Fluidkey: Daxos Teardown

Two businesses stacked on one wallet.

The substrate is a stealth-address wallet. A user gets a username.fkey.eth name; every resolution returns a fresh address not linkable to the previous ones, each backed by a 1/1 Safe smart account derived from keys the user controls. The founder demonstrated it live at [15:13] to [16:09]. VERIFIED Six queries to the live ENS CCIP-read gateway for moritz.fkey.eth returned six distinct addresses, three of which returned zero code and zero balance on Base via eth_getCode and eth_getBalance, matching the documented counterfactual behaviour.

The business is a dollar account for Taiwanese mass-affluent retail: account in about 30 seconds, KYC through Sumsub and Noah in about five minutes, US ACH and European bank details in the user's own name, automatic sweep into Morpho and Aave vaults, and Ondo tokenized US equities. The hook, at [19:50]: "they get this ACH account within 5 minutes of signing up." The founder is explicit that the substrate is not why they came, at [16:48]: "it's not, I guess, like the main problem we are solving for this new user set."

Three qualifications that change the shape of the product

• There is no Taiwan rail. The documented rails are ACH, SEPA, Fedwire and SWIFT only (docs.fluidkey.com/readme/bank). Taiwan has none of them and there is no TWD leg. Taiwanese guides describe the real on-ramp as buying USDC at a local licensed exchange (BitoPro, HOYA BIT, MAX) and withdrawing to Fluidkey. If so, Fluidkey is the leg between a Taiwanese VASP and a US broker, not "one app replacing the bank account, remittance provider, and offshore broker" as the memo says. The ACH account gets money out to IBKR. It does not get money in from Taiwan.
• Privacy does not extend to Fluidkey. This is not an ERC-5564 implementation. The docs use ERC-5564 only as a BIP-32 path constant (m/5564'/0'/8'/0'/0'/p'/n'); there is no Announcer contract and no reference to ERC-6538 anywhere. The user shares a derived viewing-key node and Fluidkey's server generates the addresses. The docs say it: "only you and Fluidkey can see all transactions and assets." Fluidkey can see every user's full balance and transaction history, forever. Disclosed, flagged by Dedaub, and weaker than "stealth" implies.
• Every component is rented. Banking and fiat: Noah, fronting SSB Bank and Lead Bank. Yield: Morpho and Aave, Gauntlet curating. Tokenized stocks: Ondo Global Markets (BVI) Limited via Bungee (the founder's "Socket" at [40:20], same firm after a rebrand). Web key custody: Privy. Identity: Sumsub. Fluidkey's own irreplaceable infrastructure is one AWS Lambda function URL in eu-west-1 serving all ENS resolution, and two relayer EOAs.

Pivot, expansion or rescue? A genuine pivot, discovered rather than chosen, executed late, unproven beyond one geography

Pivot, on the evidence: the registered corporate purpose is digital privacy, the homepage sold privacy for two years, the whole 2025 blog addressed DeFi natives, and the founder says the new users are "not deeply crypto native" [18:17] and that privacy is not the main problem being solved [16:48]. Not a rescue: no distress, team intact, change followed inbound demand, and his account is corroborated by independent Taiwanese coverage. "Taiwan kind of came to us before we started focusing on it more" [33:40].

Two things against it. The pivot is five months old in public: the site changed in April 2026, the local lead joined in June, tokenized stocks launched in late June, the bank provider changed in July. Daxos is being asked to underwrite $12.5M on a product surface one quarter old. And the acquisition story does not match the company's own documentation: the memo says "Our customer acquisition today is organic", while the docs, all dated before the memo's last edit, publish a referral programme paying 5 USDC per valid new user in Chinese-speaking regions conditional on completed KYC and a funded deposit, a live creator code at i.fluidkey.com/GRENADE published 2026-08-08, and two $30,000-threshold deposit campaigns. That is acquisition and deposit-incentive spend. It may be small and sensible. It is not organic.

And the leading Taiwan creator is a shareholder. The memo names "Astro" of Blocktrend as both a "Fluidkey angel" and one of the most respected personal finance voices in the country. Blocktrend published the Moritz interview on 2025-09-30 with no investment or affiliate disclosure in that piece, and recommends Fluidkey through referral links. A legitimate and common creator-led arrangement. It is not independent third-party validation, and it is a concentration risk as well as a strength.

