How to Separate Facts, Inference and Scenarios in Company Research

A fact is in a source with a figure attached. An inference is a step the reader took. A scenario is a conditional with its assumption named. Applied to one published year, those three tests leave two objects falling through all of them: an estimate that says so, and a subtraction that is correct and invalid. Three labels, five cells.

A classification carries as many cells as its tests produce. The three labels arrive first. Somebody wrote them down before meeting a single object, and the objects arrive afterwards, from the world. When an object passes none of the three tests, the honest move is to draw another cell, and the tempting move is to file it in the nearest one. Every defect examined here is produced by that second move rather than by any error of arithmetic or of reading.

One object in these notes already does the whole of this on its own. What is it doing?

An object comes before a word. Under how a standing bill amplifies a movement in revenue, these notes carry one drawing with one caption, and between the two of them they carry three separate things that are worth pulling apart slowly.

The first is a published total. A stationery business called Anjani Stationers Private Limited saw its standing baseThe part of a year's costs that arrives whether the works runs flat out or barely at all: rent, salaried people, insurance, the write-down on machinery. rise by Rs 24,40,000/- in a single year.

The second is a division of that total into three parts, laid out below in the words the source gives to each of them. Taken together, the three land on the total exactly, to the rupee, with nothing left over and nothing missing.

The third thing is a sentence, and it is the one most readers skate past. The sentence settles whether the division counts as a disclosureAnything a business put in writing itself, either in its published accounts or in the notes behind them. Whatever a reader afterwards derives from those accounts is not one, however carefully derived. at all. The source carrying that division prints these words alongside it, and they are its own: the split is an estimate, not a disclosure. Further on it puts the point the other way about, and the exact wording matters: the components are on the face of the statementThe main body of a published statement of profit and loss, where the prescribed line items are printed, as opposed to the notes that sit behind it and expand on some of them. even though the split is not.

One object carries a fact, a step somebody took, and the label that keeps them apart, and that label is doing work no digit in the drawing can do. Which raises the obvious question. If the three parts add up exactly, what is left to doubt?

The addition proves the addition. The addition says nothing whatever about who did the dividing. A month's electricity bill on the kitchen shelf makes the same point, with a note beside it in somebody's handwriting saying that roughly half of it went on the cooking. The total came off the bill. The half came off a memory of the month, and the word roughly is the only thing on that shelf that separates the two.

What are the three labels, and what is the test for each one?

Three definitions would be quicker to write and quicker to forget. A definition is a thing to nod along to. A test is a thing that has to be run on a line while the line is still in front of the reader. So three tests follow instead.

A fact is in a source with a figure attached. The test is whether a stranger handed the same source would write the same line. The bar is low, and a low bar is what is wanted: it keeps out everything the reader supplied and leaves in everything the source did. The harder half arrives straight away, and the harder half is where facts get manufactured most often. A stated absence is a fact only after a search, and never instead of one.

An inference is a step the reader took. The test is whether there is a step here the source did not take. Then comes the practical half: the step is named out loud, in the line itself, using the word that carries it. An inference with its step named is one of the most useful things a note can hold. An inference with its step hidden is indistinguishable, in writing, from a fact.

A scenario is a conditional about a published structure with its assumption named in the same breath. The test is a single question, taken from elsewhere in these notes rather than invented here: does the sentence need a future period to be true? A conditional does not. A conditional is true of the structure as it stands published today, whether or not anything ever moves.

Each of the three is answerable in about ten seconds. Ten seconds is the whole point of a test rather than a definition. Run them in that order, every time, on every line.

Run the three tests on one published year. What comes out?

What follows is five objects taken out of a single published year for one business, sorted one at a time rather than handed over as a finished table, because the two that fail only teach while they are failing.

Object one, two published amounts, and they are not the same kind of quantity. The first is the rise of Rs 24,40,000/- in what this business's costs come to before it sells a single register. The second is recorded elsewhere in these notes: a cash payment of Rs 21,00,000/- made at the start of the second year for 70 per cent of Chitra Binding Works Private Limited, a binding workshop. Both amounts sit in sources with figures attached, and a stranger handed either source writes the same line back. Gate one answers yes for each. Both are facts.

Object two, the three limb amounts out of the table further up. The three amounts stand exactly as they are, every amount still carrying the words its source gave it. The three limb amounts travel through all three gates without settling in any of them. The figures are printed in the source that carries them, so gate one looks satisfied for about a second, until the same source turns out to refuse to call that division anything the business itself stated. A stranger reading that source would not write a fact down. Gate two also says no. The reader took no step at all, and somebody else took the one step that was taken. Nothing about the division is conditional on anything, so gate three says no. The division passes none of the three cleanly, and a fourth cell gets drawn at exactly that point.

