ExploratoryLegacy historical outcome · not a recommendation

In this legacy historical subset, opposing the displayed direction scored more often than following it.

Across 61 qualifying Entertainment-containing events, 60 had a non-zero directional legacy outcome: 21 followed the displayed BSP direction, 39 opposed it and one was Neutral.

Status: Exploratory
Dataset: bsp-source-audit-v2
Parser: bsp-mailbox-parser-v2

Directional sample6061 outcomes; 1 Neutral

Opposite score65.0%39 of 60

Comparator50.0%Simple directional baseline

Canonical outcomes0Required market inputs absent

Examine it

What was counted?

Consolidated events whose normalised industry label contains Entertainment and has a legacy workbook-linked PQ outcome plus a non-tied displayed direction. Zero moves are Neutral and excluded from the directional rate.

Displayed direction35.0%

21 of 60

Opposite direction65.0%

95% Wilson interval: 52.4%–75.8%

Under a simple fifty-fifty binomial comparison, the unadjusted two-sided p-value is 0.0273. That asks how surprising this split would be under that narrow model. It does not correct for the wider feature search or prove a future edge.

Make it clickWhat does this p-value actually say?

A p-value asks how unusual the observed result would be if a particular chance-only model were true.

Read the full guide

Taxonomy matters

“Entertainment” is not one automatic bucket.

Exact label71.7%

38 of 53 directional; one Neutral

Label contains Entertainment65.0%

39 of 60 directional; one Neutral

The exact label produces the stronger historical rate, while the broader definition also includes Movies & Entertainment. Both are disclosed because changing the taxonomy changes the claim.

Stability

The yearly slices point the same way, but each is small.

202361.1%n = 18

202476.9%n = 13

202564.7%n = 17

202658.3%n = 12

The opposite-direction rate ranges from 58.3% to 76.9% by year. None of these small yearly slices is decisive on its own, and the 2026 slice ends in May.

Validation

It also appears later—but later data was already exposed.

A neutral calendar split uses discovery through 2024-12-31 and a later period from 2025. The hypothesis was inspected across the full archive before this split was frozen, so the replay cannot be called out-of-sample.

Discovery through 202467.7%21 of 31; 1 Neutral

Later, already exposed62.1%18 of 29

Exposed replay — not out-of-sample. A future candidate must be frozen before genuinely unseen canonical-price outcomes are examined. Changing the rule after review creates a new version and requires a new validation mechanism.

Multiple testing

One interesting result came from a wider search.

This observation belongs to the Industry × displayed-direction features family. Tradour reports 50 related industry groups at the declared sample threshold. Its p-value is therefore descriptive, not a Playbook promotion rule.

Exploratory scan: No p-value promotion; freeze a canonical-price candidate before genuinely unseen validation.

What could make this misleading?

This finding is not ready for promotion.

This is an in-sample historical comparison, not prospective evidence.
The legacy PQ outcome is linked from a later BSP email; its price source and economic horizon are not independently established.
The canonical first-public-information-session outcome cannot be calculated from the supplied files because auditable market prices and publication/session metadata are absent.
The result sits inside a wider feature scan, so its unadjusted p-value does not resolve multiple-testing risk.
The full archive was inspected before this validation structure was frozen; the later-period replay is exposed and not out-of-sample.

Audit it

How was this produced?

Direct Daily Stock Market Forecast emails were parsed across three observed HTML schema families. Forwards and replies were excluded; no APR24-15M or Gainers / Losers messages were present.

All 9,177 parsed rows agreed with the supplied workbook on every compared core field. The global raw/displayed inversion is explicit.

Legacy outcomes reproduce the workbook's next-quarter email link over a 45–140 day gap. The canonical Day 1 definition is versioned separately and remains unavailable until auditable prices, calendars and publication timestamps are supplied.