Compliance rule coding that reads every clause twice, tests every rule and cites it before it loads
Mandate rules live in a day, not two weeks, each read twice, tested and cited.
See one case, screen by screen ↓Compliance rule coding, also called investment compliance rule coding or guideline coding, is the work of turning the investment restrictions in an investment management agreement, its guidelines schedule, a prospectus or a side letter into rules the compliance engine runs before and after every trade. Compliance Rule Coding is investment compliance software that gives the rule analyst seven agents that find every restriction clause, draft each rule in plain English and engine parameters cited to its clause, read it a second time blind, test it and prepare the load. The analyst decides every interpretation, and a senior compliance officer who did not code the rule checks it.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
Compliance rule coding starts every time a mandate is won, a fund launches or an agreement is amended. A rule analyst reads the investment management agreement, its guidelines schedule, the prospectus or offering memorandum and every side letter, often in the client's own language, fills an annotation spreadsheet clause by clause, chases each defined term through the agreement and hand-codes every investment restriction into the order management system's compliance rules engine, for pre-trade compliance and the daily post-trade check. The side letter and the client's exclusion list arrive days later and restart the work.
The risk sits in single words. "Issuers in emerging markets" can mean domicile or country of risk; "5% per issuer" can mean the issuer or its whole group; "cash" may or may not include money market funds. Read the wrong way, the rule misses real breaches or blocks good trades for years. The checker often ticks the coded rule without the clause, few rules are back-tested on real holdings, and when a client, an auditor or an examiner asks how clause 3(2) is coded and tested, the trail is rebuilt from email.
Estimated split for a mandate of about 40 restrictions, coded by hand and with the solution. The sample firm's 2025 median by hand was 11 business days.
Seven agents read, draft, read again blind, test and load; the analyst decides and a different officer checks.
Pensionskasse Rheinhafen, a German pension fund, has awarded Quillfield Capital a €1,698.4 million (about $1.8 billion) segregated Global Equity Opportunities mandate, account RH01. The client funds on Monday, November 2; the rules are due live on Wednesday, October 28. Here is how its investment guidelines become investment compliance rules, screen by screen, in the working solution.
Nadia's Today screen shows 9 decisions waiting for her — 6 on Rheinhafen, 3 on rule changes — 23 mandates onboarding, 184 rule changes open with 32 due in 2 days, and 98.6% of live rules cited to a clause. The agents' morning brief reads like a conversation: the Reading agent agrees with the drafts on 37 clauses and differs on 4, and the Tester agent re-ran 64 rules overnight after a benchmark reclassification notice.
“People decide; agents never load a rule.”
Ravi Menon filed the package from document management: the Investment Management Agreement (64 pages), Schedule 2 Investment Guidelines (9 pages), a German side letter, the Nebenvereinbarung (6 pages), and the client's 212-issuer exclusion list, the Ausschlussliste. The mandate page walks the work in seven stages — documents, clauses, draft, second reading, tests, checker, load — with 47 clauses found, 3 side-letter overrides and 41 rules, each cited.
“The side letter overrides Schedule 2 in 3 places (§2: the side letter prevails). 3 clauses link to house rules already live.”
Replaying the run shows what each agent did. The Coding agent drafted 41 rules in 9 min 12 s, every element cited to its clause, median confidence 0.93. The Reading agent read all 47 clauses blind, on a different model, with no access to the drafts. The Mapping agent tied terms to data and matched 203 of 212 excluded issuers; the Tester agent ran 8 edge cases per rule and a 251-day back-test; the Load agent dry-ran 40 engine rules with 0 breaches.
“Hand-off to Nadia Brennan · 6 items need a person. Disagreements are never resolved by an agent. Nothing has been loaded.”
Side letter §3(2) caps issuers “mit Sitz in Schwellenländern” at 10% of portfolio value. Reading A, the Coding agent's draft, codes it literally — issuer domicile, country of incorporation — at confidence 0.71. Reading B, the Reading agent's, follows the side letter's §1(1) to IMA definition 1.14, Emerging Market Issuer by Country of Risk, at 0.93. The test bench shows today's transition plan at 8.7%, eight edge cases passing, and the engine parameters cited field by field.
Limit: “≤ 10.00% (‘nicht überschreiten’: 10.00% is allowed)”
One click on the definitions trail opens the source: side letter §1(1) in German, “Die in dieser Nebenvereinbarung verwendeten Begriffe haben die im Vermögensverwaltungsvertrag vom 9. Oktober 2026 festgelegte Bedeutung”, and the English beside it — terms carry the meaning given in the Investment Management Agreement of October 9, 2026. The trail runs from SL §1(1) to IMA 1.14, 1.9 Country of Risk and 1.17 Portfolio Value.
The English is the Intake agent's translation, checked by the Reading agent; the German stays the source.
