Governance & Forensics: India

Phase 11 · Elective · ~16 hours · Prerequisite: M2.08 (Forensic Case Files); this is its governance sequel

Roughly 347 entities. The great bulk of the debt in unlisted SPVs, the holdco itself unlisted, and a disclosure surface almost entirely off the exchanges. That was IL&FS, and an analyst reading the consolidated accounts saw almost none of it.

Nothing in those accounts had to be false for that to happen. The numbers can be true, audited and clean, and the value can still have left through a door nobody was watching.

Where it sits. M2.08 taught you to catch a lie inside the financial statements: fake revenue, parked costs, the accrual wedge, the cash-yield test. This elective points the same discipline at the layer above the statements: ownership and control. In India's promoter-dominated market, the most valuable analytical skill is not spotting a bent estimate; it is reading the map of who owns what, who transacts with whom, whose shares are pledged to whom, and where a listed company's cash can quietly walk to a family entity next door. This is the second India-specific forensic lab (the first was the India ten inside M2.08), and it is built around seven skills a working Indian-equity analyst uses every week: promoter-group mapping, related-party-transaction (RPT) webs, share-pledge mechanics, the subsidiary/loan "hopscotch," auditor-signal reading, the SEBI LODR Regulations 23 and 24 that govern all of it, and the public-record discipline that keeps your write-ups both useful and safe. The lens is the Indian rupee (₹); a United States dollar ($) contrast runs alongside every idea, because control-based looting is a global pattern with only local plumbing.

The one-sentence version. A promoter who controls the board, pledges the stock, and trades with the company through a web of private entities can extract minority shareholders' value entirely within the law's grey zones. The analyst's job is to map the plumbing, price the leakage, and grade the governance using only what the public record actually says.

The companies we carry. Almost every worked number below belongs to a synthetic company built for the lesson: Meridian Textiles, the Suvarna Group, Konark Power, Trishul Holdings, Nilgiri Foods, and (in dollars) Cascade Holdings. Each is labelled "(synthetic)" the first time it appears, and each is internally consistent so you can recompute it. The real names that appear (Satyam, CG Power, Fortis/Religare, Zee/Essel, DHFL, IL&FS, Yes Bank, Vakrangee, Manpasand in India; Adelphia, Enron, WorldCom, Tyco in the US) are used only for facts that a regulator found, a court adjudicated, or a company itself disclosed. Every such statement is tagged "verify," because statuses change and your duty is to check the primary document before you rely on it.

Suggested hour budget. Core content (read with the regulations open) 6h · worked examples 2h · practice set 3.5h · the applied mini-project (a full promoter-web teardown of a synthetic group) 3h · flashcards + mastery check 1.5h.

App link. This module seeds the Governance & Red Flags branch's promoter-web tier: the RPT-mapping canvas, the pledge margin-call simulator, and the LODR threshold checker. Its capstone case is CL-11.4 "The Promoter's Web at Suvarna Group," which auto-scores your RPT map, your Regulation-23 threshold tests, and your governance rating against a seeded key.

A note on facts and fairness: the absolute rule here. You are about to learn how value leaks from listed companies to the people who control them. Learning it well means naming mechanisms precisely, and naming them precisely creates real defamation risk when applied to real, living businesses and families. So the rule here is stricter than anywhere else in the program: only adjudicated or regulator-published facts about real entities may be stated as fact, always with attribution to the finder and always as a dated fact you check against the order itself; everything else is synthetic and labelled, or it is described as evidence I cannot reconcile rather than as a verdict. This is not timidity. It is the difference between an analyst whose work a fund can act on and a blogger whose work a court can enjoin. You will practise the language protocol until it is automatic, because it is both an ethical duty to the people you write about and a professional necessity for the people who rely on you.

Learning objectives

By the end you can:

1. Build a promoter-group map from public documents alone (the shareholding pattern, the annual report's related-party note, MGT-7 and BEN-2 filings on MCA21, the DRHP, and stock-exchange disclosures), showing the listed company, its subsidiaries and step-downs, the promoter and "promoter group," and the significant beneficial owner (SBO) behind any layered structure.
2. Catalogue and quantify the RPT web: identify the nine standard leakage channels (off-market sales/purchases, soft loans and advances, royalty/brand fees, corporate guarantees, asset transfers, lien-marked deposits, management fees, investments in group entities, and revenue-that-is-really-a-loan), and put a rupee number on the annual value leakage as a share of PAT and net worth.
3. Do the share-pledge math: compute loan-to-value (LTV), cover, the margin-call trigger price, the percentage decline that fires it, the top-up required, and the invocation cascade, then explain the reflexive spiral that turns a pledged promoter stake into a fuse whose length is printed every quarter.
4. Trace the subsidiary/loan "hopscotch": reconcile standalone-versus-consolidated debt to size "dark debt" in unlisted entities, follow a loan through layered subsidiaries to a promoter beneficiary, and apply the Companies Act limits (Sections 185/186) and the round-tripping test.
5. Apply SEBI LODR Regulation 23 end-to-end: compute the RPT materiality gate (10% of consolidated turnover up to ₹20,000 crore, then a banded schedule capped at ₹5,000 crore), apply the aggregation clause that defeats salami-slicing, and state who must approve (audit committee for all RPTs; shareholders for material RPTs, with related parties abstaining), plus the US Regulation S-K Item 404 contrast.
6. Apply SEBI LODR Regulation 24 to subsidiary governance: distinguish the Regulation 16 "material subsidiary" (10% of consolidated income or net worth) from the Regulation 24(1) threshold that forces an independent director onto the board (20%), and identify when a disposal needs a shareholder special resolution.
7. Read auditor signals in rank order: distinguish a scheduled rotation (Section 139) from a high-conviction mid-tenure resignation; grade Key Audit Matters (SA 701), Emphasis-of-Matter paragraphs, and qualified opinions; and locate the CARO 2020 fraud and fund-routing clauses and the Section 143(12) fraud-reporting trigger, plus the US CAM / Form 8-K Item 4.01 contrast.
8. Produce a defensible governance rating with a weighted rubric, state the specific disclosures that would change it, and write every finding in the language protocol so it distinguishes what a regulator found, what an agency alleged, and what you as an analyst can only describe as unreconciled evidence.

Prerequisites & connections

Builds on. M2.08 most directly. Its seven-plays taxonomy, its language protocol, its severity ladder (L1–L4), and its India cases (Satyam's Maytas attempt, CG Power's advances, Zee's fixed deposit, DHFL's routing, IL&FS's 347 entities) are the raw material systematised here into a governance method. From Phase 1: M1.09 (consolidation, and the parent/subsidiary boundary that "hopscotch" exploits) and M1.10 (the auditor's report, CARO, Schedule III). From Phase 2: M2.06 §4.9 (the promoter-leak play) and M2.07's NBFC/ALM panel (the funding-mismatch engine behind IL&FS and DHFL). From the ratio toolkit: M2.02 (DSO, and the receivable-that-is-a-loan) and M2.04 (cash conversion).

Feeds into. C7 (management and governance judgment) as a graded competency; M5 sector playbooks (banks and NBFCs get the group-exposure and divergence panels; every sector gets the pledge and RPT screens); M8.02's ten-minute teardown (the governance pass compressed to five documents); and E11.02 (M&A & Deals Lab, where control premia, open offers under SAST, and squeeze-outs turn these same structures into deal mechanics). The governance rubric you build here becomes a permanent input to every teardown from Phase 5 onward.

The one-sentence bridge from M2.08. There you learned that fake revenue needs a fake asset; here you learn that real revenue, in a promoter-controlled company, needs a governance audit. The fraud that survives a clean audit is the one where the numbers are true and the value simply leaves through a related door.

Two nodes cover governance and they cover different things, so the seam is worth stating before either dilutes the other. EC1.07 owns the taxonomy. That means the organisational forms and why liability, tax and disclosure differ across them, the agency problem and the Jensen-Meckling decomposition of its costs, board structures and committee design, the stakeholder map, and the integration of environmental, social and governance factors into a valuation. It is the general theory, and it is where a definition belongs. This module owns the forensics, which is a different craft with a different output. You map a promoter group from primary filings, price the leakage through a related-party web, run the Regulation 23 and 24 threshold tests, follow a subsidiary loan through the hopscotch, compute a pledge cascade on both bases, and read the auditor-signal sequence. Where EC1.07 asks what good governance is, this asks what a specific promoter did last quarter and what it is worth in rupees. Take the taxonomy there and bring it back here, and do not expect a definition to do a forensic's work.

A word on jurisdiction and currency (R6). India runs on a promoter/"promoter group" concept, quarterly pledge disclosure, and the SEBI LODR + Companies Act stack; the US runs on dispersed ownership, dual-class voting, proxy-statement disclosure, and the SEC/PCAOB stack. The mechanisms of control-based extraction are identical; only the disclosure surface differs. Every core idea below is taught in ₹ and mirrored in $.

Core teaching content

4.1 The public-record discipline and the language protocol

Everything here is built on documents any member of the public can obtain, and on a discipline about what those documents let you say. Internalise the discipline before the mechanics, because the mechanics are dangerous without it.

The source hierarchy (what each document proves). Rank your sources by what they can establish:

1. Adjudicated findings: a SEBI order, an NCLT/NCLAT ruling, a court judgment, a company's own disclosed restatement or board-committee report. These establish fact, attributed to the finder: "SEBI's order found…", "the company disclosed…". Highest evidentiary weight.
2. Regulator-published data: the shareholding pattern and pledge disclosure (LODR Regulation 31; SAST Regulation 31), the RPT half-yearly disclosure (Regulation 23(9)), MCA21 filings (MGT-7 annual return, AOC-4 financials, BEN-2 beneficial-ownership, DIR-12 directors, CHG charge-creation), rating rationales, and the audited financial statements and their notes. These are primary facts you compute from. The arithmetic is yours; the inputs are the company's own filings.
3. Allegations by an investigative agency: the ED, CBI, SFIO, or Serious Fraud offices. These are allegations, not findings, until a court adjudicates: "the ED has alleged…; trial ongoing (as of mid-2026)." Real weight as a signal, zero weight as a verdict.
4. Journalism and activist-short research: sometimes the earliest and best primary field work, sometimes wrong. Treat as reported, check the underlying documents yourself, and cite the document, not the headline.

