Bending Spoons

Everything on one page

This report does one thing: it breaks Bending Spoons' "acquire—transform—squeeze cash flow" model into numbers you can check, and labels the source grade of each number. The thirteen cells below are the entry point to every conclusion; each one notes the section it comes from.

Five failed founders, and a $10,000 acquisition

Founded in Copenhagen in 2013, named after the scene in The Matrix where a spoon is bent with the mind — "Impossible. Maybe." The next year the headquarters moved to Milan. The five founders' previous company, Evertale, failed and was liquidated, leaving $40,000; they used that money to seed Bending Spoons, and their first acquisition cost just $10,000.

Recruiting is its most aggressive marketing

Its screening of "people" is close to obsessive — its acceptance rate is two orders of magnitude below Harvard's. This is not just culture marketing, but the premise of the entire model: keep a small number of extremely high-leverage people, and hand the rest of the work to the platform and to AI.

2025 acceptance rate of about 0.036%. For comparison: Google runs at around 0.2%.

An IPO that deliberately sold very little

It filed its F-1 on 8 June 2026 and listed on the Nasdaq Global Select Market on 1 July, priced at $29 — above the top of its original range. It rose about 40% on day one, closing at $40.50. But it offered only about 9–10% of the whole company in shares, and used a dual-class structure to lock voting power in the founding team's hands.

Founder ownership (Class A shares)

Principal institutional shareholders

Ownership percentages are on an ordinary-share basis. In December 2021, Red Circle Investments, owned by Diesel founder Renzo Rosso, took a stake; it no longer appears in the above-5% disclosure table.

A closed loop that has run unchanged for 13 years

"Acquire → Transform & Optimize → Reinvest". The lineage it claims in the prospectus is interesting: Henry Singleton of Teledyne, Tom Murphy of Capital Cities, and in recent years it has studied Broadcom, Danaher and TransDigm — not one of them an internet company. What it studies is capital allocation, not product.

Two hard gates it wrote into its own prospectus

Capital deployment is accelerating

In just over two years, a single quarter's deployment is more than 10x the old full-year figure. That is both conviction and risk exposure.

Screening criteria (its publicly disclosed basis)

• Already has product-market fit, not "technology without users"
• Relies on organic traffic, not paid acquisition — ad spend as a share of revenue fell from 9% to 3%
• Subscription-based, mostly recurring revenue (subscriptions were 93% of revenue in 2025)
• Loyal users and long customer lifetimes; retention among older customers is clearly better than among new ones
• Low risk of being disrupted by AI — it states this one bluntly
• The product is embedded in the customer's workflow, so switching costs are high

By these criteria it has screened out more than 1,000 potential targets, with combined annual revenue of close to $400B; in 2025 it evaluated more than 200 in depth and closed only a single-digit number. This is a classic "better none than second-best" funnel.

Three pillars of the platform: the real moat

It uses AI more aggressively than anyone: the share of product requirement documents (PRs) written or assisted by AI rose from under 10% in 2025Q1 to more than 90% by the end of 2026Q1, of which about 70% are completed entirely by AI.

What this "in-house AI platform" actually is

One of the company's most important external narratives is that "we have a technology platform others cannot buy, so we can cut costs harder than anyone else". That claim needs to be checked in detail. Below is the complete list of in-house systems as itemised in the prospectus — nothing omitted, and nothing beyond what was disclosed.

Cash flow looks great, net income looks ugly

Revenue grew more than fivefold in three years, and operating cash flow is healthy — but 2025 net income was almost zero, because interest expense rose 337% year over year and swallowed the entire profit. This is the unavoidable price of the "borrow to buy growth" model.

The 161 of net income in 2023 includes a $103M one-time tax benefit, so do not be misled. On a TTM basis the operating margin jumps to 35%, mainly because of the structural change after large acquisitions such as Airtable were consolidated, and that should not be extrapolated as a trend.

Unpacking "adjusted profit": how the 376M is built

The company also reports "Adjusted Net Income". That number differs enormously from GAAP net income, and the prospectus sets out the full reconciliation. This is the entry point for judging earnings quality.