The Pulse exit claim OVERSTATED

On the call: [05:23] "I built another company called Pulse", [05:29] "Spent eight years living in Nigeria, Ghana and Kenya", [05:57] "we sold it for $50 million. At the end of 2022." The record: Pulse Africa was founded in 2012 by Ringier AG, a Swiss media group, and was a corporate subsidiary from the start. Harmonic has him joining June 2016 as a Business Intelligence Consultant, rising through MD Ghana, MD Nigeria and MD Africa, and Co-CEO for five months (2022-08 to 2023-01). That is 5.6 years, not eight. PitchBook records Pulse (Africa) acquired by Ringier on 2023-01-01, the parent consolidating ownership. No public price was found (checks: press searches, Ringier corporate disclosures, trade press). The memo says "a mid-8-digit acquisition"; the call says $50M. He was a genuine senior operator across six African markets ending as Co-CEO and he was there through a transaction. He did not found it, it was not a startup, and whatever he personally realised would depend on executive equity, not founder equity. A framing issue, not evidence of dishonesty, and worth raising directly and without heat.

The claims table

What he never said, and nobody asked

Total user count. Active users. KYC'd users. Retention or churn in any form. CAC, marketing spend or channel. Headcount. Burn. Runway. Cash in bank. The Non-KYC line, which is 30.1 percent of his own book. Deposit concentration. What licences Fluidkey holds anywhere. Whether $12.5M is pre-money, post-money or a cap. SAFE discount, MFN or pro rata. Use of proceeds. The cap table. Antonio's background. The words "non-KYC" are never spoken by anyone in 47 minutes.

What Daxos asked and did not get

• Marketing spend and partnerships in Taiwan. Not answered at all. [33:34] Joseph Curcio: "why did you target Taiwan first? And, and are you doing anything there for marketing specifically or have any partnerships?" He answered the first from [33:40] to [34:36] and never returned to the second. No CAC, no spend, no channel, no named partner. Joseph did not re-ask. The entire investment case is Taiwan growth, so whether that growth is bought or organic is the central question of the call, and it went unanswered. Two things Daxos did not know at the time make this worse: the 5 USDC per funded user referral programme, and the $30,000 campaigns running through the exact months being shown.
• Why users would continue. Not answered. [17:32] Mark Davidoff: "why would they continue like crypto native people in Taiwan?" He answered the acquisition half and never addressed continuation. Retention was never discussed on this call in any form. His own workbook has transacting user retention falling from 0.568 to 0.39 by month five. He had that data and did not surface it.
• Ondo dependency. Answered on the wrong axis. [39:55] Joseph asked what happens if Ondo geo-restricts. He answered that he can migrate issuers. Every alternative has the same or narrower perimeter.
• Custody of the tokenized stocks. Partially answered. [12:54] Mark asked where they are custodied. He described the minting mechanism and the liquidity benefit, never who holds the underlying shares, in which jurisdiction, under what investor protection, or what happens if Ondo fails.
• Regulatory posture. Deflected into a denial. [29:35] Mark handed him an open invitation to describe his position. He answered about what users are not doing. Mark replied [30:03] "Got it, got it. Okay. I appreciate the color there" and moved on.
• Competitors. Thin. [21:51] Jake S asked, naming ether.fi himself. Moritz discussed only ether.fi and named no competitor unprompted. He did not mention KAST, RedotPay, Zar, Sling, Infinex or any Taiwanese local player. Combined Harmonic-recorded funding of the five crypto-native names he omitted is $384.2M. He is raising $2M.
• What actually happened in Taiwan. A story, no evidence. [36:48] Nicholas Davidoff asked what actually caused the growth. [37:22] "different blogs in Taiwan talking about it. I got invited on the podcast as well there." No blog, no podcast, no date, no traffic figure, no referral rate. Substantial independent Taiwanese coverage does exist (ABMedia, CryptoCity, Grenade, penchan, Blocktrend, vocus), which is corroborative. He was not asked to name them.
• The valuation, on first asking. [09:40] Carlo asked for revenue and valuation together; only revenue was answered. He had to come back at [25:31] sixteen minutes later. He flagged he was in a car listening at [25:28], so probably inattention rather than evasion.

Candour: better than average, with one material overstatement on the number that carries the case

He volunteered six weaknesses unprompted: that most of the app is not monetised [09:47], that his own core technology is not why the new users came [16:48], that privacy does not extend to KYC data [21:21], that Ondo is far behind IBKR on breadth [14:02], that Taiwan was luck rather than strategy [34:36], and that no lead has committed [42:53]. Flat no on Robinhood. Zero closing pressure: no scarcity language, no deadline, no competing term sheet anywhere in 47 minutes, and Mark set the pace himself at [45:20]. He demonstrated the product live rather than asserting it, and he handed over the metrics workbook. All of that is real and it is not common.