The language protocol (use it in every sentence you write about a real company). State facts with the finder attributed. Mark allegations with their source and status. Reserve the word fraud for adjudications and confessions. For your own analysis, describe evidence, not verdicts: write "the promoter's pledged holding rose from 22% to 81% over four quarters while related-party advances grew to 21% of net worth — a combination I cannot reconcile with minority-shareholder value being protected," a sentence that is true, useful, and defensible, rather than "the promoter is looting the company," which is a verdict you have not earned and a lawsuit you have invited.

Why this is not optional. In a promoter-controlled market the most important facts (control, pledge, RPT terms) concern named, living people and families. Indian defamation exposure is both civil and criminal. An analyst whose language separates finding from allegation from unreconciled evidence produces work a fund can publish and act on; an analyst who asserts verdicts produces work that gets retracted, and gets the desk sued. The protocol is also simply more precise: "I cannot reconcile X with Y" states exactly what you know and exactly what you don't, which is the whole value of an analyst.

The public record is deeper in India than analysts assume. MCA21 lets you pull the financials and shareholder/charge filings of unlisted group entities for a nominal fee. That is how you see the "dark debt" and the promoter-entity accounts the listed filings hide. SEBI orders are indexed and free. The shareholding pattern discloses pledge quarterly. The gap between what is available and what analysts actually read is where your edge lives.

4.6 Share-pledge mechanics

A promoter who has pledged the controlling stake has converted a governance actor into a leveraged one. Leverage on the control block is the most mechanical, most quarterly-disclosed risk in Indian equities.

What a pledge is. A promoter borrows against shares (a loan-against-shares, LAS) from a bank, NBFC, or mutual fund. The shares are the collateral. India discloses this unusually well: the shareholding pattern shows the pledged/encumbered percentage every quarter (LODR Regulation 31), and SAST Regulation 31 requires the promoter to disclose creation, invocation, and release of encumbrance. Post-2019, "encumbrance" is defined broadly to include any restriction on disposal or on the exercise of voting rights, catching non-disposal undertakings and similar side arrangements.

The math: LTV, cover, and the margin call. The lender lends a fraction of the collateral's market value:

• LTV (loan-to-value) = loan ÷ pledged market value. Cover = pledged market value ÷ loan = 1 ÷ LTV. A 50% LTV is 2.0× cover.
• As the share price falls, the collateral shrinks while the loan is fixed, so cover falls and LTV rises. The lender sets a maintenance trigger, a minimum cover (say 1.5×), below which it issues a margin call demanding a top-up of cash or additional shares.
• The trigger price is where pledged value hits trigger cover × loan. The percentage fall to trigger is the distance from today's price to that level. If the top-up is not met, the lender invokes the pledge and sells the shares into the market.

Always compute the pledge on two bases, because the regulation does. The shareholding pattern's headline figure is the encumbered percentage of the promoter's own holding, and that is the number every screen prints. It is the wrong base for two of the three questions you are asking. Convert it to a percentage of total share capital as well, by multiplying by the promoter's stake. A 75% pledge on a 60% holding is 0.75 × 0.60 = 45% of the company, and 45% is the figure that tells you how much stock hits the market on invocation, and how far above the SAST open-offer thresholds a single buyer of that block would land.

SAST Regulation 31(4) uses the same two bases and takes whichever binds first. The promoter and the persons acting in concert must disclose detailed reasons for the encumbrance once the combined encumbrance reaches 50% of their shareholding or 20% of the total share capital, whichever is lower (check the current text and any amendment). For a 60% promoter the total-capital limb binds first, because 20% of total capital is only 0.20 ÷ 0.60 = 33.3333% of the holding, well below the 50% limb. So a promoter can be under the holding-based threshold, look unremarkable on a screen, and still owe the market an explanation. Read both numbers, every quarter, and write both into the memo.

The reflexive spiral (why pledges are fuses). Invocation dumps the control block into the market, which drives the price down, which deepens the shortfall on any remaining pledged shares and on other promoters' pledges in the same name: a feedback loop. And because the block is large (often 40–90% of the promoter's stake, and the promoter often owns a majority), the forced supply can be a double-digit percentage of the entire company's shares, overwhelming any normal liquidity. This is why 25 January 2019 saw a major Indian media promoter's listed company fall sharply in a single session on pledge-invocation fears (Zee/Essel), and why CG Power's promoter pledge near 100% sat beside the balance-sheet looting the same order described. WE3 works the full cascade.

Reading the pledge screen (level + trend + purpose + group). A pledge is not automatically damning. It can fund a benign estate settlement and sit stable for a decade. The expert reads four things together: the level (20% versus 80%), the trend (stable versus rising four quarters running), the stated purpose (personal versus funding the listed company versus funding other group entities), and the group context (a stressed, leveraged promoter beside a cash-rich listed company is motive adjacent to opportunity).

A rising pledge beside rising RPT advances is the highest-conviction promoter cluster there is.

US contrast. US disclosure of pledged shares is annual and narrative, in the proxy (Regulation S-K Item 403 beneficial-ownership footnotes), often alongside a board anti-pledging/anti-hedging policy (Item 407(i)/402 disclosures under Dodd-Frank §955, verify). Notable US pledge blow-ups exist (a well-known energy-company founder faced forced sales of pledged stock in 2008), and large founder pledges are disclosed today, but the US lacks India's quarterly, percentage-of-holding pledge line, so the Indian analyst has, uniquely, a near-real-time fuse gauge.

4.8 The governance-rating synthesis

The output of the whole method is a defensible governance rating: a grade you can put in a memo and defend to a portfolio manager or a compliance officer.

Build a weighted rubric. Score a handful of independent dimensions, each 0–2 (2 = clean), and weight them: for example, pledge (level/trend/purpose) 25%, RPT intensity and leakage 25%, auditor signals 20%, subsidiary/dark-debt opacity 15%, and board/audit-committee independence 15%. The weighted score maps to a grade (A–F). The discipline is that the inputs are all public and computable, so two analysts with the same documents should land within a grade of each other.

Apply the severity ladder from M2.08 to governance, not just accounts. L1–L2 (a stable, explained pledge; small, disclosed, arm's-length RPTs) → note it, dock the score, move on. L3 (rising pledge, rising RPT advances, a mid-tenure auditor concern, thick dark debt) → cap position size, demand specific disclosures, set written tripwires. L4 evidence cluster (near-total pledge + swelling promoter advances + a mid-tenure resignation + a lien-marked deposit for a group loan) → do not own it at any attractive price; if you publish, describe evidence, never verdicts.

State the disclosures that would change your rating. A good governance memo ends not with a verdict but with a falsifiable list: "This rises to a B if the promoter deleverages the pledge below 20%, the audit committee publishes the arm's-length basis for the royalty, and the FY-next annual report shows related-party advances reversing." That list is what makes the rating a living instrument rather than a slur, and it is the exact discipline CL-11.4 grades.

Common mistakes & how experts think differently

1. Reading the P&L and never the shareholding pattern. The novice grades a promoter company on margins and growth. The expert opens the shareholding pattern first, for the pledge percentage and its trend, because in a promoter market the control-block leverage is the fastest-moving risk and the one most likely to detonate the equity independent of operations.

2. Treating any RPT as damning, or any RPT as fine. Both are errors. RPTs are ubiquitous and mostly benign (a shared-services arm, an arm's-length supply agreement). The expert does not react to the existence of RPTs; they quantify the leakage channel by channel and test each against the arm's-length alternative, escalating only where the price gap, the forgone interest, or the excess fee is real and material.

3. Confusing scheduled rotation with a resignation. Post-2014 mandatory rotation makes auditor changes routine; the novice trades every change as a red flag and cries wolf. The expert distinguishes a clock-driven rotation (noise) from a mid-tenure resignation citing information withheld (the fire alarm), the single highest-conviction public signal in the Indian toolkit.

4. Missing the aggregation clause on Regulation 23. The novice checks whether any single RPT breaches the materiality gate and clears the company when none does. The expert sums the year's transactions with each related party, because the rule aggregates. The salami-sliced structure (many sub-threshold transactions with one party) is precisely the one designed to avoid the shareholder vote.

5. Conflating the two material-subsidiary thresholds. The 10% (Regulation 16) "material subsidiary" and the 20% (Regulation 24(1)) "independent-director-on-board" thresholds are different numbers for different obligations. The expert keeps them straight and knows a subsidiary can be material for disclosure yet below the board-seat trigger.

6. Stopping at consolidated debt. The novice reads consolidated borrowings and thinks they have the leverage picture. The expert computes dark debt (consolidated minus holdco-standalone minus listed-subsidiaries) and pulls the unlisted entities' MCA filings, because the debt that kills is the debt the exchanges never see (IL&FS).

7. Naming the beneficiary before the record does. The novice writes "the promoter is stealing." The expert writes "₹900 crore of interest-free advances reached a promoter-connected entity through three subsidiary layers; I cannot reconcile this with the funds serving the listed company", a sentence that is true, defensible, and more precise, and that lets position size, not adjectives, express conviction.

8. Ignoring the reflexive nature of pledges. The novice treats a pledged stake as a static fact. The expert models the cascade: a price fall triggers a margin call, an unmet call triggers invocation, invocation dumps a large block that drives the price down further: a spiral in which the pledge percentage is a fuse whose length is printed every quarter.

9. Believing the guarantee is free because no cash moved. A corporate guarantee for a promoter entity records no expense and often no visible asset, so the novice ignores it. The expert prices the forgone guarantee commission and books the contingent liability, because a guarantee is the listed company underwriting the promoter's risk for nothing, and it converts to real debt the day it is invoked.