Unit economics: how much "leaner" it is than peers

The per-employee numbers look very good — but they come largely from clearing out staff after acquisitions, not from organic efficiency gains. That distinction matters when assessing sustainability.

More than 50 acquisitions, not one ever sold

From a small $10,000 deal at the start to paying $1.38B for Vimeo and $1.4B for AOL. In 2026 it took four more in a row: Eventbrite, Tractive, Airtable, Miro. Everything it buys is "the wreckage of the 2021 valuation bubble" — the same assets, now changing hands for a fraction of their old valuations.

The lesson from the two together: the valuation anchors of that 2021 cohort of SaaS unicorns have failed completely, and sellers are accepting "get your money back" prices. For a buyer this is a rare window — but it will not stay open forever.

Does the machine actually work? The prospectus answers it itself

Most acquisitive companies disclose only group-level organic growth and leave you to guess. Bending Spoons' prospectus unusually goes product by product through "before the acquisition vs after" — including headcount cuts, cost reductions and ARPU change. This is the most valuable and least cited part of the document. Everything below is on the prospectus's own basis.

User scale across the portfolio (disclosed in the prospectus)

Were the companies it bought making money or losing money at the time?

The prospectus discloses the standalone operating results of some acquired businesses "from the acquisition date to the end of that year":

These three correspond to its 2024 and 2025 acquisitions respectively. The ambiguity is that these figures include acquisition accounting (amortization of intangibles, transaction costs and so on), so they cannot be taken directly to mean "it was losing money when we bought it". But they show one thing: Bending Spoons' prey is not "a cheap good business" but a company with healthy revenue whose profit is eaten by its cost structure — what it sells is precisely "replacing the cost structure".

The same construction, on four assets

When explaining revenue changes at each business, the prospectus repeatedly uses one very specific construction: "subscription revenue rose, driven by higher average revenue per subscriber, but subscriber numbers fell". This is not a one-off accident but a pricing strategy executed over and over.

The counter-example in the other direction: Meetup, WeTransfer and komoot saw ARPU and subscriber numbers rise together over the same period. So this is not "every asset is squeezed by price increases" — the more accurate description is: some assets can still improve on both (genuinely improving), while others have entered the "trade price for scale" phase. Judging which side a company is on matters far more than the group organic growth number.

Quantifying "price increases for churn": net revenue retention

The prospectus itself discloses net revenue retention (NRR), the most direct way to turn the construction above into a number. On a group basis, 2023–2026Q1 was 93%, 91%, 95%, 94% — which looks very steady. But broken down by business the differences are large, and the direction lines up exactly with "who is still improving".

Average for 2023Q1–2026Q1. Evernote 99% means almost no leakage from existing customers' payments — the price increase was nearly painless for the installed base, because those who left were never paying users; Remini 87% shows that this consumer user base, acquired through paid marketing, churns far faster than tool-type products. Run through the same system, the outcomes are completely different.

We built a control group, and it contradicted our own hypothesis

An earlier version of this report implied that app-store ratings generally get worse after a Bending Spoons acquisition. That claim has an obvious alternative explanation - the whole iOS review ecosystem has itself been getting worse for years, and plenty of apps that were never acquired also show a "lifetime average above recent average". The only way to rule it out is to run a control.

Method (identical for both groups). For each app, we pull the most recent 500 reviews from Apple's public reviews endpoint (with timestamps and star ratings), count only reviews inside a fixed observation window (2026-03-16 to 2026-09-12, 180 days) and compute the window average rating; we then subtract the lifetime average rating shown on the store page to get Δ. Both groups use the same window, the same endpoint and the same single scrape (2026-09-11/12).

Δ = lifetime average rating − window average over the last 180 days (higher means a larger decline)

Control group detail (20, none of them acquired)

Acquired group detail

What is left once the control group has run

First, Meetup really is an outlier. A Δ of 3.20 is the highest of all 32 samples, with a 180-day window average of 1.52 and 69.8% of reviews at one star. This is not "the whole market falling" - it is the only extreme value in this dataset. Given that Meetup was acquired in January 2024, well over a year ago, this is a thread worth watching on its own.