Against it, a pattern. Nine second-order figures were checked and every one runs the same way.

Three run the other way and should be credited. The memo says Taiwan is "up 10x since the beginning of the year" when the workbook supports 14.6x. It says "75%+ of Taiwanese users who sign up end up transacting" when the figure is 77.1 percent. And the headline platform number on the call matched the workbook exactly.

So the big checkable headline numbers are honest and they reconcile. The distortion is in the second-order figures, in the choice of denominator, and in the verbal restatements, which are consistently more aggressive than the written ones. The read is founder enthusiasm and loose spreadsheet hygiene, not intent to mislead. The operational conclusion is that Daxos should diligence off the documents, not off the meeting.

Nothing was committed on either side. Mark at [45:01] "I'm definitely, like, very interested", and the gating event he set at [45:10] was "once you find a lead, you know, we can. We can circle back with. With commitments."

What is good. Two people wrote the cryptography themselves and it is real, open source and audited. The CEO is still shipping commits to the core kit in 2026, which is a genuine signal. Antonio, Federico and the ODF Data / Ianum / tProof / Above lineage are one engineering team that has moved together since 2016, five ventures of continuity. The founder has genuine operating seniority from eight years in African digital media across six markets ending as Co-CEO, and he has physically relocated to the market that matters.

What is thin. Five people, of whom three are engineers, one is an advisor and one is the entire Taiwan organisation. In two years and nine months the company has added exactly one operating employee. Jennifer Hsu graduated from NCCU in 2024, spent 22 months as an Associate Consultant at Mastercard Data and Services, and now runs the region that is 65.8 percent of the identified deposit book at about three and a half months of tenure, while retaining a current role as co-founder and hackathon lead of XueDAO, a Taiwan blockchain community that is a plausible distribution channel the founder never mentioned. No compliance hire, no finance hire, no commercial leader, no second Taiwan employee. The engineering bench is two people who have never worked apart plus a CEO who commits code.

Two cap table questions. Federico Castelli is a registered board member of Fluid Privacy SA but is named as a founder in neither the memo nor Harmonic, and he is the company's primary contact email in Harmonic. What is his equity and vesting? And Moritz ran "110 Ventures" from 2023-03 to 2024-02, overlapping the founding and stopping two months after incorporation. Ask whether it holds Fluidkey shares.

CLEAN Adverse record: none found, on real checks with control tests

The founders are fully named and traceable in the Swiss register with no nominee layer, which is itself a positive.

One failed check, reported so nobody treats it as clean. An initial OpenSanctions screen returned zero matches for all four names, but control testing showed the endpoint was returning {"detail":"No API key provided."}. Those zeros were an authentication artifact and were discarded; the four primary lists above were run instead. Two checks could not be completed: the Italian Registro Imprese for ODF Data, Ianum, tProof, Daolize, Above Tech Inc. and ABOVE S.r.l. requires a paid account, so the prior companies of Antonio and Federico are unverified as to insolvency or litigation (three of four founding board members are Italian, so this is a real gap and the extracts are inexpensive); and the Swiss Federal Supreme Court judgment search refused connections from this host.

The series is a stock, not a flow. The sheet never says. Four things settle it: the founder calls it TVL at [23:59], he quotes weekly transfer flow separately at [09:29] as a different number, the asset mix table only makes sense on a balance, and the total falls in May, which a cumulative total cannot do.

Deposit growth, 2026

Fluidkey is Taiwan. Remove it and you have a $6.2M book compounding at 4.27 percent a month, which went backwards in March, and whose largest component grew 1.83 percent a month. The book is denominated in dollars but holds crypto, the Taiwan sheet shows 4.0 percent in "other crypto", and no part of the workbook separates net new deposits from mark-to-market on assets already held. A flat-to-slightly-up Non-KYC line over eight months is equally consistent with a static set of wallets whose token prices drifted. Daxos cannot tell from this pack.

The counter-argument is fair and should be stated: this is what a wedge is supposed to look like, and the founder is explicit about leaning in [34:03]. The risk is not that they focused. It is that the pre-Taiwan business, which is what existed when the roughly $5M pre-seed price was set, has demonstrably stopped growing, so there is no second leg.

FINDING The geography sheet is not computed over total deposits

The sheet is headed "% of Total Deposits". Taiwan is 45.5 percent of the August total. It is 65.2 percent of the August total less the Non-KYC line, and the six rows sum to exactly 1.000 with no residual bucket. No other denominator comes close.

