How the Blitz Index Works

BlitzLab is a research project that scores companies on their blitzscaling strategy — how concentrated capital seizes a category, and whether the moat lasts. We call it the credit rater for blitzscaling. Every quarter we judge each covered company against the same 100-point ruler, the Blitz Index, using published filings and reported data. Every scorecard is public: score, grade, the item-by-item breakdown, the reasoning, and the score history. Companies that never make it through the gates get a judgment file explaining exactly why.

What blitzscaling is

Blitzscaling is the strategy Reid Hoffman and Chris Yeh describe in Blitzscaling (2018): a company concentrates capital and moves fast to seize a whole category before competitors can respond. BlitzLab exists because that strategy is worth measuring on its own terms — not by how much money a company made, but by how completely it is changing the world. The hero of this project is not the company that earned the most; it is the company that replaced an old order with a new one. That is why the third section of the index is creative destruction, and why the project’s premise, borrowed from Schumpeter, is that the final destination of concentrated capital is not simply profit but the creative destruction of an old order.

BlitzLab calls itself the credit rater for blitzscaling, and the comparison is deliberate. A credit rater grades and publishes its reasoning; it does not tell anyone to buy or sell. BlitzLab does the same for blitzscaling strategy: one ruler, applied the same way to every company, with the score, the grade, the item-by-item breakdown, the reasoning, and the history all public. The ruler is the Blitz Index. Its weights start from the Blitzscaling Scorecard published by Blitzscaling Ventures and were adjusted from there; the starting point and the reasons for the changes are part of the published method, not hidden inside it.

The index also differs from a traditional moat rating. A moat rating asks how defensible a business is. The Blitz Index asks something else: how fast the company is seizing its category, how durable the ground it has taken is, and whether it is destroying an old order in the process. Speed is measured, but it is deliberately capped — speed alone, the backtest showed, cannot tell a future winner from a future failure. And the same ruler is applied to listed and unlisted American companies alike, so a private company can be scored on the same terms as a public one, with the lower confidence of private-company evidence shown separately from the score itself.

Two gates first

Two gates must both pass before any scoring begins. Gate 1 — the business is real: product-market-fit evidence (for mature companies, sustained growth or a second growth curve). Gate 2 — there is room: a 10-year market ceiling at least 10× current revenue. A company that fails either gate is not scored.

Gate 1 is split by stage, because the evidence of a real business looks different for a young company than for a mature one. For an early-stage or growth company — revenue below $1 billion, or unlisted — Gate 1 asks for product-market-fit evidence: at least two of repeat purchase, net revenue retention, organic growth, and backlog. For a mature company, with revenue of $1 billion or more, Gate 1 asks instead for evidence that growth continues, or for the existence of a second growth curve. A mature company being judged on a second blitz is scored on that new curve, not on its past.

Gate 2 is also split. For the early-stage company it asks whether the 10-year market ceiling is at least 10× current revenue. For the mature company it asks the same question of the new market: the 10-year ceiling of the new market must be at least 10× current revenue. The point of the gate is room — a company cannot blitzscale into a market too small to hold it.

Gate 2 is judged on a revenue basis declared before scoring begins. The bar is current (or reference) revenue × 10, and what counts as revenue is settled first: treasury and other non-operating income are excluded, gross-recognised revenue is normalised, and gross principal-transaction amounts are rejected. The gate is then tested against the most optimistic case, not the base case: if even an optimistic build-up of the market falls short of the bar, the gate fails. And the verdict file for a failed gate records a re-review trigger — the specific disclosure or figure that would cause the company to be re-examined. A failed gate is therefore a judgment with a stated way back, not a permanent verdict.

A gate failure means no score at all, not a low score. The company is published as gate-failed, with a verdict file explaining exactly which gate failed and why. That distinction matters: a low score says the company was measured and found wanting; a gate failure says the company did not qualify to be measured. Twelve companies had failed a gate in the published record at the time the method note was written: three at Gate 1 and nine at Gate 2.

The 100 points

A · Speed — 25 points

Speed asks how fast the company is growing. It has four items. A1, revenue or ARR growth, is worth 10 and looks at both the level of growth and its trajectory — whether growth is accelerating or decelerating — with growth bought through acquisition separated out and read on its own, so that bought growth is not mistaken for organic momentum. A2, capital-raising velocity, is worth 5 and asks how quickly and on how favourable terms the company can raise the capital it needs. A3, hiring and productivity, is worth 5 and weighs the pace at which the organisation is expanding against productivity, including revenue per employee. A4, market-share movement, is worth 5 and asks whether share is actually moving, not merely whether revenue is rising with the market.

Speed is capped at 25 points on purpose, and the backtest is the reason. In the 20-company blind backtest, speed could not separate future successes from future failures: companies on the edge of collapse were still fast. WeWork is the standing example. So speed is measured, but it is not allowed to dominate the score; a company cannot blitz its way to a high total on growth alone.