10. Trusting round numbers that come back. The novice sees a promoter "infuse" capital and reads confidence. The expert checks the timing: money that left the company days earlier and returned as "promoter infusion" is round-tripping: the group recycling the company's own cash so that it looks like a capital commitment. SEBI's 2023 interim and confirmatory orders at Zee/Essel described this pattern on a prima facie basis; those findings were preliminary, and appellate and settlement proceedings followed.

11. Grading the board by résumés. The novice counts independent directors and famous names and is reassured. The expert asks whether independent directors have been resigning (a real-time vote of no confidence), how active the audit committee is, whether the same directors sit across the promoter's entities, and whether the board once approved a related-party rescue (Satyam's Maytas). Engagement, not eminence.

12. Waiting for the SEBI order, or asserting the verdict before it. Symmetrical failures. Waiting for adjudication means waiting for the default; exposure decisions run on evidence, not verdicts. But writing "fraud" before a tribunal has found it is defamation risk and analytical sloppiness at once. The expert acts early on evidence and writes carefully in the language protocol. The novice does the opposite: acts late, writes recklessly.

Worked examples

Six workups, done the way the mini-project demands: every number shown and recomputed in Python for this module (R3). Synthetic companies are labelled "(synthetic)" and are internally consistent, so recompute them as your drill. Real-company statements are adjudicated/regulator-published facts, attributed and tagged "verify," per the language protocol.

WE1: Meridian Textiles (synthetic): quantifying the RPT web (₹)

Meridian Textiles Ltd (synthetic): revenue ₹4,000 crore, PAT ₹300 crore, net worth ₹2,000 crore. Its promoter controls four private entities that transact with it. From the related-party note and the loans-and-advances schedule you extract: (i) it sells goods with a fair value of ₹1,200 crore to a promoter-owned distributor at an 8% discount; (ii) it carries ₹500 crore of interest-free advances to a promoter entity (its own borrowing cost is 9%); (iii) it pays a brand royalty of 2.0% of revenue to the promoter holding company, against a 0.5% arm's-length norm; (iv) it guarantees ₹800 crore of a promoter entity's borrowing, charging no commission (arm's-length ≈ 1.0%).

Step 1: price each channel. Off-market sales leakage = 0.08 × 1,200 = ₹96 crore. Interest-free advances opportunity cost = 0.09 × 500 = ₹45 crore. Royalty excess = (0.020 − 0.005) × 4,000 = 0.015 × 4,000 = ₹60 crore. Forgone guarantee commission = 0.01 × 800 = ₹8 crore.

Step 2: total the leakage. 96 + 45 + 60 + 8 = ₹209 crore per year.

Step 3: express it against profit and capital. 209 ÷ 300 = 69.7% of PAT; 209 ÷ 2,000 = 10.5% of net worth. Roughly seven rupees of every ten of reported profit is leaving the listed company to the promoter ecosystem each year, entirely through disclosed related-party channels.

Step 4: the verdict (language protocol). "Related-party dealings — an 8% price concession on ₹1,200 crore of sales to a promoter distributor, ₹500 crore of interest-free advances, brand royalty at 2.0% versus a ~0.5% norm, and an uncompensated ₹800 crore guarantee — represent ~₹209 crore of annual value transfer, ~70% of reported PAT, that I cannot reconcile with minority-shareholder value being protected. Governance severity L3–L4. This rises to a B only if the royalty's arm's-length basis is published, the advances are recalled with interest, and a guarantee commission is charged." (Note: if the entire royalty were treated as leakage rather than only the excess, the total rises to ₹229 crore, ~76% of PAT: state which convention you used.)

WE2: The Suvarna Group (synthetic): Regulation 23 materiality threshold tests (₹)

Suvarna Textiles Ltd (synthetic) has consolidated turnover of ₹5,000 crore (last audited). Test three RPT scenarios against LODR Regulation 23. For contrast, a sister company, Suvarna Cement (synthetic), has consolidated turnover of ₹40,000 crore.

Step 1: the materiality gate for each company. Suvarna Textiles: turnover is below ₹20,000 crore, so the gate is 10% × 5,000 = ₹500 crore. Suvarna Cement: turnover sits at the top of the second band, so the gate is 2,000 + 0.05 × (40,000 − 20,000) = ₹3,000 crore (under the rule before the 2025 amendment it was the ₹1,000 crore cap).

Step 2: a single large RPT. Suvarna Textiles proposes a ₹520 crore purchase from a promoter entity. 520 > 500 → material → prior shareholder approval required, with the related party (and its group) abstaining; audit-committee approval is required regardless of size.

Step 3: three transactions with the same party (aggregation). Three separate deals of ₹180 + ₹160 + ₹190 crore with one promoter entity. Each is individually below the ₹500 crore gate, but "taken together during the financial year" they sum to 180 + 160 + 190 = ₹530 crore > 500 → material by aggregation. Structuring one large transaction as three medium ones does not escape the shareholder vote.

Step 4: the "1% structuring" tactic, defeated. Suppose every transaction is kept to ~1% of turnover: 0.01 × 5,000 = ₹50 crore each. It takes 500 ÷ 50 = 10 such transactions to reach the gate, and the aggregation clause sums them, so ten ₹50 crore payments to one related party in a year are material together. The analyst's move is simply to sum the year's dealings per party, which the half-yearly RPT disclosure (Reg 23(9)) lets you do.

Step 5: the memo line. "Aggregated FY dealings with Promoter Entity X total ₹530 crore against a ₹500 crore materiality gate (10% of ₹5,000 crore consolidated turnover); this required prior shareholder approval with the party abstaining. Verify from the RPT disclosure and the AGM/postal-ballot results whether that approval was sought and how the vote split ex-promoter."

WE3: Konark Power (synthetic): the pledge margin-call cascade (₹)

Konark Power Ltd (synthetic) has 100 crore shares outstanding; the promoter owns 60% (60 crore shares). The share price is ₹200. The promoter pledges 75% of the holding (45 crore shares) for a loan at an initial 50% LTV. The lender's maintenance trigger is a minimum cover of 1.5×; on a margin call, the promoter must top up to restore 2.0× cover.

Step 1: the loan and initial cover. Pledged value = 45 × 200 = ₹9,000 crore. Loan at 50% LTV = 0.50 × 9,000 = ₹4,500 crore. Initial cover = 9,000 ÷ 4,500 = 2.0×.

Step 2: the trigger price and the fall that fires it. The margin call fires when pledged value falls to 1.5 × 4,500 = ₹6,750 crore, i.e., a share price of 6,750 ÷ 45 = ₹150. The fall to trigger = (200 − 150) ÷ 200 = 25%. A 25% drawdown, an ordinary bad quarter, puts the control block on a margin call.

Step 3: the top-up, and why it exhausts the promoter. To restore 2.0× cover the collateral must be worth 2.0 × 4,500 = ₹9,000 crore; at the trigger it is worth ₹6,750 crore, a shortfall of 9,000 − 6,750 = ₹2,250 crore. Topping up with shares at ₹150 requires 2,250 ÷ 150 = 15 crore additional shares. The promoter's unpledged shares number 60 − 45 = 15 crore, exactly enough to meet one call and then nothing is left. The fuse is one bad quarter long.

Step 4: the invocation spiral. If the price falls further to ₹130, cover = (45 × 130) ÷ 4,500 = 1.3×; with no shares left to pledge, the lender invokes and sells. Invoked shares of 45 crore equal 45 ÷ 100 = 45% of the entire company: forced supply no market can absorb, driving the price down further and deepening the loss, the reflexive spiral in action.

Step 4b: the same pledge on the second base. The 75% figure is the encumbrance as a share of the promoter's own holding. On total share capital it is 45 ÷ 100 = 45.0000%, and that is the number the takeover code cares about. SAST Regulation 31(4)'s reasons-disclosure duty bites at 50% of the promoter's holding or at 20% of total share capital, whichever is lower. For a 60% promoter the second limb equals 33.3333% of the holding, so Konark's promoter crossed the binding threshold at 33.33% pledged rather than at 75%. A screen reporting only the holding-based figure would have called this promoter unremarkable across the whole stretch from 33.33% to 50%.

Step 5: the memo line (language protocol). "The promoter has pledged 75% of a 60% stake, which is 45% of total share capital, at ~50% LTV. A 25% price fall triggers a margin call the promoter can meet only once, with 15 crore free shares against a 15 crore top-up, and invocation would dump 45% of the company into the market. Pledge is a live L3–L4 risk independent of operations. Tripwire: any rise in pledged %, any disclosed invocation under SAST Reg 31, price within 25% of ₹200." (US contrast: this same fuse would appear only annually in a US proxy footnote, Item 403, not quarterly.)

WE4: Trishul Holdings (synthetic): subsidiary/loan hopscotch and dark debt (₹)

Trishul Holdings Ltd (synthetic) reports consolidated borrowings of ₹18,000 crore. The holdco's standalone borrowings are ₹3,000 crore; its two listed subsidiaries together disclose ₹5,000 crore. A footnote and the MCA filings reveal a chain: the holdco lent to wholly-owned Subsidiary A, which lent to step-down Subsidiary B, which advanced ₹900 crore interest-free to a promoter-owned entity, Trishul Realty (synthetic).

Step 1: size the dark debt. 18,000 − 3,000 − 5,000 = ₹10,000 crore of borrowings live in unlisted entities, 10,000 ÷ 18,000 = 55.6% of consolidated debt, disclosed only on MCA21 and in offer documents, not on the exchanges. Complexity here is the risk metric.

Step 2: trace the hopscotch. Holdco → Sub A → Sub B → Trishul Realty: by the third hop, ₹900 crore of listed-group cash has reached a promoter beneficiary outside the consolidated perimeter. Test the layering against Section 186 (no investment through more than two layers of investment companies, with exceptions): a chain this deep is either using an exception or breaching the rule, a finding either way.

Step 3: price the leakage and flag the round-trip. The ₹900 crore advance is interest-free; opportunity cost at 10% = 0.10 × 900 = ₹90 crore per year of forgone income. If any part of that ₹900 crore returns to the holdco within days as "share application money," it is round-tripping (the pattern SEBI's 2023 orders at Zee/Essel described prima facie), not a promoter infusion. Check the financing-activity lines and the RPT note together for the offsetting flows.