Unflattering in two directions at once. It overstates Taiwan's share of the actual deposit base by twenty percentage points, and it means the company has no country attribution for 30.1 percent of its book and the sheet handles that by excluding it rather than showing it as unknown. A reader taking the sheet at face value concludes Taiwan is two thirds of the business and that the geography of the book is known. Neither is true.

The retention paradox

There is no paradox. The two multiply. For any cohort, NDR equals users retained times growth in balance per surviving user, and the identity holds exactly for every cohort in the file.

The honest reading. Every dollar of NDR above 1.0 is balance growth inside a shrinking group. None of it is user growth and none of it is the product pulling churned users back. The 4.69 headline decomposes as: 61 percent of the January cohort left, and the 24 who stayed grew their balances 11.9x. The most impressive number in the pack is computed over twenty-four people, and the absolute gain behind it is $80,767, from $21,859 to $102,626. That is one or two people adding money.

There is a real positive underneath it, and it is better evidence than the headline it is buried under. The January cohort arrived at $358 a head and reached $4,276. People arrive small and grow. That is a wealth product working on the people it works on. The February cohort, by contrast, started at $3,857 a head, 10.8x January's day-zero balance, and looks recruited rather than manufactured. February is 36.3 percent of the June Taiwan book from a single month's intake. Only one of those two motions scales with marketing spend, and Daxos should ask which one the company can repeat on demand.

Three problems with how the sheet computes these

1. M4 dollar retention of 2.93 is not reproducible from the cohort dollars in the same sheet. Only two cohorts have M4 data: January at 3.597 and February at 2.721. Any weighted average must lie between them, and the dollar-weighted one is 2.759 because February is 95.7 percent of the M0 dollars. Producing 2.93 requires putting 23.9 percent weight on January, which is neither its M0 dollar weight (4.3 percent) nor its M4 weight (5.6 percent). Five of six columns reconcile. This one is a 6.2 percent overstatement of the second most quoted number in the pack.
2. The two average rows use different and inconsistent weighting, and the choice flatters both. Dollar retention is dollar-weighted. User retention is a plain unweighted average of cohort ratios, giving a 26-user cohort the same vote as a 150-user cohort. On a user-weighted basis, consistent with the dollar row, M1 retention is 53.4 percent not 56.8 percent and M2 is 45.9 percent not 50.0 percent. Weighting is applied where it raises the number and omitted where it lowers it.
3. $330,819 of January Taiwan deposits, 93.8 percent of the month, belongs to no cohort in the table, and by February the residual is $140. February through May tie out to within four hundredths of one percent, so this is not noise. Either a pre-2026 Taiwan balance vanished entirely in one month, or the January denominator is wrong. It matters because that denominator sets the "10x since the beginning of the year" headline. Restated to the cohort-consistent $21,859, Taiwan growth would read as 236x rather than 14.6x.

Concentration over time, Taiwan

The bull read: top 5 deposits fell from 97 percent to 20 percent and top 20 from 99 percent to 44 percent in six months. The base is broadening on every deposit measure.

The bear read, which is the correct one: the improvement is an artifact of the starting point and it stopped in March. Top 5 at 0.97 in January does not describe a concentrated company, it describes a company with five Taiwanese customers, and you get the transition to 0.22 for free the moment you have a real user base. Strip January and February and the trend disappears: from March to June, while deposits grew 2.64x, top 5 went 0.22 to 0.20, top 10 went 0.29 to 0.30 and top 20 went 0.41 to 0.44. Per-month trends over that window are -0.003, +0.006 and +0.016. Two of three got worse. Four months in which the Taiwan book grew by $2.28M and the top 20 share rose three points means the new money arrived at least as concentrated as the old money. The base is not broadening any more, it is replicating its own shape at larger scale.

And the series that matters most never improved at all. Top 5 volume was 0.30 in January and 0.35 in June, with a positive trend over the whole period and a steeper positive trend over the last four months. Fluidkey's only live revenue is a 30bps fee on large bank transfers, so revenue is a function of volume, not deposits. The quantity the company actually monetises is the one whose concentration has gone backwards. The company will tell the story of the deposit series. Daxos should price the volume series.

What twenty people leaving does

Two caveats, one each way. The concentration series stops in June while deposits run to August, and June is the month Taiwan added $2.1M in one step, which could distort that month; two more months would settle it. Against that, every concentration figure here is Taiwan-only. There is no concentration data of any kind for the other 54.5 percent of the book including the $3.4M Non-KYC line, so the platform numbers above assume the rest of the book is perfectly diversified, which is the most generous possible assumption. If the Non-KYC book is a handful of large wallets, and a balance that barely moves for eight months is consistent with exactly that, true platform concentration is worse than anything shown.