B · Moat — 60 points

Moat is the largest section because the backtest found that discrimination happens here and only here. It asks how solid the ground is that the company has taken, across four items. B1, network effects, is worth 18. B2, distribution scalability, is worth 18. B3, market size, is worth 12. B4, gross margin, is worth 12.

B1 is scored against fixed anchors. The top band, 15–18, is platform-type lock-in: user growth structurally raises the value of the product for every user, and switching costs are high — NVIDIA’s CUDA triangle of developers, libraries, and hardware, the iOS App Store, and Facebook’s social graph at its peak are the anchors. The 10–14 band is two-sided market liquidity, where scale itself is the product: Uber’s city-by-city liquidity, Airbnb, and the Coupang marketplace. The 5–9 band is a data or ecosystem effect that exists but is weak, such as Tesla’s driving data or an AI model’s API ecosystem. The 1–4 band is a weak effect at the level of community or brand. Zero means no network effect at all.

B2 is scored against anchors of the same kind. The top band, 15–18, is distribution where the product itself is the distribution: viral spread plus riding a giant existing network, with customer-acquisition cost approaching zero — the early virality of ChatGPT and the Dropbox referral programme are the anchors. The 10–14 band is strong virality or a large distribution partner, such as Cursor’s developer-community virality. The 5–9 band is distribution that depends on paid marketing or sales but can be repeated and scaled, such as Palantir’s bootcamp sales motion — expensive, but repeatable. The 1–4 band means distribution is the bottleneck; zero means there is no distribution strategy.

B3 and B4 are quantitative rather than anchored. B3, market size, weighs the absolute size of the reachable market and the room left in it, checked against filings and market data. B4, gross margin, weighs the level and the trajectory of gross margin — whether economies of scale are proving themselves as margin. The signature of a true blitz, in this section’s terms, is accelerating growth and expanding margins happening at the same time.

Anchors are used under three standing rules. First, an anchor is an example, not an answer: the company being scored is matched to its nearest anchor, and the difference is written out in the rationale. Second, where a company falls between two anchors, the lower score is given, by the conservative principle. Third, an item that rests on an estimate starts one band lower. That discipline is what keeps the qualitative items comparable from one company to the next and from one quarter to the next.

C · Creative destruction — 15 points

Creative destruction asks whether the company is changing the world, and it is worth 15 points across three items of 5 each. C1, the old order being replaced, is scored on anchors: 5 where the thing being replaced is clear and large — NVIDIA replacing CPU datacentres with accelerated computing, Netflix replacing DVD and broadcast with streaming —; 3 where replacement is under way but limited in scope; 1 where the company coexists with or complements the old order; 0 where it replaces nothing. C2, irreversible behaviour change, is scored the same way: 5 where a change has happened that cannot be undone — the smartphone, search, and the dawn-delivery standard Coupang created —; 3 where the change is under way, as with AI coding assistants; 1 where the change is only a convenience improvement; 0 where behaviour has not changed. C3, change in the industry’s cost structure, asks whether the cost structure of the industry itself is being remade, judged on quantitative and qualitative evidence together.

D · BlitzFailing demerits — up to −30 points

The D section subtracts. It asks for the reasons a blitzscaler might collapse, across four items, and it can take up to 30 points off the total. D1, debt and interest, is worth up to −10 and covers the scale of borrowing, the interest burden, dilutive financing, and the cost of capital; a company with no debt scores zero here. D2, revenue quality, is worth up to −10 and examines the substance of revenue: related-party revenue, a material weakness in internal controls or an adverse audit opinion, dependence on government support, and distortion in how revenue is recognised. D3, governance, is worth up to −5 and covers dual-class shares and concentrated founder voting control — measured bands run from −2 at around a third of voting power, to −3 in the 51–59% range, to −4 in the 74–85% range — as well as control by a private-equity sponsor. D4, regulatory and social cost, is worth up to −5 and covers regulatory investigations, data breaches, litigation, and safety issues.

A total demerit of −15 or more triggers the FLAG — BlitzFailing watch — regardless of the company’s total score. A high-scoring company can therefore be flagged at the same time: the flag is a warning about fragility, not a second grade. Flagged companies are checked monthly, and the current flags are shown on each company page.

Grades and the success line

Grades: S ≥ 90 · A ≥ 80 · B 60–79 · C 40–59 · D < 40. The success line is 65, set by a 20-company blind backtest (the first draft’s 80 was lowered after measurement). A total demerit of −15 or more flags the company for BlitzFailing watch; current flags are shown on each company page.

