · Account
- Window · Symbol
- Window · From
- Window · To
- Get ordersget_orders
- Get executionsget_executions
- Get bookget_book
- Get printsget_prints
- Get news windowsget_news_windows
- Signal spoofingsignal_spoofing
- Signal layeringsignal_layering
- Signal washsignal_wash
- Signal marking closesignal_marking_close
- Signal insider timingsignal_insider_timing
- Raise alertcommitraise_alert
- Monitor
- Open Case
- Refer To Tribunal