ESG governance assurance
Category
Advisory
Year
2026
A disclosure assurance engine for Indian listed companies. 2026, advisory engagement.
Authority is a property of the data, not a judgement about it. A finding you are asserting is not the same as a third-party claim you are merely surfacing. I moved that boundary into the data structure, rather than leaving it to a guideline, a review step, or a prompt.
Context
Indian listed companies publish sustainability disclosures that somebody has to assure. The assurance provider carries a specific legal exposure: a finding they are asserting is not the same as a third party claim they are merely surfacing. Blur that boundary and they own a statement they never made.
The thinking
Most systems police the boundary with policy: a guideline, a review step, a prompt instruction. All three depend on judgement holding on the day.
Stamped at acquisition
Regulatory filing portal
SEBI BRSR · MCA
Standards & assurance bodies
GRI / SASB · independent assurance
Press & investor research
reporting · short-seller notes
What I designed
Every document is classified the moment it enters, by the channel that fetched it, never by reading what it says. The tag is recorded at ingest. Inferred authority can be wrong. Recorded authority is a fact about where the document came from.
Source authority gate
The boundary is structural
The tag decides where a document is even allowed to be represented. Mandated and authoritative sources can resolve into an assertable finding. Reported sources can only be surfaced for a human. A low-authority claim has nowhere in the data to become a finding, so it cannot, regardless of what any model later decides.
Provenance by construction, not provenance by policy.
Against the company's filingsassertable
- Covered
Requirement met in the disclosure. Cited to the source.
- Thin
Partially addressed, the hard, valuable middle. Cited with the gap named.
- Absent
Required, not disclosed. A stated gap, never an accusation.
Against external contentsurface only
- Corroborated
External material aligns with the filing.
- Tension
External material contests the filing. Surfaced, sourced, routed, never auto-red.
- Orphan
External material with no requirement to attach to.
Six typed shapes
Every requirement-to-content match resolves into one of six named shapes. The register a shape belongs to is fixed by the boundary above. Three you can assert against the company's own filings, three you can only surface.
Related-party transactions · Companies Act s.188 / SEBI LODR
Filed disclosure. Issuer's RPT policy, thresholds and audit-committee approvals present; omnibus review noted.
Research note. Short-seller report alleges undisclosed related-party structuring through offshore entities.
The engine never converts a REPORTED allegation into an asserted finding. That decision was made at ingest, not left to a model at review time.
A Tension, handled correctly
The boundary earns its keep on the hardest case, where a company's own filing and a short-seller's allegation point in opposite directions. Anonymised here, because naming a real issuer beside a contested claim is precisely the move the design exists to prevent.
What the reviewer sees
Each requirement is decomposed to its elements and matched against the issuer's own filings. The output is not a score. It is a set of typed, cited findings a company secretary can act on. The interesting state is Thin: the requirement is partly met, and the system names exactly what is missing rather than passing or failing the whole thing. Market study, architecture, and this dashboard were the delivery.
Related-party transactions
- ✓Board-approved RPT policy with materiality thresholds
- ✓Audit-committee prior approval on record
- ✓Omnibus approvals reviewed quarterly
- ✗Arm's-length basis evidenced for the largest transaction
The largest related-party transaction is disclosed and approved, but the arm's-length justification the audit committee relied on isn't in the record. Surfaced for review, not marked non-compliant.
Board composition & independence
- ✓Independent directors at least half the board
- ✓Woman independent director present
- ✓Separate meeting of independent directors held
- ✓Declarations of independence on record
Requirement met against every element. Cited in full.
Vigil mechanism & whistle-blower
- ✓Vigil mechanism established
- ✓Direct access to audit-committee chair
- ✓Policy published and reachable
- ✓Affirmation that no person was denied access
Requirement met against every element. Cited in full.
Risk Management Committee
- ✓Committee constituted
- ✓Charter defines role and delegation
- ✓Met the required number of times
- ✗ESG / climate risk reflected in the risk register
Committee and charter are in place, but the disclosed risk register doesn't yet carry the climate/ESG risks the same report discusses in prose. A consistency gap, surfaced.
Managerial remuneration & ratio
- ✓Nomination & Remuneration policy disclosed
- ✓Ratio to median employee remuneration stated
- ✓Within s.197 limits, approvals shown
- ✓Performance criteria stated
Requirement met against every element. Cited in full.
Board evaluation disclosure
- ✗Manner of annual board evaluation disclosed
- ✗Criteria for evaluation stated
- ✗Outcome or action taken referenced
The annual board-evaluation statement is required and wasn't found in the filing. Recorded as a stated absence: a gap to close, not an allegation of wrongdoing.
The limit
Advisory only. I designed and specified the engine, I did not build or deploy it. Requirements and citations are synthetic, and no real issuer is assessed. There is no running system, and no performance figures are claimed. This is a design argument. Its worth is in the structure, not a benchmark.