Step 4: the memo line. "55.6% of consolidated borrowings (₹10,000 crore) sit in unlisted entities; a three-layer inter-corporate chain routed ₹900 crore interest-free to a promoter-owned entity (₹90 crore/yr forgone). I cannot reconcile the routing with the funds serving the listed group. Pull the AOC-4 filings of Sub B and Trishul Realty; test the layering against Section 186; check for same-week return flows (round-tripping). Severity L3–L4."

WE5: Nilgiri Foods (synthetic): reading the auditor-signal sequence (₹)

Nilgiri Foods Ltd (synthetic), PAT ₹110 crore, is on a five-year audit term. You read four years of audit reports in sequence. FY22: clean opinion, but a Key Audit Matter (SA 701) on "revenue recognition — sales to distributors near period-end." FY23: the KAM repeats, plus an Emphasis of Matter on the recoverability of a ₹320 crore receivable from a related party. Q2 FY24 (year 3 of the term): the statutory auditor resigns mid-tenure, citing "information and explanations not made available." Q3 FY24: a rating downgrade and a failed commercial-paper rollover; then default.

Step 1: rank the signals by conviction. In the M2.08 hierarchy, gatekeeper behaviour outranks documents, which outrank ratios. So the ranking is: the mid-tenure resignation (highest) > the EOM on a related-party receivable > the KAM > the ratios. The resignation is decisive because it is an informed insider with a franchise to lose, acting against interest.

Step 2: confirm it is a resignation, not a rotation. The term is five years and the exit is in year 3, 5 − 3 = 2 years early. This is not the Section 139 clock; it is a walk-out, and SEBI's 2019 framework means the resignation letter carries stated reasons you should read.

Step 3: size the EOM. The related-party receivable of ₹320 crore is 320 ÷ 110 = 2.9× annual PAT. If written off, it erases nearly three years of profit. An EOM on a number this large relative to earnings is not boilerplate.

Step 4: locate the fraud clauses. Read CARO 2020 clause 3(xi) (fraud noticed/reported), clause 3(iii) (loans/advances to related parties), and the fund-routing clauses; and recall Section 143(12): a suspected fraud of ₹1 crore or more goes to the Central Government via ADT-4. The RPT receivable plus the resignation should send you straight to these clauses in the last available report.

Step 5: the timing lesson. The KAM fired ~18 months before default; the EOM ~9 months before; the resignation ~4 months before. Ratios and KAMs give you early warning with many false positives; the resignation gives you late, high-conviction confirmation. Rank your tells, and your position response, accordingly. (US contrast: the resignation would surface as a Form 8-K Item 4.01 within four business days, with the auditor's letter attached.)

WE6: Cascade Holdings (synthetic): the US$ contrast, co-borrowing, RPT leak, and the control wedge ($)

Cascade Holdings Inc. (synthetic) is US-listed with a dual-class structure: the controlling family holds 18% of the economics but 62% of the votes. From the 10-K and proxy you extract: family-controlled private entities co-borrow on facilities on which the public company is jointly liable, for total co-borrowing of $2,300m, of which only $400m is carried on Cascade's balance sheet. Related-party items (Regulation S-K Item 404): a $45m/year management fee to a family entity (against a ~$10m benchmark for comparable services) and an HQ lease from a family entity at $30m/year versus a ~$18m market rent. The family has pledged 40% of its super-voting shares as margin collateral (disclosed in the proxy, Item 403).

Step 1: the control wedge. Votes ÷ economics = 0.62 ÷ 0.18 = 3.4×. The family controls the company more than three times over relative to its cash-flow stake: the structural governance risk that dual-class encodes, the US analogue of promoter control.

Step 2: the off-balance-sheet exposure. Co-borrowing of $2,300m less the $400m on-balance-sheet = 2,300 − 400 = $1,900m of joint-and-several exposure disclosed only in the notes/commitments. This is the Adelphia co-borrowing pattern: the Rigas family used the public company's credit for private borrowing kept off the balance sheet, a scheme for which John and Timothy Rigas were convicted in 2004 after the 2002 bankruptcy.

Step 3: the RPT leak. Management-fee excess over benchmark = 45 − 10 = $35m; lease excess = 30 − 18 = $12m. Combining the excess fee and the lease excess = 35 + 12 = $47m/year of questionable transfer; if you instead sum the full management fee plus the lease excess, 45 + 12 = $57m. State the convention. Every item is above the $120,000 Item 404 threshold and therefore disclosed in the proxy. The US relies on disclosure plus audit-committee review rather than a shareholder-vote gate.

Step 4: the memo line. "A 3.4× vote-to-economics wedge, $1,900m of off-balance-sheet joint co-borrowing with family entities, ~$47m/yr of above-market related-party payments, and 40% of the control block pledged as margin — a governance cluster I grade L3–L4. Verify the co-borrowing terms in the debt footnotes and the RPT terms in the proxy. This is the Adelphia template (adjudicated) in a different accounting regime — control converts the public company into the family's private bank."

Practice set

Fourteen problems, ramped guided (P1–P5, method prompted) → independent (P6–P10, you choose the tools) → timed (P11–P14, ~8 minutes each, exam conditions). Every company is synthetic unless a real name is used for an adjudicated or regulator-published fact, which you check against the order itself. Work each fully before reading the solution. Every number here was recomputed in Python before publication.

P1 (guided: RPT leakage, ₹). Amba Industries (synthetic): revenue ₹2,500 crore, PAT ₹180 crore. It sells fair-value ₹600 crore of goods to a promoter distributor at a 7% discount; carries ₹300 crore of interest-free advances to a promoter entity (borrowing cost 9.5%); and pays brand royalty of 1.8% of revenue against a 0.4% norm. (a) Price each channel. (b) Total the leakage and express it as a share of PAT.

Solution. (a) Discount = 0.07 × 600 = ₹42 crore; advance opportunity cost = 0.095 × 300 = ₹28.5 crore; royalty excess = (0.018 − 0.004) × 2,500 = 0.014 × 2,500 = ₹35 crore. (b) Total = 42 + 28.5 + 35 = ₹105.5 crore = 105.5 ÷ 180 = 58.6% of PAT. Verdict (language protocol): "~₹105 crore of annual related-party transfer, ~59% of PAT, that I cannot reconcile with arm's-length dealing; severity L3 pending the royalty basis and advance terms."

P2 (guided: Regulation 23 gate, ₹). Konkan Chemicals (synthetic) has consolidated turnover of ₹3,200 crore. (a) Compute the RPT materiality gate. (b) Is a single ₹340 crore RPT material? (c) Are four transactions with one related party of ₹90 + ₹95 + ₹85 + ₹100 crore material?

Solution. (a) Turnover is below ₹20,000 crore, so the gate is 0.10 × 3,200 = ₹320 crore. (b) 340 > 320 → material; prior shareholder approval with the party abstaining. (c) Aggregate = 90 + 95 + 85 + 100 = ₹370 crore > 320 → material by aggregation, even though each transaction is individually below the gate. The aggregation clause defeats the salami-slice.

P3 (guided: pledge margin call, ₹). The promoter of Sahyadri Steel (synthetic) owns 50 crore shares and pledges 30 crore at a price of ₹400 for a loan at 55% LTV. The lender's trigger is a minimum cover of 1.4×; a call must be topped up to 1.8× cover. (a) Loan and initial cover. (b) Trigger price and percentage fall. (c) Top-up shares needed and whether the free float suffices.

Solution. (a) Pledged value = 30 × 400 = ₹12,000 crore; loan = 0.55 × 12,000 = ₹6,600 crore; initial cover = 12,000 ÷ 6,600 = 1.82×. (b) Trigger value = 1.4 × 6,600 = ₹9,240 crore; trigger price = 9,240 ÷ 30 = ₹308; fall = (400 − 308) ÷ 400 = 23%. (c) To restore 1.8× cover: needed value = 1.8 × 6,600 = ₹11,880 crore; shortfall at trigger = 11,880 − 9,240 = ₹2,640 crore; top-up shares at ₹308 = 2,640 ÷ 308 = 8.57 crore; unpledged shares = 50 − 30 = 20 crore → sufficient for this call (8.57 < 20), but each subsequent call eats the buffer. Note that a higher initial LTV (55% vs 50%) means a smaller fall (23% vs 25%) fires the call — leverage shortens the fuse.

P4 (guided: Regulation 24 material subsidiary, ₹). Vindhya Ltd (synthetic) reports consolidated income of ₹8,000 crore and consolidated net worth of ₹6,000 crore. Its subsidiary S has income of ₹900 crore and net worth of ₹1,400 crore. (a) Is S a "material subsidiary" under Regulation 16 (10%)? (b) Does Regulation 24(1) (20%) require an independent director on S's board? (c) Name two other Regulation 24 obligations.

Solution. (a) Regulation 16 thresholds: 0.10 × 8,000 = ₹800 crore (income) and 0.10 × 6,000 = ₹600 crore (net worth). S's income (900 > 800) or net worth (1,400 > 600) exceeds: yes, material (either limb suffices). (b) Regulation 24(1) thresholds: 0.20 × 8,000 = ₹1,600 crore (income) and 0.20 × 6,000 = ₹1,200 crore (net worth). S's income (900 < 1,600) does not qualify, but net worth (1,400 > 1,200) does: yes, an independent director is required on S's board (net-worth limb). (c) Any two: audit-committee review of S's investments; S's board minutes placed before the listed board; special resolution to reduce control/holding to ≤50%; special resolution to sell >20% of S's assets in a year.

P5 (guided: dark debt and hopscotch, ₹). Godavari Infra (synthetic) reports consolidated borrowings of ₹64,000 crore; the holdco standalone shows ₹9,000 crore; three listed subsidiaries disclose ₹21,000 crore combined. A ₹1,500 crore interest-free advance is traced holdco → Sub A → Sub B → a promoter entity. (a) Dark debt and its share. (b) Annual leakage on the advance at a 10.5% cost. (c) One statutory test and one document to pull.