The funnel, and who the users actually are

The top of the funnel is genuinely excellent. A 77.1 percent signup-to-transact rate is far above the 20 to 50 percent typical for consumer fintech, because people arrive through local search and creator content with a specific job to do. The break is at the next step, and it is not shown in the pack as a step. It has to be derived: of the 77.1 percent who transact, 46.0 percent end up holding nothing at all. They came, they moved money through, and they left a zero balance. That is what Fluidkey currently is to most of its users, a rail for funding an Interactive Brokers or Firstrade account. The deposit business is what happens to the minority who stop passing through and start parking.

The $1 to $1,000 bound is arithmetic, not an estimate: 181 users cannot hold more than $181,000 if each holds under $1,000. About 170 people hold 95 percent of the Taiwan money, 23 people generate 69 percent of the transfer volume, and 20 people hold 44 percent of the deposits. This is a whale book with a consumer funnel attached to the front of it.

That is not automatically a criticism. Every wealth platform is top-heavy, and a product aimed at people with $100k to $5m in net worth should be. But whales are the most competitively mobile users in any market: 30bps is material to them, they are already courted by exchange VIP desks and private banking relationship managers, and one departure moves the headline metric by several percent.

CRITICAL What the metrics were measured during

The deposit curve was measured while users were being paid to reach and hold $30,000. Fluidkey's own terms pages document a Frontier campaign 2026-07-07 to 2026-08-01 requiring a $30,000 minimum balance; a Yield Boost 2026-08-16 to 2026-09-30 giving a 50 percent yield boost to users reaching $30,000 by 31 August and maintaining a $30,000 average through September, capped at 2,000 USDC; and a Tokenised Stock Sprint 2026-09-01 to 2026-10-15 paying 5 percent APY on $30,000 of tokenized stock, also capped at 2,000 USDC. A user parking $30,000 for two months to collect a capped $2,000 bonus is earning roughly 40 percent annualised on that position. That is a strong magnet for mercenary balances, and it gives a mechanical explanation for the exact patterns in the data: a $30,000 threshold selects for whale concentration, and paying existing users to grow balances lifts net dollar retention without doing anything at all for user retention. The campaigns end 2026-10-15. The follow-up question is the balance on 2026-11-01.

The banking provider changed six weeks before the call and every banking user was forced to re-verify identity. Docs, last updated 2026-07-08: "We are upgrading to a new banking provider, Noah ... Your former bank details will stop working on July 15, 2026. Transfers sent to the old details after that date will not be credited." The same notice says "SEPA transfers are fully back", implying SEPA had been broken, and says accounts are now "in your name, recipients of your bank transfers will now see your name, not a third party's", which means they previously were not. Taiwan deposit growth decelerates sharply right at that cutover: June +28.3 percent to July, then +9.6 percent to August. That is coincidence until proven otherwise, but it is the obvious question. Any Taiwan retention figure spanning July and August is measured across a forced re-onboarding, so users who did not re-verify present as churn that is operational rather than preference. Neither the campaigns nor the migration came up on the call.

Three independent reads that are not the founder's

Onchain, and it supports him. About $2,187,753 of stablecoin auto-earn inflow across five chains in the week to 2026-09-22, roughly 396 unique Safes funded, 768 new Safes initialised, four of five chains swept to completion with no failed windows. Against his "about a million to $2 million" [09:29] this sits at the top of his range. Deposit size distribution tells the whale story again: on Ethereum the median stablecoin deposit is $3,582, the mean $12,655, the largest in the week $302,724; on Base the median USDC deposit is $161.

App stores, and they say the active base is small. About 10 iOS ratings worldwide: Taiwan 5, United States 3, Argentina 1, Switzerland 1, Hong Kong 0, Spain 0. Google Play download band 1K+. Against a claimed "1k+ Taiwanese users" and 570 transacting cohort users, ten ratings is a normal conversion, so read it carefully rather than as a red flag on quality. What it does is independently confirm the active base is hundreds to low thousands, and it contradicts any reading of the website's "+28,000 users" as an active figure.