The Tester agent ran both readings for 251 business days: domicile would have been over 10% on March 11–13, 2026, peaking at 10.4%; country of risk peaked at 9.1%. Four holdings explain the gap — two Cayman-incorporated China issuers count only under country of risk, real China exposure the literal reading misses. Nadia chooses Reading B with the reason code “Defined term governs” and sends it to Tomás Ferreira; a one-line client confirmation is drafted for Ravi Menon in German and English.
Coding agent: “I agree the defined term should govern. I have not changed the draft: the choice is yours.”
The client exclusion list, side letter §4(6), is matched by ISIN, then LEI, then name with the parent hierarchy. 203 of 212 issuers matched; 7 renamed or merged issuers have proposed matches at confidence 0.91–0.99, and 2 need a person — one delisted, one ambiguous name. The rule stays “Test failing” until they are resolved: a pre-trade compliance hard block and a daily post-trade check, with list changes applying ten business days after notice.
The Mapping agent never guesses an identifier below 0.90 confidence; it asks.
Nadia sends the 34 agreed, passing rules in one step and R-RH01-014 with her decision. Tomás opens it: the German clause and the English, the rule, Nadia's decision and reason, 8 of 8 edge cases, 0 days over in a 251-day back-test, and the version of the Coding agent that drafted it. He works a four-point checklist — clause read in its original language, defined terms applied, tests cover the clause, decision sound — then signs off or returns it with a reason code.
On a rule he coded himself: “You coded or decided this rule. Segregation of duties: another checker must sign it.”
Package RP-2026-1019-RH01 compiles in the engine's test environment as 40 engine rules and 1 manual control. The dry-run on the 64-holding transition plan shows 0 breaches and 1 warning: emerging markets at 8.7% against the 8.5% warning level. Against production it is 41 new, 0 changed, 0 retired. Nothing reaches production before a checker's signature on each rule and Helen Achebe's release; after the load the agent verifies the first pre-trade check and the first overnight batch.
“Only the Head of Investment Compliance releases packages over 25 rules.”
From the audit trail, one export collects the mandate's 47 source clauses in German and English, both readings for every rule, the analyst's decisions with reason codes, checker sign-offs, edge cases and 12-month back-tests, the engine dry-run and load results, and the agent and model versions. The pack — PDF report, CSV event log and JSON bundle — comes with a SHA-256 manifest, and the export itself is logged. It answers what an auditor testing investment guidelines compliance asks: which clause, which reading, who signed and what was tested.
“Every file is listed in a SHA-256 manifest so a regulator or auditor can prove nothing changed after export.”
The same mandate, seen by the five people who carry it from award to audit — what their week looked like, and what it looks like now.
Seven agents read, draft, read again blind, test and load; the analyst decides and a different officer checks.
Collects the mandate package from document management on award and starts the investment management agreement review: classifies each document, reads every page in its original language with English beside it, splits the clauses, builds the document hierarchy and finds where a side letter overrides the guidelines schedule. It watches for amendments.
Drafts each rule in plain English and in engine parameters — scope, numerator, denominator, data field, limit, warning and test timing — maps it to the rule templates and cites every element to its clause, with its confidence.
Reads each clause again, independently, on a different model family, following the definitions across documents, and writes its own rule. Agreement is shown; a difference goes to the analyst with both readings.
Maps defined terms to the data that measures them — country of risk, issuer group, ratings, vendor revenue data — and resolves exclusion-list names to ISIN, CUSIP and LEI.
Builds edge-case portfolios from the clause's own words, runs today's holdings and a 12-month back-test under each reading, shows the holdings that make the readings differ, and re-tests affected rules after every change.
Builds the engine package, dry-runs it in the engine's test environment, diffs it against production, loads after release, verifies the first pre-trade check and overnight batch, and rolls back on request.
Writes every step to the audit trail, assembles evidence packs with a hash manifest, back-cites legacy rules that have no clause, and runs the yearly re-certification of the library.
Decides; a senior officer who did not code it checks. The agents propose; a named person decides.
Ask about any mandate, rule, clause or change — the answer comes with its source, and the agents can act on it.
Which Rheinhafen rules still need a decision?
Six: four where the two readers disagree — R-RH01-014 emerging markets (domicile or country of risk), R-RH01-006 single issuer (issuer or group), R-RH01-009 cash (money market funds), R-RH01-031 thermal coal (second limb) — and two failing tests, R-RH01-027 derivatives (v2 to test) and R-RH01-033 exclusion list (9 identifiers).
Why did the agents disagree on the 10% emerging-markets limit?
The side letter says issuers “mit Sitz in Schwellenländern” — with their seat (domicile) in emerging markets. But its §1(1) gives its terms the IMA's meaning, and IMA 1.14 defines an Emerging Market Issuer by country of risk. Under domicile, the back-test shows 3 days over 10% (March 11–13, 2026); under country of risk, none.
