bd@
The public premise, tested
Korean peak rank did not rank-order U.S. Top 10 entry
The test covers 1,618 title-season-category records that appeared in Korea across five years of Netflix weekly Top 10 data. The score was Korean peak rank and the label was U.S. Top 10 entry.
The threshold was fixed in writing before fitting. The result does not describe buyer judgment: per-title availability, marketing, placement and every internal buyer input are unobserved.
FIG 1 · The sample
Two hundred and seventy records reached number one in Korea. Two hundred and three never entered the U.S. Top 10
203 of them never entered the U.S. Top 10. This count is descriptive. The model result below uses all 1,618 title-season-category records that appeared in Korea.
What came back
- Every record −0.04
- nothing threshold premise dead
- Every record 0.501
- 0.40 a coin 0.60
Point estimates · 1,618 records · 94 country charts · 497,320 chart rows · one source on both sides
How the pre-registered threshold is applied
| Out-of-sample R² | Verdict, fixed in advance |
|---|---|
| Below 0.20 | Korean chart data alone does not clear the threshold. Continue to a direct audience test |
| 0.20 to 0.50 | Inconclusive. Widen the sample and run it again |
| Above 0.50 | The tested public signal clears the threshold. Stop |
Why a zero here is a finding and not a failure
Handed every country except America, the same estimator reaches an R² of 0.85 and an AUC of 0.99. That control is close to circular by construction, and it shows the pipeline can recover structure when the supplied variables carry it. The comparison supports reading the first result as specific to Korean home performance rather than as a pipeline failure.
FIG 2 · Robustness
Availability is unobserved. Narrower exposure screens did not resolve the null
Per-title U.S. availability is not in the public data. Restricting the sample to records that reached at least forty country charts, or that appeared in at least one of Britain, Canada, Australia or Ireland, moved the AUC between 0.39 and 0.55. The result did not resolve into an ordering signal.
| Check | What remains true |
|---|---|
| Exposure screens | Records reaching at least forty country charts, or appearing in Britain, Canada, Australia or Ireland |
| AUC range | 0.39 to 0.55 across the restricted samples |
| Still missing | Whether each title was available in the United States, plus marketing, placement and internal buyer inputs |
FIG 3 · Scope
What the test covered
Six things a buyer can see before they pay.
- Home chart runtested
- Cast recognitionnot tested
- Budgetnot tested
- Festival recordnot tested
- Genre comparisonsnot tested
- Pre-sale interestnot tested
1 of 6. The result covers one public signal and none of the buyer's internal inputs.
And five things this data cannot separate
- AvailabilityNot in the data. A title never licensed here could not chart here
- Below tenth placeInvisible. Eleventh and four-hundredth both read as absent
- Demand on its ownNot separated. A top ten includes what the platform promoted
- TimingCorrelation, not sequence. Many titles open the same day in both
- This instrumentUntested. That is the next study
It found one public signal incomplete. It did not find what completes it.