Solution. (a) Dark debt = 64,000 − 9,000 − 21,000 = ₹34,000 crore = 34,000 ÷ 64,000 = 53.1% of consolidated borrowings. The majority sits in unlisted entities. (b) 0.105 × 1,500 = ₹157.5 crore/year forgone. (c) Test the chain against Section 186's two-layer investment-company limit; pull Sub B's and the promoter entity's AOC-4 (financials) from MCA21 to see where the ₹1,500 crore lands and whether it returns (round-tripping).

P6 (independent: pledge reflexivity, ₹). Chambal Media (synthetic): 200 crore shares outstanding; promoter owns 42% and has pledged 95% of that holding at a price of ₹250, for a loan at 60% LTV. The lender's trigger is 1.5× cover. (a) Loan. (b) Trigger price and percentage fall. (c) If invoked, what percentage of the company hits the market, and why is that the core risk?

Solution. (a) Promoter shares = 200 × 0.42 = 84 crore; pledged = 84 × 0.95 = 79.8 crore; pledged value = 79.8 × 250 = ₹19,950 crore; loan = 0.60 × 19,950 = ₹11,970 crore. (b) Trigger value = 1.5 × 11,970 = ₹17,955 crore; trigger price = 17,955 ÷ 79.8 = ₹225; fall = (250 − 225) ÷ 250 = 10%. A mere 10% fall fires the call, because the LTV is already high at 60%. (c) Invoked shares = 79.8 crore = 79.8 ÷ 200 = 39.9% of the entire company: forced supply of ~40% of the shares that no market absorbs, so invocation itself craters the price and can trigger a change-of-control/open-offer event. The high LTV plus near-total pledge is the danger, not the pledge alone.

P7 (independent: Regulation 23 aggregation/structuring, ₹). Palar Textiles (synthetic) has consolidated turnover of ₹12,000 crore. In one financial year the promoter structures 12 monthly "service" payments of ₹105 crore each to a promoter entity, arguing each is immaterial. (a) The gate. (b) The aggregate and the verdict. (c) What is the tell?

Solution. (a) Turnover is below ₹20,000 crore, so the gate is 0.10 × 12,000 = ₹1,200 crore. (b) Aggregate = 12 × 105 = ₹1,260 crore > 1,200 → material by aggregation, requiring prior shareholder approval with the party abstaining, despite each payment being ~105 ÷ 12,000 = 0.88% of turnover. Aggregation runs within a financial year, so payments spread across two years are summed year by year. (c) The tell is the pattern: many identical, sub-threshold, recurring payments to one related party is the classic structure-below-the-gate design; sum the year and it collapses. Confirm from the half-yearly RPT disclosure (Reg 23(9)) and check whether an omnibus approval was used to avoid transaction-level scrutiny.

P8 (independent: auditor-signal ranking). Rank these five signals from highest to lowest conviction, and say why the top one outranks the rest: (i) a scheduled auditor rotation under Section 139; (ii) an increase in audit fees; (iii) a new Key Audit Matter on related-party receivables; (iv) an Emphasis of Matter on going concern; (v) a mid-tenure auditor resignation citing information not made available.

Solution. Ranking: (v) > (iv) > (iii) > (ii) ≈ (i). The mid-tenure resignation (v) is highest because it is an informed gatekeeper with inspection access and a franchise to lose, acting against commercial interest and outside the rotation clock: the loudest public signal available. The going-concern EOM (iv) is a serious auditor statement short of walking out. The KAM (iii) flags where audit stress sits but is not itself negative. A fee increase (ii) and a scheduled rotation (i) are routine noise; (i) especially, because post-2014 rotation is clock-driven and expected. The novice error is treating (i) as a red flag; the expert reserves conviction for (v).

P9 (independent: SBO tracing, ₹/%). An individual holds 60% of Entity X (unlisted), which holds 30% of listed Deccan Foods (synthetic). No one appears on Deccan's SBO register. (a) Compute the individual's indirect interest. (b) Is she a significant beneficial owner under Section 90? (c) What filings should exist, and what does their absence suggest?

Solution. (a) Indirect interest = 0.60 × 0.30 = 18%. (b) 18% ≥ 10% → yes, an SBO under Section 90 and the SBO Rules. (c) She should have filed Form BEN-1 and the company Form BEN-2 on MCA21; an independent director / audit committee should have satisfied itself of the register's accuracy. An empty SBO register beside an obviously layered 18% indirect interest is a compliance red flag and a mapping finding. Pull MGT-7 and the BEN filings to confirm.

P10 (independent: US$ contrast: control wedge + RPT). Summit Media Inc. (synthetic, US) has a dual-class structure: the founder holds 15% of economics and 55% of votes. Item 404 discloses a $50m/year management fee to a founder entity (benchmark ~$8m) and a $25m HQ lease (market ~$16m). Net income is $200m. (a) The control wedge. (b) Total questionable RPT transfer (excess basis) and its share of net income. (c) How does US disclosure of all this differ from India's?

Solution. (a) Wedge = 0.55 ÷ 0.15 = 3.7×. (b) Management-fee excess = 50 − 8 = $42m; lease excess = 25 − 16 = $9m; total excess = 42 + 9 = $51m = 51 ÷ 200 = 25.5% of net income. (c) Both items exceed the $120,000 Item 404 threshold and are disclosed in the proxy/10-K with audit-committee review, but the US imposes no shareholder-vote materiality gate and no quarterly pledge line. It relies on disclosure, independent directors, and litigation, where India (LODR Reg 23) forces a shareholder vote with the related party abstaining. Same leakage, different control regime.

P11 (timed: pledge margin call, $). A US promoter-CEO pledges 8 million shares at $75 for a margin loan at 50% LTV; the maintenance trigger is 1.6× cover. (a) Loan. (b) Trigger price and percentage fall. (c) Which proxy item discloses this?

Solution. (a) Pledged value = 8 × 75 = $600m; loan = 0.50 × 600 = $300m. (b) Trigger value = 1.6 × 300 = $480m; trigger price = 480 ÷ 8 = $60; fall = (75 − 60) ÷ 75 = 20%. (c) Regulation S-K Item 403 (beneficial-ownership footnotes disclose pledged shares), typically alongside the board's hedging/anti-pledging policy (Item 407(i)/402, verify). In India the same fuse would appear quarterly in the shareholding pattern and via SAST Reg 31.

P12 (timed: Regulation 23/24 combined, ₹). Narmada Ltd (synthetic): consolidated turnover ₹6,000 crore, consolidated net worth ₹4,500 crore. It has a subsidiary with income ₹700 crore and net worth ₹950 crore. (a) The RPT materiality gate. (b) Is the subsidiary "material" (Reg 16)? (c) Does Reg 24(1) require an independent director on its board?

Solution. (a) Turnover is below ₹20,000 crore, so the gate is 0.10 × 6,000 = ₹600 crore. (b) Reg 16: income limb 0.10 × 6,000 = 600 (700 > 600 ✓) and net-worth limb 0.10 × 4,500 = 450 (950 > 450 ✓) → material on both limbs. (c) Reg 24(1): income 0.20 × 6,000 = 1,200 (700 < 1,200) but net worth 0.20 × 4,500 = 900 (950 > 900 ✓) → yes, an independent director is required (net-worth limb). The subsidiary clears the 10% bar comfortably and just crosses the 20% net-worth bar.

P13 (independent: governance rating rubric, ₹). Score Tapti Corp (synthetic) on five weighted dimensions, each 0–2 (2 = clean): pledge 25%, RPT leakage 25%, auditor signals 20%, subsidiary/dark-debt opacity 15%, board independence 15%. The evidence: pledge 88% and rising (score 0); RPT leakage ~70% of PAT (0); one mid-tenure auditor concern but no resignation (1); dark debt 52% of consolidated (0); a functioning but promoter-heavy board (1). (a) Compute the weighted score and grade. (b) State the disclosures that would raise it.

Solution. (a) Weighted score = (0×25 + 0×25 + 1×20 + 0×15 + 1×15) ÷ (2×100) = (0+0+20+0+15) ÷ 200 = 35 ÷ 200 = 17.5% → grade F. (b) It rises materially only if: the pledge is brought below ~20% and stabilises; the RPT leakage channels are proven arm's-length or reversed (royalty basis published, advances recalled with interest); dark debt is consolidated or explained; and the board adds genuinely independent directors with an active audit committee. State them as a falsifiable list, not a slur.

P14 (timed: governance teardown + language protocol, ₹). In eight minutes, produce the verdict for Purna Ltd (synthetic): promoter pledge 88%; "loans and advances to related parties" grew from 6% to 21% of a ₹2,500 crore net worth over two years; dark debt is 52% of consolidated borrowings; the statutory auditor resigned mid-tenure last quarter citing information withheld. (a) Quantify the RPT-advance growth. (b) Write the one-paragraph verdict in the language protocol with a severity and two tripwires.

Solution. (a) RPT advances went from 0.06 × 2,500 = ₹150 crore to 0.21 × 2,500 = ₹525 crore, a 525 − 150 = ₹375 crore build in two years, a 3.5× increase in advances to the ecosystem. (b) "Four independent governance signals converge at Purna Ltd: promoter pledge at 88%, related-party advances up from ₹150 crore to ₹525 crore (6%→21% of net worth) in two years, dark debt at 52% of consolidated borrowings, and a mid-tenure statutory-auditor resignation citing information withheld. I cannot reconcile this cluster with minority-shareholder value being protected; the mid-tenure resignation is the highest-conviction signal and, with the pledge, drives an L4 evidence pattern. I do not own this at any price I would call attractive, and if I publish I will describe evidence, not assert a verdict. Tripwires: any pledge-invocation disclosure under SAST Reg 31; the incoming auditor's first opinion and any CARO 3(xi) fraud remark; the FY-next RPT disclosure showing whether advances reverse."

Applied mini-project: the promoter-web teardown

Budget ~3 hours. You will produce a complete governance teardown of a synthetic promoter group and defend a rating: the exact deliverable CL-11.4 auto-scores.

