Never delete a missed call — why the scorecard is built into the machine (Part 3)

Part 3 of five, on the system's heart: a mechanism that makes our own track record impossible to quietly edit.

The world of investment content has a curious custom: wins get told again and again, while misses quietly disappear.

You've seen the post that says "told you it would go up." You have never seen the one that says "everything I recommended last month went down." It isn't even malice — humans are simply built this way. Nobody wants to remember their misses, and nobody wants to hear about them.

My system has a feature called answer-checking — a scorecard. This chapter is about why I made a machine do it.

The easiest person to fool is yourself

Every day the system picks "dip candidates" — fundamentally healthy companies being sold off — from the day's decliners. When I started, I believed I remembered how my past candidates had done.

Then I made the machine keep the records, and compared. Memory and records were not even close. The winners I remembered vividly; the losers had simply vanished from my memory. Like a child who only remembers the subjects they scored well in.

Before you ever mislead a reader, you mislead yourself. That's why the scorecard runs on mechanism, not willpower.

What the machine's scorecard looks like

Every weekly report carries a section scoring the previous issue's candidates: subsequent returns and status — rebound confirmed / still monitoring / conditions failed — tabulated mechanically. My hands never touch it. Leave room for hands, and one week you'll inevitably think "I'd rather not publish this one."

Monthly, the horizon extends: how did past cohorts do, and how would the same criteria have performed across a decade of historical data.

Banning hindsight, structurally

Just as important as the scorecard is pre-registration.

When we test a new hypothesis — say, "dips after dividend increases perform better" — the hypothesis, the period, and the pass/fail criteria are registered in the system before the test runs. Changing the story after seeing the results ("well, what I really meant was...") is forbidden by construction, not by good intentions.

The consequence: most hypotheses die. The majority of everything we've tested came back "no significant difference" or "looks promising, only works in one specific era." A small number survived. But those survivors carry a fact with them: they survived without hindsight — and that fact is recorded too. That's where the value is.

Publishing misses looks like a loss. It isn't.

Honestly, publishing misses looks commercially foolish. A weekly that says "last week's candidates went down" cannot compete with a lively sales pitch.

Two reasons we do it anyway.

First, you cannot improve without a record of your misses. Example: declines in recently-IPO'd stocks look like dips but tend to be falling knives — a pattern that only became visible after stratifying our recorded misses. Today it lives on in the product as a rule: anything listed less than three months gets demoted to a caution label. Misses aren't shame. They're raw material.

Second, the readership it selects. Readers hunting for someone who "calls it right" eventually leave — nobody calls it right forever. But readers who want to watch the verification process stay. Those are the people I want to spend years with.

Not a predictor. A verifier.

My X profile says, in Japanese, "a verifier, not a predictor." That's not modesty — it's the design philosophy.

I could not build a machine that predicts the future. What I could build is a machine that remembers every judgment I've ever made and makes hiding from them impossible. For an individual investor, I've come to believe the second machine is worth more.

Next in Part 4, the grubby stuff: data breaks silently — the ingestion that stopped for 4.5 months, the corrupted financials, and the fake -91% crash caught 90 minutes before publication.

Tsukiyo Research publishes a weekly systematic report on Japan's uncovered small caps — every issue opens with the scorecard. Subscribe here.

This series is a personal build-and-operate log. Nothing here is investment advice or a recommendation to buy or sell any security.