Why We Killed the Winning Score (and How to Read the Tape Instead)

8 min · interactive · Intermediate

For eight months, Majorka's headline metric was the Winning Score — five weighted inputs collapsed into one tidy number. On June 10, 2026, we killed it. Not quietly deprecated: killed, with receipts. This lesson is the post-mortem and the replacement — how to read the tape: receipts, 7-day adds, cooling, and the discipline of saying "not enough evidence" out loud.

What the score promised

One number, sortable, no thinking required. That's the pitch of every spy tool's score, including our old one. You opened the dashboard, sorted descending, and the "best" products floated to the top. It felt like research. It compressed a hard, uncomfortable question — is anyone actually buying this right now? — into a green badge.

The Winning Score (retired) blended five weighted inputs into a single 0-100: order velocity, margin potential, supplier reliability, market demand, competition density. It looked rigorous. It had a table of weights. Operators built entire shortlists on it — this very lesson used to teach you to filter at "Score ≥ 80" and never launch below 75.

We're not telling you this to dunk on a dead feature. We're telling you because every other tool you will ever use still works exactly this way, and the failure pattern is identical.

Why composite scores lie

Weights are opinions wearing a lab coat. Our score said Velocity 40%, Margin 20% — but nobody ever validated 40 over 30, and when we audited the velocity inputs themselves, they didn't survive. A composite hides its weakest input behind its strongest. The number looks precise; the precision is costume.

Walk through what actually happens inside a composite:

The weights are unfalsifiable. Is velocity worth 40% of a buying decision, or 25%? There is no experiment behind the number — someone chose it because it felt right. Change the weights and the "winners" reshuffle, which tells you the ranking was never a property of the products. It was a property of the opinion.

One rotten input poisons silently. On June 10, 2026 we audited our own velocity columns — the 40%-weighted, highest-leverage input — and refuted them. The stored values didn't survive contact with the raw snapshots they were supposedly derived from. Every score that ever leaned on them was contaminated, and the score's tidy 0-100 surface gave you no way to see it. That audit is why the score died the same day.

A score can't say "I don't know." A composite always returns a number. Thin evidence, contradictory evidence, no evidence — the formula doesn't care; it outputs 73 with the same confident typography as a number backed by a thousand observations. The most important thing a research tool can tell you — "the evidence isn't there" — is structurally impossible for it to say.

So we stopped saying it. Modeled scores were purged from verdicts. What replaced them is deliberately less impressive-looking, and that is the point.

What replaced it: the tape

The tape is the stream of things we actually observed on real listings, with dates attached. Four words carry the whole vocabulary:

Receipt — an observed, dated event on a real listing — an order add we actually saw, not a number we modeled. A receipt can't be reweighted, recalibrated, or argued with. It either happened on that date or it didn't.

7-day adds — the orders a listing added in the last seven days. This is the live read. A listing with 80,000 lifetime orders may have done all of them last year. Seven-day adds tell you what's happening now; lifetime counts tell you what already happened to someone else. When this lesson's old version told you to "catch velocity early," this is the honest version of that instinct: recent, dated adds, counted — not a modeled growth coefficient.

Cooling — when the adds a listing was putting up stop arriving — the tape going quiet on something that used to print. Cooling is the single most valuable warning in dropshipping, because the famous products — the ones every spy tool still ranks highly on lifetime volume — are very often the cooling ones.

Verdict — a call built only from receipts. If the receipts aren't there, you don't get a softer score. You get an abstention.

Notice what's missing: no weights, no 0-100, no tiers. The tape doesn't rank the whole catalog for you, and that loss is real — sorting by a score was fast. But it was fast the way a fabricated answer is fast.

Abstention is a feature

On June 10, the system looked at 395 candidates for a cooling call. It served 1 and suppressed 394. The 394 weren't "fine" — they didn't clear the evidence bar, and the honest output for "I don't know" is silence, not a confident guess. Every tool you've used before filled that silence with a score.

Sit with the ratio for a second: one call in 395. A scored system would have ranked all 395 and let you act on the top decile. Most of that decile would have been noise dressed as signal — and you'd have paid for the costume in ad spend.

So when Majorka abstains on a listing you paste, read it correctly:

An abstention is not a no. It's "not enough evidence yet." Park the listing and re-check in a few days — evidence arrives or it doesn't.
An abstention is not a malfunction. It's the system refusing to do the thing we killed it for doing.
A tool that never abstains is not more capable than the tape. It is less honest than the tape.