Part A: build the map (45 min). From the packet provided in the app (a synthetic shareholding pattern, RPT note, subsidiary list, pledge disclosure, and two years of auditor's reports), draw the promoter-group map: the listed company; its subsidiaries, step-downs, associates and JVs with holding percentages; the promoter/promoter-group entities outside the perimeter; the directed transaction and loan edges; and the SBO at the top. Mark every entity for which you would pull an MCA21 filing.

Part B: quantify the leakage (45 min). Build the RPT-leakage table: one row per channel (off-market pricing, soft advances, royalty excess, forgone guarantee commission, lien-marked deposits, management fees, revenue-that-is-a-loan), each with its rupee amount, the arm's-length benchmark you used, and the residue line where it lands. Total the annual leakage and express it as a share of PAT and net worth.

Part C: the threshold and pledge tests (30 min). Test each material RPT against the Regulation 23 gate (10% of consolidated turnover up to ₹20,000 crore, banded above it to a ₹5,000 crore cap) and apply the aggregation clause per related party. Classify each subsidiary under Regulation 16 (10%) and Regulation 24(1) (20%). Compute the pledge LTV, cover, trigger price, and percentage fall to the margin call, and state the invocation supply as a percentage of the company.

Part D: the rating and memo (30 min). Score the weighted governance rubric (Part 4.8), assign a grade, and write a one-page memo in the language protocol: conclusion first, the three findings that matter with severity, the leakage in numbers, the falsifiable list of disclosures that would change the rating, and the dated tripwires you will monitor.

Scoring rubric (self-grade honestly, or swap with a peer):

Pass standard: ≥13/16 with no criterion at 0, and the memo readable by a portfolio manager who has not seen the packet.

Case Lab: run case CL-11.4 in the app's Cases tab for this module.

Reading & resources

• SEBI LODR Regulations 2015, Regulations 16, 23, 24, 24A, 30, 31 (with the RPT amendments effective 2022–2023). The primary text; read Regulation 23 and its definitions word for word, then the half-yearly RPT disclosure format. sebi.gov.in. [Free] [Advanced]
• SEBI (SAST) Regulations 2011, Regulation 31 (encumbrance disclosure). The pledge-disclosure regime, including the broadened post-2019 definition of "encumbrance." [Free] [Advanced]
• The Companies Act 2013, Sections 90 (SBO), 139 (auditor rotation), 143(12) (fraud reporting), 177 & 188 (RPTs), 185 & 186 (loans, investments, layers). With the SBO Rules and the Companies (Auditor's Report) Order (CARO) 2020. mca.gov.in. [Free] [Advanced]
• SEBI enforcement orders. Read the adjudicated governance cases in full: the CG Power order (2019); the Zee/Essel interim and confirmatory orders (2023); the Fortis Healthcare fund-diversion order (2022); the Vakrangee orders. Note precisely what each found versus what remains alleged. sebi.gov.in → Enforcement → Orders. [Free] [Advanced]
• NFRA audit-quality orders (nfra.gov.in), the IL&FS/IFIN audit orders as worked examples of gatekeeper failure and its consequences. [Free] [Advanced]
• MCA21 (mca.gov.in): pull one real unlisted group entity's MGT-7, AOC-4, and BEN-2. The single best exercise here: see for yourself how much of the "dark" perimeter is actually public for a nominal fee. [Free] [Intermediate]
• Ind AS 24, Related Party Disclosures (and the older AS 18), the accounting definition of related parties and the disclosure architecture behind the RPT note. [Free] [Intermediate]
• SA 701 (Key Audit Matters), SA 705 (Modified Opinions), SA 706 (Emphasis of Matter). The auditing standards behind the opinion ladder you learned to rank. [Free] [Advanced]
• US contrast: SEC Regulation S-K Item 404 (RPTs) and Item 403 (ownership/pledges); PCAOB AS 3101 (Critical Audit Matters); Form 8-K Item 4.01 (auditor change). Read one real proxy's "Certain Relationships and Related Transactions" section to see the US disclosure regime in action. sec.gov. [Free] [Advanced]
• Tamal Bandyopadhyay, Pandemonium: The Great Indian Banking Tragedy (2020). The IL&FS/DHFL/Yes Bank credit cycle from inside: the promoter-web and funding-mismatch patterns as narrative. [Paid] [Intermediate]
• Institutional proxy-advisory research (IiAS, SES in India; ISS, Glass Lewis in the US). How professionals grade RPTs, pledges, and board quality at scale. Read their public reports as models of defamation-safe governance writing. [Free/Paid] [Intermediate]

The Modern Analyst's Addendum

Everything above teaches this skill from first principles, by hand. That is how you learn it, and the mastery check still tests it that way. This addendum shows how a working analyst amplifies the same skill today. It adds; it never replaces. (R1/R10)

AI-Augment this skill

{
  "skill": "Promoter-group and SBO mapping, the nine RPT leakage channels priced against arm's-length benchmarks, the Regulation 23 materiality gate and its aggregation clause, Regulation 24 and the subsidiary/loan hopscotch, the dark-debt reconciliation, pledge LTV/cover/trigger/invocation math, the auditor-signal rank order, and the language protocol that governs every sentence you write about a real company",
  "use": "Governance forensics has the cleanest division of labour in the program: the retrieval is enormous and mechanical, the judgement is small and irreducible, and the tool belongs entirely on the first side. §4.1's claim that India's public record is deeper than analysts assume is precisely a claim that there is more here than a human will read — which is a machine's job. Four uses. (1) LONGITUDINAL EXTRACTION, the one that changes what you can do at all: pull the Ind AS 24 related-party note, the loans-and-advances schedule and the contingent-liabilities note out of six or eight consecutive annual reports into one table, one row per counterparty per year, with a note reference in every cell. §4.3's residue — advances creeping up as a share of net worth, receivables from associates outrunning the revenue they supposedly arise from, guarantees swelling — is a *trend*, and nobody assembles eight years of it by hand. (2) THE ENTITY LIST for your MCA21 requests: have it read the subsidiary schedule, the related-party list and the charge disclosures and emit every entity named, with the document each appeared in. That list is what turns §4.5's dark-debt subtraction from a number into a specific set of AOC-4 filings to pull. (3) THE DOCUMENT-AGAINST-A-FIXED-QUESTION-LIST read: hand it a resignation letter, a CARO 2020 annexure or an auditor's report together with a closed list of questions — which clauses are answered adversely, does 3(xi) report a fraud, are 'information and explanations' qualified, what changed against last year's report — and require it to answer only those, quoting the text it relies on. (4) THE INNOCENT EXPLANATION: ask for the strongest defensible account of every flag you have raised, because a flag you have not steelmanned is a flag you will over-read, and §4.3's second common mistake is exactly that. Now the prohibition, which is the sharpest in this program and is not a style note: never ask a model to characterise a named person, family or company. Give it a name and a suspicion and it will return a fluent, confident, sourceless allegation, because that is what a language model does with those two inputs — and you will have manufactured a defamation exposure out of a probability distribution over tokens. Roles, entity types and case categories in the prompt; the names stay in the primary documents you open yourself.",
  "tools": ["Document/PDF AI — multi-year related-party notes, loans-and-advances schedules, CARO annexures, resignation letters and MCA21 filings into one referenced table", "Coding assistants — the governance panel, the leakage table, the two pledge closed forms and the screen's posterior below", "Chat assistants — the innocent-explanation steelman for every flag, and a pre-mortem on your own rating before you publish it"],
  "prompt": "Extraction only — do not characterise, rate or accuse any person or entity. From the attached annual reports for FY<A> to FY<B>, build one table: financial year, counterparty exactly as named in the note, relationship exactly as stated, transaction type mapped to my nine channels (off-market sales/purchases, advances, royalty/brand fee, guarantee, asset transfer, lien-marked deposit, management fee, investment in group entity, receivable from an associate), amount, and the note number and page it came from. Leave any cell you cannot source blank rather than inferring it, and list separately every figure you could not tie to a note. Then, as a second and separate output, list every entity named anywhere in the subsidiary schedule, the related-party list or the charge disclosures, with the document each appeared in — that list is my MCA21 filing request, not a finding about anyone.",
  "verify": "Three checks, in this order, and the third is the one this module exists for. (1) EVERY CELL TRACES. Open the note reference and read it. An extraction table is a claim about documents, and the failure mode here is specific: a fabricated amount for a transaction that does not exist looks exactly like the amounts that do, and it will be the one a lawyer asks you about. (2) THE ARITHMETIC IS YOURS. The leakage total, the Regulation 23 gate and its per-party aggregation, the pledge closed forms, the SBO chain multiplication and the dark-debt reconciliation are computed by you from the extracted inputs, never reported by the tool — §4.3's whole point is that the product is one number a portfolio manager can act on, and you must own it. (3) THE OUTPUT IS A QUEUE, NOT A FINDING. The quantitative lens below computes what a governance screen's posterior actually is against a low base rate of adjudicated failure, and the answer is that the large majority of what any such screen flags is an honest company under stress. A screen therefore schedules your reading; the reading produces evidence; and §4.1's language protocol — finding, allegation, or unreconciled evidence — is the exact vocabulary that posterior entitles you to use.",
  "diy": "The gate is unaided. You draw the promoter-group map and multiply the SBO's interest through the chain; you price all nine leakage channels against benchmarks you can name, total them, and state them against PAT and net worth; you compute the Regulation 23 gate from the turnover bands (10% up to ₹20,000 crore, then ₹2,000 crore plus 5%, then ₹3,000 crore plus 2.5%, capped at ₹5,000 crore) and apply the aggregation clause party by party; you keep Regulation 16's 10% apart from Regulation 24(1)'s 20%; you do LTV, cover, trigger price, percentage fall, top-up and invocation supply on paper; you subtract standalone and listed-subsidiary borrowings from consolidated to size dark debt; you rank the auditor signals with the mid-tenure resignation on top; and you write the verdict in the language protocol. Nothing in this callout is admissible in that room — and the language protocol is graded there too, because it is the part of this module that carries legal consequences."
}

Modern Data Analysis

By hand first. You priced four leakage channels to ₹209 crore and set them against ₹300 crore of PAT; you took a ₹5,000 crore turnover down to a ₹500 crore materiality gate and then defeated the salami-slice by summing three transactions to ₹530 crore; you walked a pledge from ₹9,000 crore of collateral through a ₹150 trigger price to an invocation that would dump 45% of a company into the market; you subtracted ₹3,000 crore and ₹5,000 crore from ₹18,000 crore and found ₹10,000 crore of borrowing living where the exchanges cannot see it. Every one of those stays in your hands. What code adds is not the arithmetic, which is small. It is two things the page cannot give you. The first is the panel: the same computations run across every company and every quarter, which is what turns a teardown into a screen. The second is an honest posterior on what the screen's output means, the harder half and the one almost nobody attaches.

