Introducing FiDUS Buy Side

Your thesis, and what happened next.

Write down why you hold a company, in your own words. It becomes the assumptions it rests on, and every filing that arrives afterwards is read against them. You hear the moment one is touched. It never tells you what to do.

1,975companies · statements as of any date

143checklist items · nine frameworks

18sector operating-KPI libraries

5 minworker cadence on the live filing stream

Margin expansion holds. EBITDA margin reaches about 25% in FY27.

You wrote it down.
Every filing is read against it.

A worker polls the live filing stream every five minutes, not on a schedule someone has to remember. Most filings pass through dark. When one bears on an assumption you hold, that assumption alone answers, with the quote that drove the call and the filing attached.

LUPIN · the filing stream · classify_evidence_impact

May 2026JunJulAug 2026
7 May 2026FY26 results · integrated filing · 44 lines as filed · passed through 6 Aug · 22:02Q1 FY27 results · revenue ₹8,277 cr, +32.0% YoY · EBITDA margin 29.6%, +204 bps YoY, −366 bps QoQ bseindia.com/xml-data/corpfiling/AttachLive/5b9e63ec-7102-47fa-b4c9-37b33c391a83.pdf 12 Aug · 17:47Transcript of the Q1 FY27 call · 18 pages · read by the promise ledger
01EBITDA margin reaches about 25% in FY27needs review · 52
02Revenue grows high single digitneeds review · 52
03No generic entry on Tolvaptan or Mirabegron before FY28passed through
04Pithampur remediation closes without an import alertpassed through

impact ∈ {supports, weakens, invalidates, needs_review} · key signal: margin · a directional call under 55 is demoted to needs review · a quote not copied verbatim from the filing is discarded · thesis health 70 → 66

needs review · 52

Promises

What was said. What was filed.

A guidance line is captured verbatim from the call with the date it was filed and held open. When the number that settles it is filed, one to four quarters later, the analyst marks it delivered, delayed or missed against that filing. Nothing is marked kept until a filing says so.

“We would like to reiterate our earlier guidance of high single-digit revenue growth for the organization with EBITDA margins of around 25% during the year.”

Vinita Gupta, CEO, LupinQ1 FY27 earnings call · 7 August 2026 · page 5 of 18filed with NSE 12 Aug 2026, 17:47

Ramesh Swaminathan · page 10

“We are being a little more cautious … the EBITDA margins would be in the range of 24% to 25%.”

Same page as the guidance · bears on assumption 03

“…the overall sales in the U.S. will be impacted by competition in key products like Mirabegron and Tolvaptan this year.”

buyside_management_promises · LUPIN
promise_text "We would like to reiterate our earlier guidance of high single-digit revenue growth … EBITDA margins of around 25% during the year" metric margin filed_at 2026-08-12 17:47 status open
openon trackdelayedmisseddelivered
filed so far Q1 FY27 · EBITDA margin 29.6% settles on FY27 results · not yet filed status is changed by the analyst, against the filing · automated settlement is on the bench

Checks that run anyway

Scored by the business it is actually in.

The computable items from nine frameworks run on every covered company without being asked, each with the number it came from. A metric the filings do not carry is unknown, never a pass. And a manufacturer’s rules once flagged a wealth manager for an implausible 65% margin, which is simply what that business earns, so financials are never scored by that lens.

LUPIN · manufacturing lens

margin · leverage · cash conversion · dilution

cum CFO / PAT, 8 yearscritical1.93×pass
borrowings / EBITDAcritical0.58×pass
ROCE32.7%pass
3-year capex / depreciation1.25×pass
share count CAGR, FY19–FY260.15%pass

HDFCBANK · bank lens

basis: class-correct metrics from filing XBRL, not the manufacturing lens

gross NPA, % of advancescritical1.15%pass
net NPA, what provisioning has not coveredcritical0.38%pass
assets / equitycritical7.8×pass
net interest margin3.11%pass
cost to income, median39.65%pass

In short

One lens for everythingFiDUS
A bankInsolvent at 7.8× assets to equityLevered by design; leverage flagged only past 20×
A wealth managerImplausible 65% marginFee subtype: no loan book, so no credit checks
An NBFCOne Borrowings lineFunding leverage on all three Ind-AS lines
Class fromA sector label someone typedThe XBRL taxonomy the filing declares
A missing metricZeroUnknown, counted, never a pass

It produces evidence. It never produces a recommendation. The judgment stays with the analyst.