Read one yourself

The single paste is free. Here's the walk:

Find a candidate listing on AliExpress — something from your Lesson 2.3 or 2.4 shortlist.
Paste its URL into Majorka's search. The system pulls what we've observed on that listing.
If receipts exist, you'll see dated order-add events — each row is a thing that happened, with a date. Read the 7-day adds against the lifetime total: recent-and-arriving beats big-and-stale, every time.
If the tape is quiet, you'll get an honest miss: "not enough evidence yet." No score, no guess. That message is the product working, not failing.
If the listing is cooling, treat it as the strongest signal on the page. Something that used to print and stopped is the late-curve trap from Lesson 2.2, caught in the act.

What's free and what's paid, stated plainly: the single paste and its reading are free. Running a board of your own listings with alerts on them is the Scale plan — the lesson you just read is free forever either way.

Why this matters

The Winning Score post-mortem is not really about Majorka. It's about the question you should now ask of every research tool, including ours: is this number something you observed, or something you modeled? Demand the date. Demand the event. And when a tool can't show you either, treat its confidence as decoration. The operator who reads receipts — and respects abstentions — enters products on evidence. Everyone else is sorting by an opinion in a lab coat.

June 10, 2026 — one call served, 394 suppressed

This is the day the discipline became visible in production, and it's worth replaying as a worked example because the numbers are exact and ours.

The cooling detector looked at 395 candidate listings — products whose tape history made them worth checking for a "this has gone quiet" call. The old composite would have scored all 395 and surfaced a ranked list. The tape system instead asked one question per listing: do the receipts clear the evidence bar for this specific call?

Result: 1 served, 394 suppressed. Hand-verified, exact.

The one served call carried its receipts with it — the dated adds that used to arrive, and the silence where they stopped. Anyone could audit it. The 394 suppressed weren't ranked lower; they produced nothing, because "probably fine" and "probably cooling" without evidence are both guesses, and guesses had just gotten our headline metric executed.

The operator takeaway: when you paste a listing and get an abstention, you are seeing one of the 394. The system had a chance to impress you with a confident answer and declined. That refusal is what you're actually paying attention for — it's the difference between research and theatre.

Action items

Paste one candidate from your current shortlist and read its tape end to end.
Write down, in one sentence, whether the verdict came from receipts or the system abstained. If it abstained, write the date you'll re-check.
Re-read the same listing after 7 days. Did adds arrive? Did the abstention resolve? That delta is the velocity lesson from 2.2, observed instead of modeled.
Take any spy tool you currently pay for and ask its headline number the post-mortem question: observed or modeled? If you can't find a dated event behind it, you know what the number is wearing.

Next lesson: the live walkthrough — building a 20-product shortlist in under an hour using Majorka, AliExpress, TikTok and Meta Ad Library together. The full research workflow, end to end.

Sources

Majorka engineering — Winning Score retirement and velocity-column audit, June 10, 2026
Majorka cooling detector — 395 candidates evaluated, 1 served / 394 suppressed (hand-verified), June 10, 2026
Composite scoring vs single-metric ranking — academic e-commerce research, Manning et al 2017

Module 02 — Product Research — Reading the Tape Before the Crowd

The hardest skill in this business. Data-driven frameworks for spotting products at the beginning of their curve, not the end.

Lessons in this module

The 4 Types of Winning Products (and which you should pick) · 11 min
Problem-solvers, wow-factor, impulse, evergreen — the trade-offs of each.
Trend Velocity — Catching a Winner at Day 10, Not Day 60 · 13 min
How to read a velocity curve and when to pounce.
AliExpress Signals That Actually Matter · 9 min
Ignore reviews. Watch orders, store age, and "recently ordered" pulse.
TikTok Search for Product Discovery (the right way) · 10 min
The search strings that surface rising products, not viral replays.
Meta Ad Library — Reverse-Engineering Competitor Winners · 12 min
How to tell a test from a scale, and steal the ad angle without the copy.
Why We Killed the Winning Score (and How to Read the Tape Instead) (this lesson) · 8 min
The post-mortem of our own headline metric — and the receipts, cooling signals and abstentions that replaced it.
Building a 20-Product Shortlist in Under an Hour · 15 min
Live walkthrough: from dashboard to validated shortlist, fast.
Beyond AliExpress — Apps That Surface Different Winners · 9 min
Sell The Trend, Dropispy, AdSpy — when each adds signal Majorka does not.
The 1000-Product Shortlist Problem · 11 min
How to filter from 1000 candidates to 5 testable products in under 90 minutes.