Today's workflow. Three changes, and the first is a piece of algebra the module performs seven times and never states in closed form.

The pledge fuse is price-free and size-free. Let the promoter pledge S shares at price P, so collateral is S·P and the loan is L = LTV·S·P. A margin call fires when collateral falls to cover_trigger · L, i.e. at a price P′ where S·P′ = cover_trigger · LTV · S · P. The S cancels and so does the rupee amount:

• Fall to trigger = 1 − (cover_trigger × LTV). Nothing else. Not the share price, not the block size, not the loan.
• Top-up shares, as a fraction of the pledged block = (cover_restore ÷ cover_trigger) − 1. The LTV cancels here too.

That is why WE3's 25%, P3's 23%, P6's 10%, P11's and q17's 20%, A7's 25% and B10's 10% are one line of code, and why WE3's 15 crore top-up shares is just 45 × (2.0/1.5 − 1) and P3's 8.57 crore is 30 × (1.8/1.4 − 1). The practical consequence is that a pledge screen runs on an entire index from the quarterly shareholding pattern alone, once you assume the lender's terms. The only company-specific input the fall-to-trigger needs is the LTV, and the only thing the invocation supply needs is pledged shares over shares outstanding, both of which are disclosed. This is not a shortcut for the exam; it is what makes §4.6's screen a screen.

The governance record becomes a panel. One row per company-quarter, and the columns are the six signals India actually discloses: pledged percentage of promoter holding and its four-quarter change (LODR Reg 31, quarterly); aggregated related-party dealings per counterparty against the Reg 23 gate (Reg 23(9), half-yearly); related-party advances as a share of net worth (annual note); auditor exit and whether it falls on or off the Section 139 clock, with the substance of the stated reasons; independent-director resignations and their stated reasons (Reg 30 announcement feed); and dark debt as a share of consolidated borrowings, which needs an MCA21 pull. The frequency and the perimeter are where India differs, and the differences are specific: the pledged-percentage line is quarterly here and an annual proxy footnote there; related-party dealings arrive half-yearly and per-counterparty here against an annual narrative under Item 404 there; the substance of a resigning auditor's reasons is required here; and the unlisted perimeter where §4.5's dark debt lives is a filing fee away on MCA21 and has no US registry at all. The US is faster on exactly one of them, and it is worth conceding: an auditor change is an 8-K Item 4.01 within four business days, with the outgoing auditor given a chance to respond. So the asymmetry is not uniform. The honest summary is that India gives you a higher-frequency fuse gauge, a structured RPT feed and an unlisted perimeter, and the US gives you a faster gatekeeper alarm. §4.2's remark that India's pledge disclosure is richer "in this one respect" is where the panel stops being an observation and becomes an instrument.

And the screen gets a posterior attached before it is used, not after. That is the quantitative lens, and it is the reason this section stops at building the panel rather than at ranking it.

Tools & sources (IN + US). pandas for the company-quarter panel and the per-counterparty aggregation, numpy for the screen arithmetic, matplotlib for the pledge-versus-advances scatter that makes §4.6's "highest-conviction promoter cluster" visible. India: the BSE and NSE corporate-filings portals for the quarterly shareholding pattern (the pledged/encumbered line, machine-readable), the half-yearly related-party disclosure under Reg 23(9), and the Regulation 30 announcement feed where director and auditor exits land first; SEBI's enforcement-orders index (sebi.gov.in → Enforcement → Orders), which is the single most underused corpus in Indian equity research: free, searchable, and a catalogue of the mechanisms enforcement has actually found rather than the ones commentators imagine; NFRA's orders (nfra.gov.in) for the audit-quality tier above the statutory auditor; MCA21 (mca.gov.in) for MGT-7, AOC-4, BEN-2, DIR-12 and CHG filings on the unlisted entities that hold the dark debt, at a nominal per-document fee; screener.in and Tijori for the ten-year standalone-and-consolidated panels the dark-debt subtraction needs; and the published governance research of the Indian proxy advisers (IiAS, SES) as working models of defamation-safe writing at scale. US: EDGAR full-text search (sec.gov/edgar/search) across DEF 14A "Certain Relationships and Related Transactions" sections and 8-K Item 4.01 auditor-change filings, which is how you build the US auditor-exit panel; EDGAR XBRL companyfacts for the multi-year financial panel; PCAOB AuditorSearch for the engagement-partner and firm history behind an audit, and PCAOB inspection reports and enforcement orders for the firm-quality tier; the SEC's Accounting and Auditing Enforcement Releases (AAERs), which are the US analogue of the SEBI-order corpus and are the standard research dataset for exactly the base-rate question the lens below turns on.

# Governance screen: the pledge closed forms, then the screen's own posterior — R3, in-session
from math import comb

# 1) The two closed forms the module computes case by case (§4.6, WE3, P3, P6, P11, A7, B10).
fall  = lambda cover_trig, ltv:      1 - cover_trig * ltv        # price-free AND size-free
topup = lambda cover_restore, cover_trig: cover_restore / cover_trig - 1   # also LTV-free
print([round(fall(c, l), 4) for c, l in ((1.5,.50), (1.4,.55), (1.5,.60), (1.6,.50))],
      round(topup(2.0, 1.5)*45, 3), round(topup(1.8, 1.4)*30, 3))   # [.25 .23 .10 .20] 15.0 8.571

# 2) What a k-of-n red-flag screen is actually worth. p = base rate of adjudicated/disclosed
#    governance failure; q_bad/q_ok = per-signal fire rates in each population. ALL ASSUMED.
def ppv(p, n, k, q_bad, q_ok):
    tail = lambda q: sum(comb(n, i) * q**i * (1-q)**(n-i) for i in range(k, n+1))
    s, f = tail(q_bad), tail(q_ok)
    return p*s / (p*s + (1-p)*f), s, f
for k in (1, 3, 4):
    v, s, f = ppv(0.01, 6, k, 0.55, 0.10)
    print(k, round(v, 4), round(s, 4), round(f, 5))     # 2.09% | 32.19% | 77.84%  (independent)

# 3) The same conjunction when the six signals share one promoter-stress factor.
import numpy as np
from scipy.stats import norm
def ppv_rho(p, n, k, q_bad, q_ok, rho):
    z = np.linspace(-8, 8, 20001); w = np.exp(-z**2 / 2); w /= w.sum()
    def tail(q):
        pz = norm.cdf((np.sqrt(rho)*z - norm.ppf(1-q)) / np.sqrt(1-rho))
        return (w * sum(comb(n, i) * pz**i * (1-pz)**(n-i) for i in range(k, n+1))).sum()
    s, f = tail(q_bad), tail(q_ok)
    return p*s / (p*s + (1-p)*f)
print([float(round(ppv_rho(0.01, 6, 4, 0.55, 0.10, r), 4)) for r in (0.0, 0.2, 0.5)])
#      [0.7784, 0.3214, 0.1227]  — a correlation of 0.2 costs 46 points of posterior

Verify. Prove the engine against the module's own worked figures before it goes near a live panel, and the pledge closed forms make that cheap: the fall-to-trigger formula reproduces 25% (WE3), 23% (P3), 10% (P6), 20% (P11), 25% (A7), 10% (B10) and the $90 → $72 twenty-percent fall of q17 with no other input, and the top-up formula reproduces WE3's 15 crore shares and P3's 8.57 crore exactly. Then the rest: WE1's 96 + 45 + 60 + 8 = ₹209 crore, 69.7% of PAT and 10.45% of net worth (and ₹229 crore / 76.3% on the full-royalty convention, which is why WE1 tells you to state which one you used); WE2's gates of ₹500 crore and ₹1,000 crore, the ₹530 crore aggregate, and the ₹10,000 crore turnover above which the cap binds and the gate stops scaling; P7's 24 × 49 = ₹1,176 crore from payments each worth 0.41% of turnover; WE4's ₹10,000 crore of dark debt at 55.6% of consolidated and P5's ₹34,000 crore at 53.1%; WE5's related-party receivable at 2.91× annual PAT; WE6's 3.44× control wedge, $1,900m off-balance-sheet and $47m / $57m on the two conventions; P10's 3.67× wedge and 25.5% of net income; P13's weighted 17.5% grade-F; P14's ₹150 → ₹525 crore, a 3.5× build. Every one reproduces. One convention to hold straight when you code the pledge screen: the fall to trigger uses the lender's maintenance cover, while the top-up uses the restore cover, and the two are different numbers in every worked example above. A screen that uses one where it means the other will report a fuse that is either comfortably long or already lit.

Quantitative lens

There is exactly one quantitative fact that governs forensic work, it is the one most often skipped, and it is not about any screen's cleverness. A red-flag screen with high sensitivity, applied to a low base rate, produces flag sets that are overwhelmingly composed of honest companies. The screen is a triage instrument and never a verdict. Earlier modules have already established the arithmetic in two other settings and this addendum does not repeat it: M2.07's modern layer runs a synthetic-firm harness that measures what the Beneish/Sloan/Altman/Piotroski battery does to precision at a 2% manipulation prevalence, and M7.05's shows the same collapse for macro crisis signals and proves the asymmetry that specificity, not sensitivity, is the only lever that matters. Read both. What is new here is the India-specific question they cannot answer: India discloses more governance signals than the US does, so what does the extra disclosure actually buy? All figures below are computed from the assumptions stated in each sentence; they are properties of that arithmetic, not measurements of the Indian market.