Web traffic, and it does not corroborate the growth story. Harmonic monthly visits to fluidkey.com: 11,080 (Oct 2025), 11,640 (Nov), 7,788 (Dec), 8,691 (Jan 2026), 3,431 (Feb), 8,447 (Mar), 4,475 (Apr), 2,402 (May), 10,250 (Jun), 14,020 (Jul), 9,047 (Aug). August is below the July peak and below October and November 2025, with no clear trend, while Taiwan deposits rose 14.6x over the same period. Two fair mitigations: the Taiwan product is a mobile app so the website is a marketing page, and third-party estimates below 15,000 visits a month are unreliable. A weak negative, not a contradiction, and the only independent traffic series available.

And one thing nobody has explained. In the first week of February 2026 there were 387 EURC auto-earn deposits on Base totalling 538,761.85. In the week to 2026-09-21 there were zero on Base and 1,976.06 across three deposits on Ethereum. Both Base windows were swept to completion, so this is clean. The European side of the business appears to have stopped, which converges with the July notice saying SEPA had been broken and that EUR stablecoin ramps were still "coming soon". Spain is still shown as 5.6 percent of deposits.

What is live

That is the whole revenue base: 30 basis points on large bank transfers, generated by roughly 23 people who produce 68.9 percent of volume.

Modelled run-rate on today's book

Three things that break the forward model

1. The transfer fee is 30bps on the call and 10bps in the memo, a 3x difference on the only revenue line that exists. [30:45] "That's 30 basis points today." Memo: "bank transfer fees (10bps)". The charitable reading is that 30bps is today's rate and 10bps is the intended rate at scale, which is a normal thing to plan. It is never presented as a planned price cut, and the entire $120,000 base is generated at 30bps. At 10bps on the same volume it is $40,000. Any revenue bridge assuming volume scales while the rate drops 3x needs volume growth to outrun the price cut before it produces the first extra dollar.
2. The yield fee takes a quarter of the product's main draw. 1 percent APY on a measured gross of 4.40 percent is 23 percent of the user's return, on a rate Fluidkey does not set and cannot protect. Untested against user behaviour, and it will be tested on a base where 61 percent of each cohort already stops transacting by month five.
3. The $100M milestone is stated two different ways and only one is credible.

At $100M the founder's $2M to $3M implies a 2.0 to 3.0 percent take. Applying his own component rates: 0.759 percent from yield, plus transfers at 0.459 to 0.918 percent at the memo's 10bps or 1.376 to 2.753 percent at the call's 30bps, plus 0.068 percent from trading. Total 1.29 percent at 10bps, 3.58 percent at 30bps. So the milestone is reachable only if the fee stays at 30bps and volume scales in proportion to deposits. At the memo's own rate it lands at $1.29M, well short of the bottom of the range.

The longer-horizon claim does not close at all. Memo: "At 100,000 customers and $1b in deposits, this leads to $50m in annual revenues", implying $500 ARPU. That is a 5.0 percent take against a modelled 1.29 to 3.58 percent, so $12.9M to $35.8M rather than $50M, and closing the gap requires card and lending, neither of which exists, to carry roughly half of all revenue. And the ARPU assumption does not survive the company's own distribution: $1bn across 100,000 customers is a $10,000 average balance, worth $100 at a 1 percent yield share, leaving $400 to come from transfers, trading, interchange and lending. For any one of those alone the average user would need roughly $400,000 of annual transfers, or $200,000 of annual trading, or $80,000 of card spend, or a $20,000 permanent loan balance. Fluidkey's own data says 73.7 percent of Taiwanese users generate under $1,000 of annualised transfer volume and together account for 0.1 percent of it. The $500 ARPU is calibrated on the whale, not on the user. Today's realised figure is about $120 per Taiwan user.

Burn, runway and price

A $2M raise at today's team size is three to four years, which is genuinely comfortable. The memo's use of funds is "scaling creator-led distribution, expanding local marketing and engineering capacity, and shipping the card and lending roadmap", so headcount and paid spend both rise. A realistic post-raise burn of $1.2M to $1.5M gives 16 to 20 months, which is adequate but not generous against a $100M deposit milestone.

Wealth and neobank platforms are commonly valued at 2 to 5 percent of AUM. At 3 percent this book is worth $340,048 and at 5 percent $566,747. Fluidkey is priced at roughly 22 to 37 times what the existing deposit base is worth on a conventional AUM multiple. That is not automatically an objection, since nobody prices a pre-seed on its current book. It locates precisely what is being bought: essentially none of the $12.5M is the current business. All of it is the forward case, and the forward case is stated two different ways in the two primary documents.

