Back
Public prediction log

Receipts, not promises

Every pre-breakout alert our radar emits is timestamped, then re-measured daily against real YouTube view counts. This log shows them all — the wins and the misses. No cherry-picking: an honest signal has to survive in public.

Outcome tracking started July 2, 2026. Numbers get published here only once their sample is statistically meaningful — not before.

21.4×more likely to grow

Fresh window, zero overlap with anything published before. An early-phase alert kept growing over the next 5 days in 47.7% of cases, vs a 2.2% baseline across every tracked signal-day — 21.4× more likely to keep growing than a random trend. Small sample, so read the floor, not the headline: even at the bottom of the 95% confidence interval the edge is 9×.

Wide-sample cross-check (not independent — this window contains the earlier published ones): across Jul 2 – Aug 12, 2026 the same measurement gives 11.1× (43.6% vs 3.9% baseline, n=374 from 84 unique signals).

And on real YouTube view counts, re-polled daily after each alert: 66% of early alerts gained at least 20% more views (n=97 measured alerts across 35 unique signals, median +39%). No base rate exists for that one — we only re-poll videos we alerted on — so it is a real-world check, not a multiplier.

Backtest on 65 measurable early alerts from 30 unique signals, Aug 6 – 12, 2026 · baseline from 4,302 signal-day observations · baselines shift as the tracked pool changes, so compare alert-vs-baseline within one run, never across runs · hit-rates are window-specific too, so a bigger multiplier is a property of the new window, not proof of improvement.

A harder bar, published for the first time: alerts that grew at least 1.5× within the same 5 days. Across Jul 2 – Aug 12, 2026 that happened to 3.7% of early alerts vs 0.45% of all tracked signal-days — 8.3× more likely. Read this one carefully: it rests on 14 wins (95% CI 4.1×–16.7×), and it comes from the same wide window as the cross-check above — so it is not independent of what we published before. We held it back for four runs until the interval cleared 1× twice in a row.

Re-measured weekly from a fresh backtest — next run: Aug 20, 2026.

The window is the product.

Volume-ranked trend feeds surface a topic at its loudest — which is often after its window has closed. This log works the other way: the call goes out while the signal is still climbing (phase-labeled), the outcome is re-measured in public, and each entry below shows whether its window is still open today.

Put the proof on your own page

A live badge that reads from this log. It refreshes on its own and never shows a hit-rate before the sample is real.

Lolik Receipts — live proof badge
<a href="https://lolikai.com/receipts">
  <img src="https://lolikai.com/receipts/badge"
       alt="Lolik Receipts — live prediction log" width="320" height="96">
</a>

Method: alerts are emitted by the Signal Vault each morning (phase-labeled: early = accelerating below its prior peak, momentum = at its all-time high). Views are re-read daily from the YouTube Data API. Topic-band alerts (search/news) aren't view-trackable yet and are excluded. Full methodology →