Second, the store's lifetime average rating is not credible for any app. In the control group AllTrails sits at 4.89, InShot at 4.85, Luma at 4.93; these numbers look like "users are very satisfied", but they equally mask a recent decline of 0.2-0.4 points. Taking a 4.8 rating as proof of a healthy product is a systematic misreading.

Third, the one observation that genuinely holds up has nothing to do with acquisitions.​ Vimeo's rating collapse happened in mid-2023, and komoot has been shedding points since 2022 - both of them earlier than the acquisitions (before November 2025 and before March 2025 respectively). The only directional judgement this data can support is this: when it picks a target, the reputation is already deteriorating but subscription revenue has not yet broken. The bad reputation is already in the price while the cash flow is still intact - which is precisely the necessary outcome of its screening logic.

Its most important number has the word "estimate" in the definition

For a company that grew up on acquisitions, organic growth is the only metric that answers whether the machine is really creating value. Which is exactly why the way this metric is defined deserves to be read word for word.

Three key phrases. "The same businesses" excludes the contribution of anything newly acquired in the period - reasonable, and standard industry practice. "The same period of the prior year" anchors the comparison to the same season - also reasonable. What actually needs watching is "estimated pre-acquisition revenue" - when a company buys a business mid-year, calculating that year's organic growth requires the business's revenue in the same period of the prior year, and revenue Bending Spoons did not own and which was never audited can only be estimated.

This is not fraud, and standard practice does permit it. But it means: the numerator of this company's most important growth metric is audited, the denominator is estimated. And the lower the denominator is estimated, the higher the reported organic growth rate. For a business that acquires every quarter and whose targets are often private companies (with historical data supplied by the seller), this is a definition that has to be watched continuously.

The actual numbers: a clearly declining curve

The 13% in 2025 is the peak - but in that year it also completed large acquisitions including Vimeo, AOL, Brightcove, Harvest, komoot and MileIQ, so the "estimated" component of the denominator was at its heaviest. The next two quarters fell to 6% and then 3%. Over the same period total revenue grew +126% year over year (2026Q2). That is the whole story: revenue is exploding, but the explosion comes from buying, not from growing.

$3.3B buys three companies, pro forma full-year net income of $22.37M

Between November 2025 and March 2026, Bending Spoons bought Vimeo, AOL and Eventbrite one after another. All three were treated by the SEC as significant acquisitions, so the prospectus attaches audited financials for each - the densest and least-quoted passage in the whole document. Read the three tables side by side and they answer one question directly: what, exactly, did all that money buy?

Annual revenue basis: full-year 2025 for AOL and Eventbrite (the prospectus's as adjusted basis); full-year 2024 for Vimeo, because it had already turned to a loss by 2025. Consideration is the purchase consideration disclosed in the prospectus.

AOL: this one was a bargain, not a hot potato

The common read on AOL is that it is an ageing advertising asset. The primary numbers do not support that. AOL's 2025 revenue was $633.4M, split as follows:

Subscriptions are 59% and advertising 40% - this is not a pure advertising asset. And the trend runs against intuition: subscription revenue slipped from $375M to $372M (−1.1%), while display advertising climbed from $213M to $240M (+12.7%). AOL's operating margin is 52.7%, with net income of $183M. Bending Spoons paid $1.454B for a cash cow earning $334M of operating profit a year, equal to 4.4x operating profit.

Vimeo: on earnings this one is expensive, and it was already loss-making pre-acquisition

Vimeo's revenue barely moved across 2023 and 2024 ($417.2M → $417.0M), and 2024 operating profit was only $18.68M - 72.7x operating profit against a $1.359B consideration. More important, the 11 months before the acquisition (2025-01-01 to 2025-11-23) reversed direction:

So Vimeo did not get worse after being bought - it was already losing money when it was bought. That explains Bending Spoons' motive: what it buys is not a good business but one where "the revenue is still there and the cost structure has rotted" - the type of asset with the most room for rework. The near-total reorganisation of Vimeo after the deal is that logic carried forward.