Show the 3 breach days
Under domicile: March 11, 10.2%; March 12, 10.4%; March 13, 10.1%. Marisol Brands rose 18% after its results on March 10, lifting the domicile count. Under country of risk the same days were 8.9%, 9.1% and 8.8%.
Which live rules have no clause citation?
89 of 6,412 live rules (1.4%), all coded before 2022. The Evidence agent has proposed a citation for 83; the riskiest is R-10442, a legacy 5% single-issuer limit traced to the fund's prospectus, page 14. One, R-13902, has no clause at all and may be a house rule.
13 screens from the working solution, on its sample data. Pick one to see it large.
What counts as ready, tested and released are settings, changed only with a second administrator and logged.
| Setting | Default | Choose from |
|---|---|---|
| Both readers must agree for “ready” | On | On · Off; otherwise the rule goes to the analyst with both readings |
| Confidence floor for a reading | 0.90 | Below it, the reading is flagged even if both agree |
| Warning level | 85% of limit | Share of each limit that raises a warning |
| Back-test window | 12 months (251 days) | Holdings history the Tester agent uses |
| Minimum edge cases per rule | 6 | A rule with fewer is not “tested” |
| Release approval above | 25 rules | Packages larger than this need the Head of Investment Compliance |
| New mandate rules live | 3 business days before funding | Business days before funding |
| Rule change, from receipt to load | 5 business days | Analyst decision 1 business day from hand-off; checker sign-off 1 business day from sent |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample mandates and holdings · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (SEC enforcement actions as summarized in Sidley's 2025 fiscal year review of SEC enforcement against investment advisers: a $4 million penalty, October 21, 2024, and a $2.7 million penalty for issuer concentration of 30% to 89% against a 10% limit). People, firms, clients, mandates, funds and holdings named on this page are fictional — characters and sample data in the working solution.
Compliance rule coding, also called investment compliance rule coding or guideline coding, is the work of turning the investment restrictions in an investment management agreement, its guidelines schedule, a prospectus or a side letter into rules the compliance engine runs before and after every trade. Compliance Rule Coding is investment compliance software that gives the rule analyst seven agents that find every restriction clause, draft each rule in plain English and engine parameters cited to its clause, read it a second time blind, test it and prepare the load. The analyst decides every interpretation, and a senior compliance officer who did not code the rule checks it.
Yes. The Intake agent reads each document in the language it was signed in and puts the English beside it; the clause is shown in the original with the deciding words highlighted. Side letter compliance works the same way: in the sample mandate, a German side letter overrides the English guidelines schedule in three places, its §2 says the side letter prevails, and each of those rules cites the side letter.
The Tester agent builds at least six edge-case portfolios from the clause's own words — exactly at the limit, a Cayman-incorporated China issuer, a market reclassified mid-year — and back-tests 12 months of holdings under each reading. The Load agent then dry-runs the package in the engine's test environment and diffs it against production before anything is released.
The rule goes to the analyst with both readings side by side, their confidence, their citation trails and what each would have flagged in the back-test. An agent never resolves a disagreement; the analyst chooses, edits or returns the rule with a reason code, and the choice is recorded.
People decide everything that matters. The analyst decides each interpretation, a senior compliance officer who did not code or decide the rule checks it, and the Head of Investment Compliance releases packages over 25 rules. The agents never sign, never load a rule without a checker's signature and a release, and never change a signed record, and a settings change needs a second administrator.
Investment guideline monitoring does not stop at go-live. The agents find every rule the change touches, re-test it across the portfolios that carry it and bring only the results that change. In the sample, a benchmark reclassification re-tested 64 rules in 118 portfolios; 3 results change, and one fund would fall below its 85% emerging-markets name policy, so a notice to the portfolio manager is drafted.
Yes. Every rule carries its clause, both readings, the decision and reason code, the checker's signature, its tests and its load receipt. One export builds the evidence pack for a rule or a mandate with a SHA-256 manifest, and the reports feed the annual compliance program review under Advisers Act Rule 206(4)-7 and Investment Company Act Rule 38a-1.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your interpretation policy, a sample of past mandate packages and your compliance engine's rule set — and it goes live once every quality gate has passed. We will walk you through it on one of your own mandates first.
Of 412 overnight alerts, 9 reach a person, and every close cites its clause and approver.
1-min demoSEC Marketing Rule ComplianceEvery claim proven to its record, every local version checked, approved the same day.
1-min demoTrade SurveillanceEvery alert closed with a cited case file, and a second reviewer on every close.
1-min demoWe'll run Compliance Rule Coding on one of your own mandate packages — the IMA, its guidelines schedule and its side letters.