Decomposing an Indian statute into obligations you can actually assign
How to turn an Act, its Rules, the state amendment and three circulars into atomic obligations with an owner and a deadline — including the judgement calls that decide whether the register survives.
Obligation decomposition is the least discussed and most consequential activity in compliance. Done well, everything downstream — ownership, control mapping, evidence, testing — becomes tractable. Done badly, you have an expensive index of section numbers.
Indian statutes make it harder than most, for reasons worth being explicit about.
Why Indian law is harder to decompose
The obligation is rarely in one place. A single requirement typically lives across the Act, the Central Rules, a state amendment, a notification prescribing the form, and a circular clarifying the interpretation. Decomposing from the Act alone produces an obligation that is incomplete and often wrong on the operative detail.
Applicability is threshold-driven and multi-dimensional. Whether an obligation applies depends on headcount, turnover, installed capacity, hazardous process classification, state, and sometimes the date of establishment. The same section binds one plant and not the one next door.
The form is the obligation. For a large share of Indian compliance, the real requirement is "file Form XYZ by this date in this state's prescribed format". The Act says maintain records; the operative obligation is a specific form with specific columns notified separately.
Amendments are frequent and quiet. Notifications amending rules are published continuously and are not always accompanied by a consolidated version. The current text of a rule is frequently not available anywhere as a single document.
What an obligation looks like when it is finished
For each obligation, capture:
| Field | Why it matters |
|---|---|
| Obligated party | Occupier, employer, principal employer, director, KMP — these are different people |
| Required action | The verb; what must actually be done |
| Trigger | Periodic, event-based, or condition-based |
| Threshold | Headcount, turnover, capacity, category — with the source of each |
| Applicable states | Where it applies, and how the variant differs |
| Deadline | Absolute date, or period from trigger |
| Prescribed form | The actual form number and its notifying instrument |
| Evidence | What demonstrates compliance to an inspector |
| Penalty and personal exposure | Amount, and whether imprisonment attaches to a named individual |
| Citation | Act section, rule, notification and state amendment |
The last two fields are the ones most registers omit and most users want.
Personal exposure in particular changes behaviour. An obligation that carries imprisonment for the named occupier gets attention that a monetary penalty on the company does not, and that information belongs on the record rather than in a lawyer's memory.
Four judgement calls
Granularity
Decompose until each item has exactly one accountable owner and one testable outcome. Stop there.
"Maintain prescribed registers under the Factories Act" is too coarse — different registers sit with different people. Decomposing each column of each register is too fine and produces a register nobody maintains.
The form problem
When a rule requires a filing, is the obligation "file the return" or "file Form 21 by 30 April in the state format"?
The second. The first cannot be tested and does not tell the owner what to do. This means obligations must be maintained against notifications that change the form, which is more maintenance and is the whole point.
State variants: fork or attribute?
If the state difference is a parameter — a different threshold, a different date — hold it as an attribute of one obligation.
If the state has prescribed a different form, a different register or an additional return, fork it. Trying to express a genuinely different Gujarat return as a parameter of the Central obligation produces something that reads correctly and cannot be acted on.
Definitions and thresholds
A large share of Indian compliance disputes turn on a definition — what counts as wages, what counts as a factory, who is a worker. Attach definitions to obligations; do not register them as obligations. But do record which definition version you applied, because it changes and the change is consequential.
Maintenance is the whole game
A register is accurate on the day it is built and decays from there. Three mechanisms keep it alive:
- Citation anchors that survive amendment, so a notification can be located against the obligations it affects rather than requiring a re-read.
- Forking on change, so the prior version and its assessments are preserved.
- Carry-forward proposals, so a reviewer confirms whether an existing assessment still holds rather than starting again.
Most registers we are asked to reconcile against source are between eighteen months and four years out of date. Their owners usually know, which is why they asked.
How long it takes
For a significant Act with its Rules and principal state variants, expect 200 to 700 obligations and four to ten analyst weeks with review. Automation reduces extraction time substantially and does not remove review — the judgement calls above are exactly where a model errs in ways that are invisible downstream.
Written by Rohit Menon, Principal Regulatory Analyst
Part of the team that builds and maintains the Regulens obligation library and platform. If you disagree with something here, we would genuinely like to hear it — get in touch.