Eventbrite: paying 1.7x revenue for a loss-making company

Eventbrite's 2025 net revenue was $291.8M, with an operating loss of $26.04M and a net loss of $10.52M. The $505M consideration equals 1.7x revenue - the cheapest multiple of the three, but also the least profitable of the three. In the first 10 weeks of 2026 (to the 9 March closing) it lost $34.03M net, on seasonality plus restructuring costs.

The three deals together: real profitability on a pro forma basis

The prospectus provides pro forma statements with all three folded in. This is the most honest presentation of full-year net income:

And on a pro forma basis, 2026Q1 has already turned to a loss

This is a comparison in the prospectus that is easy to skip. BSP reported 2026Q1 net income of +$27.47M; but fold in the 10 weeks of Eventbrite before the closing and the pro forma figure becomes a net loss of $13.07M (−$0.02 per share). The reason sits in the table: Eventbrite lost $34.03M net over those 10 weeks, plus acquisition- and financing-related adjustments.

The other side of integration: 1,830 people, only a few hundred left

The prospectus contains an unusually candid passage. Across 2025Q4 and 2026Q1 the company added 1,830 full-time employees through the acquisitions of AOL, Eventbrite and Vimeo. Then, immediately:

1,830 people in, a few hundred left: the implied cut is over 80%. This is the company's own disclosure, not a media estimate. It is the same move as the earlier Evernote (341→60) and StreamYard (154→44) cases, only five times the scale.

The same company, with leverage of 2.19x or 5.16x

This is the section of this report most worth reading more than once. The low leverage ratio — the number long used to argue "this company is solid" — is low not because debt is small, but because its definition of "profit" is very loose.

What the covenant actually says (prospectus wording)

The prospectus defines these two terms in full as follows (direct translation):

Two leverage ratios at the same point in time

The denominators differ by about 2.3x, and leverage moves from 2.2 to 5.2. The threshold is 4.00x, so one can say precisely: on the company's covenant basis it has nearly a full turn of headroom; on the accounting basis it has already crossed the line. Neither number is false, but only one of them is quoted regularly.

Run it yourself: drag the parameters and watch leverage move

This is this report's own scenario model, not company disclosure. Base values are taken from primary sources: net debt $4.13B, covenant-basis EBITDA $1.85B, GAAP-basis EBITDA $800M, pro forma interest expense $368M; it assumes 70% of debt is floating-rate and that newly borrowed debt costs 8.5%. New acquisitions are estimated conservatively as contributing no EBITDA (that is, loss-making targets like Miro).

Three preset scenarios

Also note: the prospectus discloses a second credit facility / note document with a leverage cap of 3.25x, likewise based on the adjusted EBITDA basis. The incremental term loan borrowed in 2026Q1 has a principal of $1.61B + €300M (the latter signed 2026-01-30), used to pay the consideration for AOL, Vimeo and Eventbrite. The definition details of the two documents require reading the credit agreements attached to the registration statement (this report did not compare the exhibits clause by clause).

The release is not a single day — it is the 48 hours after a quarterly report

The free float is only 9–10%, so "when does the lock-up release" is the most certain event for this stock over the next three months. This clause is written very specifically in the prospectus, and it is far earlier than the common "180 days after listing".

Unpacked, the clause imposes three actual constraints:

1. The baseline is 180 days — counted from 2026-06-30, no later than 2026-12-27.
2. But it can come early — the trigger is the "second quarterly report", with at least 125 days elapsed since the prospectus date (that is, 2026-11-02).
3. The Q2 report was already released on 2026-08-13, and that was the first. So the "second" is the Q3 report, which on the company's cadence (Q2 in mid-August) falls in early November 2026.

An easily overlooked exception: the company can still issue stock for acquisitions

One exception listed in the prospectus deserves to be flagged on its own: the lock-up does not restrict the company from issuing shares of up to 10% of the common stock outstanding after the offering in a merger, joint venture or strategic transaction (provided the recipients also sign a lock-up agreement). In other words, the lock-up locks up shareholders, not the company's ability to acquire. This has already been used in the Miro deal — some Miro shareholders rolled $295M into new BSP shares (see §15).

