What the Blitz Index Measures — and What It Doesn't

In a single day we pre-registered three hypotheses about score-based entry signals, ran them against the data, and rejected all three. This is the synthesis: what the failures have in common, what the score genuinely does, and the one lead that is still alive.

7 min read · 3 hypothesis tests · pre-registered 2026-10-13 · Not investment advice

One day, three rejections

The research program behind this article began with a straightforward question: if the Blitz Index score measures how well a company is executing a blitzscaling strategy, can score-derived events tell you when to act? Three candidate answers were formalized, registered in HYPOTHESES_PREREG_2026-10-13.md before any data was touched — definitions, entry rules, and falsification criteria frozen in advance — and then tested. The falsification rules were mechanical, the kind that leave no room for post-hoc reinterpretation: if the event-return median came in at or below the benchmark, reject.

Each test was published in full as it completed — the profitability-inflection study ("The Profitability Turn Was Not the Buy Signal"), the score-momentum study ("A Rising Score Did Not Predict a Rising Stock"), and the threshold-crossing study ("Crossing 65 Was Not a Buy Signal Either"). What follows is the combined scoreboard, then the pattern that connects them.

Three rejections in one day would be a bad day for a vendor. For a research program, it is a clean day — every criterion was fixed beforehand, every failure is on the record, and the failures rhyme. The question is what they rhyme with.

The common thread: price is decided at entry

In all three studies, the same force explained the results better than any score-derived signal: the entry multiple and the market cycle. The score-momentum test's negative-Δ winners — Palantir 2022, DoorDash 2022, CrowdStrike 2022 — were all entered at the 2022–23 bear-market trough. The threshold-crossing test's few +3-year winners were all entered at cycle troughs (the COVID collapse, the 2022 lows), making the crossing signal inseparable from the multiple. The profitability-turn study found that by the time an inflection is announced in a 10-K, the expectation is already in the price. Different signals, same mechanism: a great business bought at despair and the same business bought at euphoria are not the same outcome at all, and the score knows the difference between the businesses but not between the prices.

The second shared lesson is statistical hygiene: means lie here; medians and win rates tell the truth. Winner-take-all markets produce thick tails — AppLovin +564.9% and Palantir +345.7% in the inflection panel — and two or three such observations drag any mean into positive territory while the median and the SPY beat rate stay negative. That is exactly why the pre-registration mandated medians and win rates as primary statistics before the data was opened: it removed the temptation to let the tails do the arguing.

A third point, worth saying plainly: these rejections are unusually sturdy. Two of the three panels (the estimated-score panels behind H1 and H3) were constructed with knowledge of outcomes — a hindsight bias that works in favor of the hypotheses, since a known-good year invites a generous score. Claims that fail with the wind at their back are more robust than claims that fail against it.

What the score actually does: it finds the failures

If this were the whole record — three rejected timing signals — the natural conclusion would be that the score is useless. The opposite record says otherwise, because the score was never built to time entries. Look at what it was built to do and the evidence is consistent:

It grades business failure. The 2019 blind score for WeWork was 29/D; the company collapsed to $45 million from a $47 billion valuation. In the v1.1 valuation study's D-grade cohort (10 observations), seven fell 70% or more. In the ARK 2021 backtest at the cycle peak, the score avoided five of the seven deep blowups — including Invitae's bankruptcy. One caveat, reported in the same spirit: Blue Apron's stock tripled after acquisition despite the business dying — the D grade predicts business failure, not shareholder returns. That distinction is the point, not a footnote.

The gates work. The G1 gate has now rejected three pre-revenue hopefuls on identical grounds — OKLO (no power revenue, NRC application not even filed), JOBY (no commercial revenue, no quantified backlog) — while conditionally passing ASTS ($1.3B backlog, satellites in orbit) and CIFR ($11.4B contracted demand, deliveries started). The gate's question is always the same: is there quantified, contracted demand? Answering it uniformly is what makes it a gate rather than an opinion.

So the scoreboard reads like this: the Blitz Index measures business quality — with demonstrated, registered evidence on the failure side — and it does not measure price timing. A business scoring 65+ has crossed into what the model considers blitz-quality execution. Whether that business is cheap or expensive is a different instrument's job.

The one lead still alive: H4

There is exactly one score-plus-price combination in our research that has not been rejected: the expansion zone from the v1.1 valuation study — score ≥ 65 combined with a low P/S percentile relative to the company's own five-year history. In the test split it averaged +249% over three years (median +102%), with five of seven observations expanding. It is also the only direction consistent with the ARK Test 2 result at the 2022 trough, where correlation between score and forward return flipped positive (r = +0.307): quality combined with a low multiple is the pattern that keeps surviving.

But it survives on a sample of eleven — six expansions out of eleven zone observations. That is a lead, not a finding. The registered next step is H4: revalidating the expansion zone on the 150-observation panel (the Track 2 effort), adding ten to twenty new historical blind observations. If it fails to replicate, that failure will be published with the same ceremony as these three. A research program that only publishes its wins is an advertisement.

BlitzLab's position

Three things follow from today's results, stated as commitments rather than conclusions.

First, scores will continue to be published with price context, never as entry signals. The site shows the Blitz Index score alongside financial data and valuation percentiles; the score grades the business, the percentiles describe the price, and the reader combines them at their own responsibility. Nothing on this site will ever tell you when to buy or sell.

Second, the hypotheses come back when the data is bigger. All three rejected tests were flagged for retest on the expanded panel: the H2 verdict was sensitive to the n=9 paired sample, the H3 verdict needs a wider Δ distribution than the −9-to-+19 range this panel offered, and the H1 verdict rests on seven crossings. Rejection on small samples is a scheduling decision as much as a scientific one — and the registration files will be waiting, unchanged, for the rerun.

Third, the pre-registration discipline stays. The reason these three rejections read cleanly is that no definition moved after the results were known — the profitable-turn test kept its painful sensitivity caveat, the crossing test kept its n=4 trough confound, the slope test kept its untestable sub-claim. That discipline is now the program's house style for every hypothesis that follows, including H4.

The one-sentence version of this whole day: the Blitz Index tells you which businesses are executing blitz-quality strategy; it does not tell you which of them are cheap. Knowing the difference is the entire point of publishing the rejections.

BlitzLab scores businesses against a scholar's standard — not investment advice. This article synthesizes three pre-registered hypothesis tests (H2, H3, H1; registered 2026-10-13; verdicts: all rejected). It reports descriptive statistics of past patterns. Past patterns are not predictions. No price targets, no buy/sell recommendations. Figures: GAAP, SEC-filed; estimates labeled as estimates. Companion pieces: "The Profitability Turn Was Not the Buy Signal", "A Rising Score Did Not Predict a Rising Stock", "Crossing 65 Was Not a Buy Signal Either".