Is the quarterly score telling us anything?

After the Neuland case study, I had to ask the uncomfortable follow-up. Did the score trail actually tell us something there, or did it just look good in hindsight? One good example proves nothing. So I spent a day stress-testing it.

The method was simple. Take the companies with the longest score histories on the platform — we ended with 18 of them. For each one, line up three things side by side: our quarterly score trail, the company's actual reported profits, and the stock price. Then ask: who moves first, who agrees with whom, and where was the money?

What the score is

A thermometer for the business. For most companies, when the quarter's profits improved, the score went up that same quarter. When profits fell, the score fell. For Neuland and Navin Fluorine the match was almost one-to-one. That sounds obvious until you remember what it means: the score compressed six years of concalls and presentations into one line that tracked business reality — without anyone reading 24 transcripts.

What the score is not

A crystal ball. The score does not know next quarter. Navin Fluorine scored 8.0 on a call held just before its worst year began — because management didn't know either, and the score can only read what's in the documents. I tested this properly: the forward-looking parts of the score (guidance, Q&A tone) predict the next quarter no better than the backward-looking parts. If anyone tells you their sentiment model sees the future, hold your wallet.

A price signal. The stock sometimes moves before the score, sometimes after, sometimes ignores it for years. Time Technoplast's score stayed high and steady while its stock went up 12x and then drifted down 20% — the business compounded calmly through both.

The pattern that paid

Here's the interesting part. The money, in every case I found, was in disagreements between the score and the price.

• Neuland's score crashed to 3.8 within weeks of the stock's all-time high. The stock fell by a third over the next three months and went nowhere for a year.
• Navin's score recovered while the reported numbers still looked ugly. The stock roughly doubled after.
• TD Power printed its two highest scores ever right before profit growth jumped to +50% and +83% — and the stock did almost 3x after.

Three for three. When the trail and the market disagreed, the trail was right. That is a small sample and I am saying it carefully — but it's the clearest thing this research produced.

The shape tells you what you own

Plot the trail's 4-quarter average (it's now drawn as a dashed line on every company's score chart) and the shape itself speaks. A staircase means the business is changing gear — that was Neuland's mix shift. A V means a cycle turning — that was Navin. A flat high band means a steady compounder — that was TD Power, where flat is not boring, flat is the whole point. Sit.

When the score and the P&L disagree, check who's lying

Two companies looked like failures at first — NH and MCX, where the score barely tracked reported profits at all. I assumed the score was missing something. It wasn't. When NH's profit dropped by a third last quarter, the scorer had read the numbers, identified that a just-acquired UK business was distorting the consolidated accounts, and scored the underlying hospitals — which actually improved. The reported profit line was the misleading series, not the score. Same story at MCX with one-off technology costs.

I find this strangely reassuring. Every time I benchmarked the score against another series and they disagreed, it was worth checking which one was lying. Twice, it wasn't the score.

What's next

The "disagreement pays" pattern deserves a real test, not three good stories. That means writing the rule down before looking — what exactly counts as a disagreement — and then measuring what stocks did after every such moment across the whole history. It also means more history: we're re-scoring roughly 470 older quarters to triple the dataset (total AI cost: about two dollars, which still amuses me).

If the pattern survives that test, you'll read it here. If it doesn't, you'll read that here too.