13 Form 3 filings, 0 Form 4 filings

The prospectus shows that the founders and executives still hold the overwhelming majority of the voting power after the offering. So the question is direct: is anyone selling? The filing record on EDGAR gives the only answer available today.

Filing record (as of 2026-09-11)

How to read that zero. The lock-up has not yet expired, so "nobody is selling" is the result of a structural constraint, not evidence of confidence — reading it as a positive is over-interpretation. But it does show two useful things: first, the filing channel is working; second, nobody has obtained an early waiver. That point has practical significance: the lock-up gives waiver rights to three lead underwriters (Goldman Sachs International, J.P. Morgan Securities LLC, Allen & Company LLC), who could in theory release shares. If any Form 4 appears before November, that is a signal that a waiver was used or that the release came early.

Fewer disclosure obligations: this is a "foreign private issuer"

The company itself lists this as a risk factor in the prospectus, in direct wording:

Since the IPO the company has filed only 5 6-Ks (07-28, 08-04, 08-13 Q2 results, 09-04, 09-10 Miro acquisition). No 10-Q, no 8-K. This means there is no event-driven immediate disclosure obligation — to know what the company is doing you must read each 6-K one by one, and you cannot rely on any automatic alert. For a research analyst, this is an information cost to build in in advance.

Calendar: three things in the next three months

From $17.5B to $1.36B: an acquisition that shrank by 92%

On 10 September 2026, Bending Spoons announced the acquisition of Miro. It is its most striking deal since listing, because the price is itself a market verdict.

Bought cheap, and here is the cost

Its integration playbook is brutally effective, but it is not gentle. Every item below rests on a public record, not on speculation.

Does what he says square with what the books can verify

The previous sixteen sections rest entirely on SEC filings and primary scraping. This section uses a different source: the company's number one, speaking in his own words. The reason for including it is not that the story is good, but this: where a model that has been repeatedly validated can be matched, point by point, in the books by its own telling, its credibility rises another notch; and the one point that fails to match is precisely the one most worth remembering.

The source and its reliability. The raw material is an interview with co-founder and CEO Luca Ferrari on David Senra's podcast, written up in Chinese by Shensi Circle, a Chinese tech publication on WeChat, on 2026-09-14 (this report did not listen to the podcast audio itself). It is not a primary document but a retelling of an oral account, so this section treats it as secondary: anything that can be checked against the prospectus or the books is checked item by item; anything that cannot is explicitly marked single-source.

Start with the comparison table

20 people, $100M

The headline number to look at first is Evernote. The CEO's version: bought in 2023 for about $200M, with annual revenue of about $90M at the time, about 350 employees and roughly break-even; within eighteen months of taking over, the team was cut to 50–60 people, the codebase rewritten, the cloud infrastructure rebuilt and monetisation redone. It is now run by 20 people, with revenue above $100M and very high margins.

The prospectus side (primary) matches: Evernote 341 → 60 (−82%), fewer than 20 at the application layer. The two sources are independent and reach the same conclusion.

It is easy to tell this as an "efficiency miracle", but the more accurate reading is one of weighting: group revenue per employee rose from $1.12M to $2.57M (§03), which reflects not only higher efficiency across the board but also a few assets pushed to the limit pulling the average up. If Evernote's revenue really is above $100M, it accounts for about 5% of the group's $2.04B of revenue, while its headcount share is under 1%. The same "revenue per employee" metric differs between assets by two orders of magnitude — which is also why the layout in §12, with "per Spooner 1.12 / 1.64 / 2.57 placed alongside 0.97", deserves caution.

He admitted this strategy outright

This upgrades the finding in §08 from a "suspicious pattern" to "a position confirmed in his own words". That section found that subscription revenue at Evernote, Issuu and StreamYard was rising while subscriber numbers were falling, with almost identical wording across the three assets. At the time the most that could be said was that it did not look like a coincidence. Now it can be written as a conclusion: a falling user count is not a side effect of the rebuild, but one of the rebuild's objectives.