Object three, a sentence: the purchase caused the fall in profit. Every input underneath it is published. Operating profitWhat a year's trading left after the costs of running the business, taken before interest and before tax. In an Indian filing it is not a printed line and has to be assembled. was Rs 53,00,000/- and then Rs 41,50,000/-, a fall of Rs 11,50,000/-, and the payment for the shareholding is published too. No source anywhere writes that sentence, so gate one says no. Gate two says yes, and it says exactly where: the step is sitting inside the single word caused. The sentence is an inference, and the way to keep it useful is to leave the word caused standing in the open where a reader can argue with it.

Object four, another sentence: the fall would have been smaller had the largest account paid in the same time as the others. The sentence is a conditional. Its assumption can be named. The sentence carries no weight and no rank and no period. Gate three answers no to its own question, and no is the answer a conditional gives. The sentence is a scenario. The scenario cell carries the heaviest refusals of all, and what such a sentence may actually claim comes further on.

Object five, an operation rather than a sentence. The two amounts in object one, with the smaller subtracted from the larger. The arithmetic runs. The arithmetic returns an answer. And the operation is invalid. The two amounts are not quantities of the same kind, and why that matters is worked out properly further on. No source publishes that answer, so gate one says no. Gate two says no in a way worth pausing on. A step was certainly taken. Naming a step does not rescue it when the step was never a legitimate move to begin with. Gate three says no. The operation passes none of the three, and a fifth cell gets drawn at exactly that point.

Two objects came out unfiled. What does that establish about the labels?

Count what ended up on the board. Three labels went in at the top and five cells came out at the bottom. Everything else rests on the rule, so do that subtraction yourself and then say the rule out loud. A classification carries as many cells as its tests produce.

The rule is not a courtesy extended to untidy evidence. The rule is the arithmetic of the situation. Labels get written down in advance by somebody who has not yet met the objects, and the objects turn up afterwards, from the world, in whatever shape they happen to be. Nothing guarantees the two numbers agree, and on this year's evidence they came out two apart.

So what happens if the five are pushed back into three, one wrong filing at a time? Each one is a real defect with a cost that can be pointed at.

File the estimate as a fact and the label dies on the spot. The reader downstream takes somebody's working for the business's own statement, and nothing in the digits warns them. The digits were never wrong in the first place.

Filing the estimate as an inference takes credit for work somebody else did. The person who actually did the dividing drops out of the record, and out with them goes the one address a later reader could have written to.

Filing the invalid operation as a fact is what happens by default whenever nobody is watching. The arithmetic is correct. Not one digit objects to anything. There is nothing for a checker to catch, and that is precisely the problem.

Every one of those three defects is produced by the classification rather than by the reader, and the fix is therefore another cell and not more care. A form with three boxes for how the journey to work was made, walked or cycled or came by bus, handed to somebody who cycled half the way and walked the rest, shows the same thing. The person who cycled half the way is not being careless. The form has three boxes and the morning had two halves.

What is an estimate that says so, and why is it better evidence than a missing disclosure?

Somebody's labelled working beats a disclosure nobody ever made, and the thing doing the beating is the label rather than the figure underneath it. The claim reads backwards on first meeting, so work it through rather than accepting it.

A missing disclosure hands a reader nothing to argue with. There is no figure to check, no worker to question and no way for anybody to be wrong about it. The position feels safe and is in fact the weakest one a note can occupy. There is nothing there, so nothing can be built on it and nothing can correct it.

A labelled estimate hands the same reader three things they can actually do this afternoon. They can put it in front of somebody who knows the trade and ask whether that division looks about right. They can check it against the published components, printed on the statement itself even when the division of them is not. They can disagree with it outright, replace it with a division of their own, and write down why they moved it. A blank offers none of the three.

Hence the working rule, and it is short enough to keep. An amount and the words fixing its status are a single object, so carry both of them or carry neither. Carrying the three limb amounts forward without the words that call them an estimate is not a shortening. The shortening is a change of meaning made entirely by deletion, and nobody downstream can see that it happened. The digits are identical either way.

And then the sharper half, the sentence most likely to change how somebody works. An exact reconciliation is not evidence of a disclosure. A division worked out from the outside can be built to reconcileTo make two separately built figures meet, item by item, with nothing unexplained left over at the end. A statement reconciles when the parts and the total agree. to the last rupee, and this one was. A division that failed to add up would have been questioned inside a paragraph. A division that closes exactly looks like a statement of fact and gets waved through without a second glance. The precision is what makes it dangerous rather than what makes it safe.

When is a subtraction correct and still invalid?

No automated check catches this cell, so start with the rule and work the instance afterwards. Before subtracting two figures, say what each one is a quantity of.

Do that here, in words rather than in figures. One of the two amounts is a rise in a recurring annual cost base. The rise in the cost base arrives again next year, and the year after that, until something changes it. The other is a one-off cash outlay for a shareholding. The cash outlay happened once, on one day, and a balance sheet carries it as one line that will never come round again.