What Fluidkey holds today: nothing, anywhere. No FINMA licence, no Swiss SRO affiliation, no Taiwan VASP registration, no securities licence in any jurisdiction. The company says so itself in the memo: "As we expand our feature set, we will pursue Swiss SRO affiliation to offer some financial services directly" and "Taiwan is rolling out their new VASP framework ... We are preparing to partner with licensed entities and get relevant licenses as the business grows." Credit where due, the memo claims no licence it does not have. The legal theory is that self-custody plus licensed partners keeps Fluidkey outside every perimeter.

Also on the record: the Swiss company's registered corporate purpose is software and digital privacy, not financial services, and while there is a general commercial catch-all, the object clause does not describe the business the company now runs. An acquirer's counsel will read the object clause against the product. The company also has no auditor; it waived limited audit at incorporation, so no Fluidkey financial statement has ever been independently reviewed. The Dedaub and Ackee reports are code audits of open-source components, not financial audits.

The "Non-KYC Deposits" line, handled squarely

The workbook carries a line item literally named "Non-KYC Deposits". It was $3,008,292 in January, 60.7 percent of the book, and $3,414,995 in August, 30.1 percent. It is the second largest line in the company and it grew 13.5 percent in eight months, so it is being diluted, not wound down. In 47 minutes the founder never names it, never quantifies it, never says what happens to it, and nobody asks. The phrase "non-KYC" is never spoken on the call by anyone. It is also absent from the memo, which reports only the $5.2m Taiwan figure, so an investor reading the memo alone would not know that 30 percent of deposits sit outside any KYC perimeter.

The benign reading is probably right for custody. The line almost certainly means users who never completed identity verification because they never used the Bank feature, which is the legacy V1 stealth-wallet cohort. The flatness across eight months fits a cohort nobody is marketing to. The architecture supports the claim: keys in a Privy embedded wallet on web or the secure enclave on mobile, 1/1 Safe accounts, a published third-party audit of the stealth kit, and an independently verified counterfactual property. The Terms say "We do not: offer custody and back-up of your private keys ... engage in any regulated financial activity." The founder's answer to Jake's direct question at [23:15] is the standard wallet-software argument, at [23:26]: "it's a self custodial wallet in the end. Right. So the same way you don't have to KYC to use Metamask or Rabi, you don't have to do that to use fluidkey. But yeah, as soon as we touch any regulated Fiat Rails you do have to KYC there." On the AMLA, MiCA and FinCEN FIN-2019-G001 tests, a provider that never has power of disposal over client assets is generally outside the perimeter and the identification duty sits with the CASP or the regulated partner. That analysis holds.

The qualification is the whole of the remaining risk, and it is unresolved. Nothing in the documentation gates Earn or Tokenized Stocks on identity verification. The KYC guide scopes verification to the Bank feature only. The Fee Structure page describes Auto-Earn and tokenized stock trading with no verification precondition. The Tokenized Stocks page pushes eligibility onto the user rather than gating it. And the founder's own sentence reads, on its natural construction, as putting yield and stocks on the pre-KYC side of the line: [20:32] "you have the account ready, you can start sending and receiving stablecoins already." [20:40] "You can start earning yield, getting access to tokenized stocks as soon as you want to set up your ACH details and any bank rails that's where you'll have to kyc."

That sentence is ambiguous as transcribed and nothing here is built on it alone. Read with [23:36], where the KYC boundary is drawn at "any regulated Fiat Rails" and nowhere else, and with the documentation's silence, the probable position is that an unverified user can allocate to Morpho and Aave vaults and can buy Ondo structured notes. If that is right, the self-custody characterisation holds for custody but not for distribution, and the line stops being a dormant legacy artifact and becomes an active distribution channel for instruments carrying eligibility conditions. This is the single highest-value verification step in the whole diligence and it costs one afternoon. It was not resolvable from public sources.

What an acquirer or a Series A lead will do with this line

Not read it as fraud. Read it as three things. It is an unpriceable pocket: a regulated acquirer, a Robinhood or a broker as Mark raised at [27:36], cannot inherit 30 percent of a balance sheet whose customers it cannot identify or geolocate, so the outcome is carve-out or run-off with non-converting deposits written to zero. It is a label that reprices the whole book, because "Non-KYC Deposits" written in the company's own hand gets read aloud in an investment committee, and a bank partner's financial crime team does not have the luxury of the charitable reading. And it is a governance signal: the favourable reading is available and probably true, and the company has not done the twenty minutes of work needed to make it obvious by renaming the line, showing the cohort as unattributed geography, or explaining it in the pack.