The second layer is more worth thinking about: this strategy can hold only because the parent company is not bound by a single metric. A public-market company cannot do it — one quarter of falling user numbers and the valuation falls first. So the real dividend of the "permanent capital" label is not "we can choose not to sell", but the ability to endure a stretch of ugly data in exchange for longer cash flow. The trade-off is equally clear: this road ends by trading a user base that can only be felt for revenue that can be quantified.

"We are not a PE firm" — and that claim can be checked

The CEO is visibly impatient with the label. His distinction: PE firms buy in order to cut costs and sell after 5–7 years, so they cannot integrate deeply (integrate too deeply and it cannot be separated out for sale); they buy to hold forever, and deep integration is itself the source of value.

The part that can be checked: across the 50-plus acquisitions catalogued in this report, not one disposal of material scale exists. That claim holds up. The part that cannot: "the value created by integration exceeds the price paid" is a claim about capability, and only results can prove it.

People: more than 100 signals, and a centralised hiring team

What came before was the accounts; this is the people — because this company's product is, in the end, "running the same assets with fewer people".

• Anti-experience. A preference for hiring new graduates rather than seasoned executives. The reason given is that experience has a short shelf life ("the way software was written in 2010 barely overlaps with 2026"), and that "you can get worse through experience" — working for a long time in a low-standard environment makes low standards look normal.
• More than 100 signals. For candidates with no work history the sample of outcomes is too small, so they turn to behavioural signals: academic records, personal projects, test scores, and even how a candidate treats support staff who have no influence on the hiring outcome. People who are courteous to the "interviewer" but arrogant towards a "small role", they say, typically also reveal collaboration problems once in the job.
• Hiring and exit decisions are centralised. Not left to the business line's hiring manager, but handled uniformly by a dedicated Talent Team. The reason is to hedge against distorted incentives: a business line short of people will tend to hire "someone who can start now" rather than the better long-term hire.
• All job titles abolished. From senior engineer to VP, none are used, and the company does not care what employees write on LinkedIn. According to the CEO, since then nobody has spent a minute on "why have I not been promoted".

These points mesh closely with the company comparison in section 19. Constellation's model is decentralised — buy and do not intervene, with each business autonomous; Bending Spoons is centralised — hiring, engineering, monetisation and compensation all pulled into headquarters. This section gives the reasons for centralisation (uniform standards, a large sample, the ability to move people across roles) and, in passing, its cost: one centralised standard governing 50-plus assets at once leaves very little margin for error.

"Ruthless simplification": a rule that works only against itself

They have an explicit rule: the burden of proof for any proposal that adds complexity lies with the proposer. To add a feature, a process or a rule, you have to prove it is worth it; the person objecting has nothing to prove — saying "I do not think we should add it" constitutes resistance.

This rule explains a question left unexplained earlier: why 50-plus acquisitions packed into one platform have not pulled it apart. Organisational complexity does not grow linearly — going from 3 parts to 4 adds not only one more part but its connections to all the other parts. Most companies' complexity is the accumulation of "every decision looked reasonable on its own". One premise in the CEO's formulation is worth noting: he believes the organisation is able to tolerate a state in which "there is always more that could be done than can be done" (the accompanying practice is deliberately giving high-potential employees workloads beyond their comfort zone). This is structurally the same as the hardest number in this report: $2.57M of revenue per employee — if that holds, someone has to be permanently overloaded.

"I am a sceptic about numbers" — and that sentence must be measured with the same ruler

The CEO describes himself as sceptical about numbers: numbers are only an approximation of reality, and when information is insufficient one should return to first principles. His example is concrete — when discussing pay rises, even where A/B testing showed no significant effect he still supported the rise, because the experiment only sees the people who clicked on a recruitment ad in the short run, not the long-term compounding of pay on reputation and on word of mouth among students.

That sentence has two sides, and both belong in here. The good side is long-termism, and it explains why the company dares to play "lose user count first, collect cash flow later". The other side is that the same company, in its external disclosure, leans heavily on "adjusted", "credit-agreement basis" and "pro forma" — the metric in §12 that folds "anticipated cost savings" into EBITDA and thereby pushes the leverage ratio down to 2.19x was produced by a company that describes itself as sceptical about numbers. This is not an accusation but the same ruler: if the CEO holds himself to scepticism about numbers, that scepticism should apply to its own external measures too.