Neither one is a part of the other, and neither one can properly be taken away from the other. Their difference is therefore a number with nothing to be a quantity of, in the same way that the gap between a monthly rent and the asking price of a flat is a number with nothing to be a quantity of. Both are in rupees. The subtraction runs without complaint. The answer is not anything.

Now the part that earns this cell a block of its own, and the same part explains why the mistake survives being made. The answer that operation returns sits Rs 10,000/- away from a figure that genuinely is published, the goodwillWhat a buyer paid over and above the fair value of the share of net assets it bought. It appears only on a purchase, and it is arrived at by subtraction rather than by measuring anything. of Rs 3,50,000/- that arose on the purchase. Coincidental errors turn up already dressed as a confirmation, and that is what makes them the worst kind there is. Somebody who runs that subtraction and then finds a published amount sitting almost exactly on top of the result does not go back and ask whether the operation was allowed. The accounts have apparently just told them that it was.

Hence the general rule that comes out of it, and it applies far beyond this one instance. Ask which subtraction or which division made a figure, rather than asking which figure it turned out to be. A number keeps no memory of how it was arrived at, so the only place left to look is the operation standing behind it.

A number printed once gets quoted, and a number that means nothing is worse in circulation than in nobody's hands.

What may a sentence about something nobody observed actually claim?

That third cell has been standing on the board since the sorting, and it carries the heaviest refusals here, so take it up properly now.

The distinction is settled elsewhere in these notes rather than invented here. Set a forecast beside a conditional. The first asserts an outcome. The second asserts only the link running between a stated assumption and whatever would follow from it. Nothing at all separates the two in their arithmetic. Everything separates them in what is being claimed.

Then the refusals, and they are absolute rather than a matter of taste. No probability. No weight. No likelihood. No case named as the central one. No case named as the good one or the bad one. No ranking of one against another. No naming of a period nobody has lived through. And no set of cases presented as though it covered the possibilities.

Elsewhere in these notes a refusal takes exactly that shape, and the shape is worth seeing rather than the rule. Four paths are drawn, none is called likelier than another, none is called central, and none is called a scenario either. A great deal is set down while almost nothing is claimed, and the drawing is still useful.

The sorting does its best work of all on object four, so object four comes back up now. The sentence was: the fall would have been smaller had the largest account paid in the same time as the others.

Its assumption can be named, and here it is, entirely out of published lines. The Sunrise Public School group, an eleven year customer and the largest account on the book, was collecting at about 171 days against about 110 days for the other thirty five accounts, with Rs 38,00,000/- outstanding, and it stopped paying. The assumption is simply that it had paid in the time the others took. The sentence is a conditional. The conditional needs no period nobody has lived through. The conditional carries no weight and no rank. It passes.

Nothing published anywhere connects when that account paid to what the year's operating result was, so the sentence carries no number at all. The published evidence connects that account's collection periodHow long a customer takes to settle, worked out from what it still owes against what it was billed and expressed as a number of days. A comparison between customers, not a rule anybody agreed to. to the money still owed, and the money still owed is a different thing from the year's profit. So the honest output is the conditional standing on its own with nothing attached.

A sentence can be a perfectly well formed conditional and still carry no arithmetic, and the honest output is then the conditional with nothing attached. Building a case on a company's own published lines is covered separately under How to Build Business Scenarios for a Company.

Is an absence a fact?

Answer it in one line and do not soften it. An absence is a fact after a search and an inference before one. The same words, in both cases. The same tone. The same air of having read something. Only the history behind the sentence differs, and the history is invisible in the sentence itself.

A searched absence is among the most useful lines a note can carry. A searched absence hands the next reader one precise thing to go and ask somebody for, and a single printed sheet turning up can knock it over. Being knocked over that easily sounds like a weakness and is the whole of a searched absence's strength: a sentence somebody can disprove is a sentence somebody can build on.

An unsearched absence is a step the writer took, dressed up as a reading. An unsearched absence is the worst kind of inference on this board. Nothing in the sentence shows that a step was taken at all. Every other inference leaves a word behind, a caused or a therefore or a because, that a careful reader can catch. An unsearched absence leaves nothing.

So the practical form is one clause long. Where the search was made goes in the same sentence as the absence, together with the words that would have to turn up for it to be wrong. A sentence that cannot be disproved has not been offered as evidence.

The sentence there is no chemist in the lane is true if the speaker walked the lane end to end this morning, and an assumption if the speaker looked in from one end and saw no green cross. The two sentences are word for word identical and only one of them is worth anything to whoever hears it.

Anjani Stationers Private Limited, Chitra Binding Works Private Limited and the Sunrise Public School group are invented.
Educational material. Not advice on any investment, tax, budget or market position.