By construction

It never tells you what to do.

Buy / Sellthe one output it is built not to have

Filedthe only thing a signal can settle against · never a price, never a consensus

thesis health 66 · starts at 70 · moved only by evidence the analyst has not rejected · a number to argue with, not to act on

Never learns from the share price.

Every signal is settled by a filed number. A quote the model did not copy verbatim from the filing is discarded before it is shown.

Never accuses.

Written into the forensic framework itself: “a research checklist, not a fraud accusation engine.” Outputs say risk flag or needs review, and a high-severity flag triggers deeper review, not a decision.

Not built, deliberately.

Promise settlement against the filed number is the analyst’s today. There is no structured falsifier on an assumption yet, and no accuracy figure is published for the classifier until its evaluation has been run.

The library

Nine frameworks. 143 items.

Each drawn from a named body of published investing and forensic-accounting work. The computable items run on every covered company; the rest are surfaced as questions for the analyst, never scored. Plus 18 sector operating-KPI libraries, every claim carrying the management quote it was read from and its page.

31

Forensic fraud checklist

Cash conversion against profit, dilution, goodwill against net worth, dividend behaviour under stress.

two critical items
27

Financial statement analysis

Revenue quality, margin drivers, balance-sheet stress. The largest single library.

15

Unified forensic-alpha

Earnings quality and financial shenanigans; receivables and inventories against sales.

O’glove · Schilit
14

Management quality

Owner orientation, capital returned against raised, auditor and related-party signals.

14

Behavioural

Where a thesis is most likely to be fooling its author, as questions the analyst answers.

analyst-read · no filing trigger
12

Business quality and moat

Return on capital through cycles; cash backing of reported profit.

11

Capital and credit cycle

Capex against depreciation, leverage into a downturn, position in the cycle.

10

Valuation

Expectations embedded in the price against what filings support.

never a target price
9

Capital allocation

Buybacks, dividends, acquisitions, and whether shareholders were diluted to pay for them.

Where it sits

Not a screen. Not a feed.

The other ways a desk watches a position: a price alert that fires on a move already priced in, or a filing feed you read yourself and hope you do not miss the sentence that matters. Here is where each one stops.

A price alertA filing feedFiDUS Buy Side
Watchesthe share priceevery documentyour reasoning, as the assumptions it rests on
Fires ona move already priced ineverything, so nothingthe filing that bears on an assumption you hold
Saysup or down—supports · weakens · invalidates · needs review, with confidence and the quote
What was promised—a transcript to readthe line verbatim, its filing date, a status the analyst settles
Forensic checks——a 143-item library · the computable ones run unprompted, each with its number
A banksame alertsame feeda different lens: NPA, funding leverage, cost to income
A missing metric—a blankunknown, counted, never a pass
Buy or sellimplied—never, by construction
When you doubt ittrust the chartread it all againopen the filing · it is attached to the signal
Every FiDUS cell is an output on this page or a documented contract in the monitor.captured 16 Sep 2026

What it does not carry, by design: broker estimates and consensus, price targets, intraday ticks. If it was never in a filing, it cannot settle a signal.

Setup

Hear it where your desk is.

Write the thesis once in the web app. Alerts arrive on Telegram with the filing attached, and the same monitor is reachable over the API for a desk that runs its own tooling. Hosted, in development, access by request.

Next on the bench

Built in the open.

Not shipped yet, listed anyway, so everything above stays exactly what the monitor does today.

Promises settled by the filing.

Last year’s guidance checked against this year’s filed number automatically: met, missed, or quietly dropped.

A falsifier on every assumption.

A filed number that breaks an assumption breaks it without a model in the loop.

Alerts weighted by position.

Today a filing on a six percent position and the same filing on a watchlist name are indistinguishable to the alert engine.

Design-partner access

Your thesis, and what happened next.

For funds, PMS and family offices. Send a thesis you already hold, two or three sentences, and we return the assumptions it extracts and what the last four quarters of filings did to each one.

Get access

Contact

Siddhi Nagvekar

FiDUS · design-partner access

Contact

Shreyas Mashelkar

FiDUS · design-partner access

Request
access.

Tell us who you are and which product you want to see. A person replies, not a sequence.