What is genuinely scarce here is not "the ability to buy cheap"

People who can hunt down cheap assets are everywhere. What sets Bending Spoons apart is two other things.

Bending Spoons vs Constellation Software

Put Bending Spoons next to Constellation Software (founded in 1995 by former VC Mark Leonard, an acquisition-driven compounding machine returning about 30% annualized since listing), and a deeply counterintuitive conclusion emerges: they are two opposite implementations of the same philosophy.

Constellation's financials are on a USD basis and its market data on a CAD basis (C$1 ≈ $1.42). Both companies were substantially repriced by the market over the past 52 weeks: CSU −39.8%, BSP down about 33% from its high. The market's premium for "acquisition-driven compounding machines" is contracting.

The essential difference between the two models

In one sentence: Constellation's bet is "I select well, and I let capable people run it themselves"; Bending Spoons' bet is "I can rebuild it, and I can do it 50 times". The latter's return curve is steeper, but its tolerance for execution error is close to zero — and it currently carries 5.2x EBITDA leverage.

Distilled from the real criteria disclosed in the Bending Spoons prospectus, plus the comparison with Constellation. Every item needs a number; "feels good" is not accepted. Checked state is saved automatically in the browser and is still there next time you open it.

Checked state is stored in the local browser (localStorage) and is not uploaded. Use it to screen deals item by item on site during due diligence.

Where every number comes from, and which conclusions we changed

The evidence in this report has three tiers. Anything from a primary SEC filing is marked Primary; anything from an authoritative third-party report is marked Secondary; anything this report scraped or calculated itself is marked Original. Any judgment without a source label is treated as this report's analytical opinion, not as fact.

Primary sources (SEC EDGAR, CIK 0002004711)

Secondary sources

Forbes (pro forma profit of $22M, later confirmed as $22.37M by the primary pro forma table), The Economist, WSJ, TechCrunch (the Miro deal), Wikipedia (company history, used only for cross-checking and never as a sole source); Luca Ferrari's interview on the David Senra podcast (Chinese-language compiled version, Shensi Quan, 2026-09-14) — used only in §17, checked item by item against primary sources, with anything that could not be verified flagged as single-source. Every secondary conclusion is compared against primary data in the body; where they disagree, the primary source governs.

Original data: the app store controlled experiment (§09)

The reproduction script is kept in the working directory tools/store_audit.py, and the raw reviews are stored review by review in /tmp/bs/rev/ (acquired group) and /tmp/bs/rev_ctrl/ (control group). The script can be re-run using the same window logic, so the result can be independently reproduced.

What changed in this version (corrections log)

Known limitations and open items

• The company itself states that MAU and paying customers are summed across products and cannot be de-duplicated, so any "group total users" figure cannot be compared directly with a single product's number.
• Pro forma statements and the TTM basis are inconsistent. The pro forma figures assume the three acquisitions closed at the start of the period, while TTM is the actual consolidated result; the two cannot be subtracted directly.
• The review data is not a random sample. Dissatisfied users are more inclined to write reviews, so we only make relative comparisons between groups, never absolute readings.
• The control group is matched by category, not by size or age, so differences in review volume bring differences in the time span covered within the window.
• Google Patents queries were blocked, the patent route could not be completed, and the technology section relies entirely on the prospectus's own list.
• The filers of the two SC 13G forms could not be extracted (on EDGAR that form is JS-rendered plus XML and index scraping failed).
• The App Store data does not cover Google Play, and this report makes no statement about the Play side.
• The CEO interview in §17 is a retelling of an oral account, not a primary document, and not audio this report listened to directly. Within it, Evernote's "20 people, revenue over $100M" and "more than 50 in-house tools" are single-source and unconfirmed by SEC filings.
• The 2026Q3 results have not yet been released, so the exact lock-up effective date, Q3 organic growth and insider behaviour after the lock-up expires are all items still to be verified.