Start with the collapse, on governance's own numbers. Take one screen (say the composite of §4.6's four-dimensional pledge read) with a generous 90% sensitivity and 90% specificity, and let p be the annual probability that a listed company turns out to have suffered material, adjudicated or disclosed governance-driven value destruction. At p = 2% the posterior on a flag is 15.5%; at p = 1% it is 8.3%; at p = 0.5% it is 4.3%; at p = 0.2% it is 1.8% and you are reading 56 files for each true one. Nobody knows what p is; that is the honest state of the field. The posterior is very nearly linear in it at these levels, so halving your assumed base rate halves your answer. A precision figure quoted without the base rate it assumed is not a number. The lever, as M7.05 proves in the macro setting, is specificity: holding sensitivity at 90% and p at 1%, pushing specificity from 90% to 99% moves the posterior from 8.3% to 47.6%, while pushing sensitivity from 60% all the way to a perfect 100% at fixed 90% specificity moves it only from 5.7% to 9.2%. Catching everything buys three and a half points. Being quiet buys thirty-nine.

Which is why conjunction is the only real instrument, and why India is where you can build one. A single quiet signal is hard to construct; several noisy signals that must fire together is the practical route to specificity, and §4.6 already says so in words: "a rising pledge beside rising RPT advances is the highest-conviction promoter cluster there is." Put numbers on it. Assume six disclosed signals, each firing on 55% of genuinely failing companies and 10% of honest ones, at a 1% base rate. Requiring any one to fire gives a posterior of 2.09%, worse than the single-screen case, because six chances to trip one alarm is a less specific instrument than one alarm. Requiring three of six gives 32.2%. Requiring four of six gives 77.8%. That is the mathematical content of the word "cluster," and it is a fifteen-fold improvement over any-one-fires.

Now the finding that should change how you read your own screen: almost all of that gain is an artefact of an assumption you cannot defend. The 77.8% above assumes the six signals are conditionally independent: that knowing a company's pledge is rising tells you nothing about whether its related-party advances are rising. That is false, and it is false for a reason this module spends eight sections establishing: a stressed, leveraged promoter beside a cash-rich listed company is one underlying condition that drives the pledge, the advances, the lien-marked deposit and the auditor's discomfort at the same time. Model that with a single common stress factor and let the within-population correlation be ρ. At ρ = 0 the four-of-six posterior is 77.8%. At ρ = 0.2, a correlation most analysts would call negligible, it is 32.1%. At ρ = 0.4 it is 15.9%, and at ρ = 0.5, 12.3%. A correlation of one fifth costs forty-six points of posterior. The mechanism is worth holding onto: correlation raises the co-firing rate in both populations, and because there are ninety-nine honest companies for every failing one, the rise in the honest population is what dominates. Correlated evidence is much weaker evidence than the same count of independent evidence, and a governance cluster is correlated evidence almost by construction.

So what does India's richer disclosure actually buy? Not what the intuition says. Compare a six-signal Indian surface against a three-signal US one, both at ρ = 0.5 and a 1% base rate. At the same proportion, three of six against two of three, the posteriors are 7.89% and 7.87%: identical, because at a fixed proportion more signals means more chances for an honest company to trip the threshold. What the six-signal surface buys is access to operating points the three-signal surface cannot express at all. Requiring five of six reaches 19.0%; requiring all six reaches 30.1%, against a best-case 16.2% for all-three-of-three, roughly double the posterior, at the price of sensitivity falling from 29.8% to 18.0%. That is the honest form of the India advantage: not a better screen at the same setting, but a reachable stricter setting, bought with disclosure and paid for in missed cases. And it is the arithmetic behind §4.8's rubric using five weighted dimensions rather than one, and behind the L4 ladder requiring a cluster (near-total pledge and swelling advances and a mid-tenure resignation and a lien-marked group deposit) rather than any single tell.

Then the sentence the whole module turns on. At the strictest operating point available (all six signals co-firing, ρ = 0.5, p = 1%) the posterior is 30.1%, which means roughly seven in ten of even those flags are an honest company. Push the other way and ask what a verdict would require. To justify 95% posterior confidence at these assumptions, the screen would need a false-positive rate of about 0.0096%, one honest company in ten thousand, and by the rule of three you would need roughly 31,000 clean company-years with not one false alarm to establish a rate that low at 95% confidence. No analyst has that, no vendor has that, and no published governance screen has that. The verdict is not merely imprudent; it is unestimable. Which means §4.1's language protocol is not a lawyer's hedge bolted onto the analysis. It is the correct output format of the instrument. "Related-party advances rose from 6% to 21% of net worth while the promoter pledge went from 22% to 81%, and I cannot reconcile that with minority-shareholder value being protected" is exactly the sentence a 30%-posterior instrument entitles you to write. "The promoter is looting the company" is a claim pitched at a confidence level the public record cannot support — and one a court will ask you to support. The ethics rule and Bayes' rule are the same rule, which is why the module puts §4.1 first.

And the practical output is a work queue with a size. Run the six-signal panel over a 500-name universe at p = 1%, ρ = 0.5, requiring three of six: 40 names flag, about 3.2 of them are real, and about 1.8 genuine cases are missed entirely. Tighten to four of six and it is 20 flags, 2.5 real, 2.5 missed. Neither is a finding; both are a reading list, and the second one is a week of one analyst's document work. Sized that way, the screen does the only job it can do honestly: it decides which forty annual reports you open, in what order, this quarter. §4.8's falsifiable change-list is what you write at the end of each one.

Honest limits, stated at length because this is the module where a confident-looking number does the most damage. Every input above is an assumption, and the most important one is unmeasured. The base rate p is the number nobody in this field has: what you could compute from the SEBI-order and AAER corpora is the rate of detected, pursued and adjudicated failure, which is a lower bound on the thing you care about and a non-randomly sampled one. Enforcement finds what it looks for, looks where it has been told to look, and adjudicates years after the fact, so a base rate estimated that way is biased downward by everything that was never caught and biased toward the mechanisms regulators had already learned to recognise. The per-signal fire rates (55% and 10%) and the correlation (ρ) are likewise chosen to be plausible and generous, not measured; the direction of the finding survives wide variation in them, but the specific posteriors do not, and the correct use of this section is to re-run it with your numbers rather than to quote 12.3%. The one-factor structure is itself a simplification: real governance signals cluster into subgroups (the two pledge signals move together more tightly than either moves with the auditor signal), which a single factor cannot represent and which would change the conjunction arithmetic in ways this model does not capture. The screen also treats each company-year as an independent draw when promoter groups and sectors cluster, so the effective sample behind any estimated rate is smaller than the row count suggests. Nothing above is a statement about how many Indian listed companies are badly governed; it is a statement about what a classifier can support at a stated base rate, which is a different and much narrower claim. And the deepest limit is one no arithmetic reaches: this section is about whether, and the entire practical question in governance is when — a pledge fuse whose length is printed every quarter still says nothing about which quarter the price falls 25%.

Do it in code: build the panel as one row per company-quarter with the six disclosed signals; run the pledge screen on the two closed forms so it works index-wide from the shareholding pattern; require a conjunction rather than any-one and state k explicitly; attach a posterior to every screen output with the assumed base rate written in the same sentence; report the flag count, the expected true count and the expected misses as a work queue rather than a ranking; and write every finding in the language protocol, because the posterior is what makes that vocabulary correct rather than merely cautious. See G1/QM1.01, G2/QD1.02 and G5/AI0.06.

Where this goes next: galaxy cross-links

• G1 · QM1.01 Probability, Distributions & Estimation: conditional probability, base rates and the posterior arithmetic above, plus the interval estimation that tells you how little a rate estimated from a few hundred company-years actually pins down.
• G2 · QD1.02 Signal & Factor Construction: turning the six disclosed governance signals into a scored, backtestable factor: thresholding, winsorising the pledge and advance series, and the survivorship and look-ahead traps that make a governance backtest flatter itself.
• G2 · DA2.03 Deep Learning, NLP & Time Series: reading a corpus of resignation letters, CARO annexures and enforcement orders as text rather than one document at a time, which is the only way the SEBI-order index becomes a pattern source rather than a case library.
• G5 · AI0.04 Retrieval & RAG: grounding every extracted figure in the note it came from, with citation, which is the technical form of this addendum's first verification check.
• G5 · AI0.06 Failure Modes & Guardrails: the confident fabricated allegation is the highest-stakes model failure in this program, and this is the module on why it happens and how a workflow is built so that it cannot reach a published sentence.

Teach it back & journal

Feynman prompt. Write a one-page explainer for a smart 15-year-old titled "How a family can quietly drain a company they don't fully own." You must use three of the ideas above. Pick from: the RPT web (the company buys and sells with the family's other businesses on bad terms), the pledge fuse (the family borrowed against the shares, so a price fall forces a fire sale), and the subsidiary hopscotch (money hops through company-owned businesses until it reaches the family, off the map). End with your own three-document routine (the shareholding pattern, the related-party note, and the auditor's report) in words a teenager could follow. No jargon: if you cannot explain "pledge" without the word "collateral," you do not own it yet.

Journal reflection. Open your decision journal and write the governance-calibration entry for this work: (1) the one real Indian case from M2.08 or this module whose promoter-web you would most likely have rationalised at the time (the growth story, the famous name), and the honest reason why; (2) your personal bias on the pledge screen: do you over-punish any pledge, or wave away a rising one because operations look good? (3) the single disclosure you would now always pull that you did not before (MCA21 for an unlisted group entity? the half-yearly RPT disclosure? the resignation letter?); and (4) one sentence you commit to writing in every governance memo: a falsifiable change-list line, so the rating stays an instrument and never becomes a slur. Date it; the app will ask you to reread this entry before CL-11.4.

This module's flashcards and mastery quiz are wired into the app: see the node's Quiz and Reviews.