The success line of 65 is not an arbitrary number. The first draft of the model set it at 80; the 20-company blind backtest measured the average score of the successes at 67.2, and the line was lowered to 65 on that evidence. The backtest scored 20 companies using only the information available at the time, with the answer key set by business outcomes three years later rather than by share price. It called 16 of the 20 correctly, detected all 10 of the failures, and caught 6 of the 10 successes. Its central finding shapes the whole model: speed did not discriminate, and discrimination happened only in moat.

Grades and the success line are separate things, and the gap between them is intentional. Grade B runs from 60 to 79, so a company can hold a B grade and still sit below the success line — the 60–64 band is exactly that: B grade, success line not reached. The grade describes where the total falls; the line describes whether the backtest says that total is consistent with success.

Score and confidence are shown as two separate axes, in the manner of a credit rating plus an outlook: a listed company is shown as a score with a confidence level, and an unlisted company as a range with a low confidence level. Low confidence never lowers the grade; it is reported alongside the score so that readers can see how much evidence stands behind it. The long-form structure of a published scorecard follows the same order as the model itself: gates, then the radar, then the two items that moved most, then the demerit check, then a two-sided assessment that states both the case for the company and the case against it.

Frozen at v1.0

The model is frozen at version 1.0. Weights, thresholds, and the qualitative anchors are not quietly changed.

Frozen means frozen for at least four quarters. Past scores are not silently corrected: the only exception is an arithmetic error, and even that is corrected only with approval and with a correction note left in the record. The discipline exists so that a score from one quarter can be compared with a score from another — if the ruler changed between quarters, the history would be meaningless. Every company page therefore shows the version the score was struck under, alongside its quarterly history.

If the model is ever revised, the revision is done in the open: every past score is recalculated under the new version, and both versions are published side by side. New indicator ideas are not folded into the live model in the meantime; they are recorded only on the v1.1 research track, where some forty candidates are waiting. Starting v1.1 requires a panel of more than 150 observations including failures, a condition that had not yet been met at the time the method note was written. Until then, v1.0 is the ruler, and the ruler does not move.

Evidence standards

Scores rest on published filings and reported data, with SEC filings first. Figures are labeled confirmed, estimated, or reported; figures that cannot be verified are left out. Scores are revisited each quarter, and sooner when earnings change the evidence.

Each quarterly re-score follows the same working sequence. First a data pack is built, with SEC EDGAR filings as the primary source and a scoring date, T, fixed in advance; information published after T is not used, so a score is always a judgment on the evidence available at that date. Then the Gate 2 revenue basis is declared and the gates are judged. Then each item is scored against its anchor, with the rationale written out, and uncertainty is labelled confirmed, estimated, reported, or unverifiable. Then an arithmetic self-check is run: the item sums must equal the section sums, and the section sums must equal the total. The score file feeds the quarterly history, and the site integration re-runs a full arithmetic audit centrally before anything is published.

Earnings seasons — February, May, August, and November — are the score-update seasons, and a score is revisited sooner whenever earnings change the evidence. Large movements are treated as the story of the quarter: an item moving by 8 points or more is covered in the main body of that quarter’s write-up, not buried in a footnote. Figures are labelled by their status throughout: audited revenue is kept distinct from run-rate and annualised figures, unlisted-company figures are labelled as reported, and anything estimated is marked as an estimate. Numbers are presented with both sides of the case: the tone of the project is narrative plus numbers, with no exaggeration, no certainty, and the evidence in public view.

What a score can and cannot tell you

A Blitz Index score measures the quality of the business as a blitzscaler. It does not predict what the share price will do, and the project’s own tests say so plainly. The first round of pre-registered hypotheses — that crossing the threshold, turning profitable, or a rising score slope would predict returns — was rejected on all three counts: the score does not predict future returns. Two later hypotheses passed provisionally and are reported as such, not as settled law. The first is the Speed–Moat Gap: companies whose speed far outruns their moat failed far more often than companies where the two were in balance. The second is Quality per Dollar: the score divided by the price-to-sales ratio carried signal, except at market bottoms, where it went silent.

The accumulated interpretation is a single sentence: the score measures the quality of the business, and the entry price governs the return. Failure, on this evidence, comes in two forms: speed without a moat, and a moat bought at too high a price. That is why the site publishes the score, the grade, the breakdown, and the reasoning — and never a view on whether to buy, hold, or sell at today’s price.

What this is not

Research only. Not investment advice. No buy, sell, or hold recommendations, no price targets, and no earnings or share-price forecasts. The full methodology is also published as a PDF: Blitz Index methodology (PDF). Every article is listed on the Research index.

The standing disclaimer is fixed: “For education and research purposes only. Not a recommendation to buy, sell, or hold any security.” The words buy, sell, and price target do not appear in the scoring work at all, and the index itself is named the Blitz Index precisely because it measures the degree of blitzscaling — not how attractive an investment is. A high score is a statement about a business; it is not a statement about a stock.