Tokenized stocks, handled squarely

What the user actually holds. Not a share. An Ondo tokenized stock is a structured note issued by Ondo Global Markets (BVI) Limited, a bankruptcy-remote SPV, backed 1:1 plus a buffer by the corresponding security held at a regulated custodial broker-dealer, with no voting or information rights. Ondo Finance Inc., a Delaware corporation, provides tokenization services and holds equity in the BVI issuer. A base prospectus is approved by the Liechtenstein FMA and notified into certain EEA states, with Final Terms for one product filed with the Malta FSA under ISIN VGG7001AAJ30. Distribution outside the US is under Regulation S. Trading runs on BNB Smart Chain through Bungee with just-in-time mint and redeem during market hours.

Are they securities? Almost certainly yes in substance. Transferable debt instruments whose return is referenced to listed equities, issued under an EEA-approved prospectus and distributed under a US private-placement safe harbour. Relying on Regulation S is itself an admission that securities law applies.

The good news, and it is real. Ondo's eligibility page prohibits the US and fourteen other jurisdictions and restricts the EEA, UK, Switzerland, Singapore and Hong Kong to qualified or professional investors. Taiwan appears on neither list. So Fluidkey's largest market is, from the issuer's compliance perspective, permitted retail territory, and Fluidkey is not knowingly breaching its issuer's terms in Taiwan. That removes what would otherwise have been the worst finding in this section.

The bad news is specific and it is Hong Kong. Hong Kong is Fluidkey's second market at 7.7 percent of deposits, and the founder puts it at "about 10% of our total activity" [36:00]. Ondo restricts Hong Kong to Professional Investors, a roughly US$1 million portfolio threshold, while Fluidkey's own stated target customer is "$100k to $5m in net worth". The majority of the declared target band in Hong Kong does not qualify. Singapore is also professional-only for Ondo, and Singapore is named in Fluidkey's referral programme as a qualifying region and by the founder as a next market [38:59]. And Fluidkey's own eligibility disclosure omits both. Its page names "the EEA, UK, and Switzerland" as professional-only and tells users to "check whether you are eligible on Ondo's eligibility page". No jurisdiction attestation, no professional-investor certification and no blocking was found anywhere in the documentation. Combined with the Non-KYC finding, the likely position is that Fluidkey knows neither the residence nor the investor status of many of the people buying Ondo notes.

The three consequences, with very different probabilities. The likeliest and most damaging is commercial, not legal: Ondo, with a prospectus-approved programme and over $1bn TVL, has every incentive to enforce its distribution representations, and the cheap remedy is to geo-restrict or terminate the integration, which removes the feature the memo calls the thing that makes Fluidkey a primary wealth account. The founder's substitutability answer [40:12] does not cure this, because every alternative issuer has the same or narrower perimeter. Second, regulatory exposure in Hong Kong under the SFO and Singapore under the SFA, where the realistic outcome at this size is a cease-and-desist and a public warning naming the app rather than prosecution. A public regulator warning is the tail that matters, because it is permanent, searchable, and fatal to the bank and issuer partnerships the whole model rests on. Third, investor recourse, bounded today by the roughly $873,000 in tokenized stocks, scaling linearly with the feature the company intends to scale fastest.

And the estate tax claim, which is the memo's actual moat

The memo's headline wedge against IBKR: "Because these assets are structured as non-US-situs, they are shielded from the 40% US estate-tax exposure that offshore broker accounts carry." Four things about it. One, the underlying tax fact is real: a non-resident alien has a US estate tax exemption of only $60,000 on US-situs assets with rates to 40 percent, and using a foreign broker does not change the situs of US-listed stock. Two, the structuring argument is arguable but untested: a debt obligation of a foreign issuer is generally non-US situs, so a BVI note tracking Apple is a defensible position, but it is a position and no ruling, opinion letter or authority on tokenized equity situs was found (checks: IRS international taxpayer pages, Bogleheads non-resident alien guides, Wealthspire, AbitOs and Skybound technical guides, Ondo's own documentation). Three, and decisively, Ondo itself makes no such claim on its product page or in its eligibility documentation, and Fluidkey's own product docs do not make it either, saying instead "Nothing in Fluidkey or this page is investment or tax advice." So it is a tax conclusion marketed as settled, to retail, about a product the supplier does not describe that way, in the sales document and not in the product document. Four, it is not proprietary: it is a property of the instrument, so any Ondo or xStocks distributor can copy it word for word.

THE RISK THAT CAPS THE RATING Taiwan

The framework. Taiwan's AML registration regime for VASPs took effect 2024-11-30, with applications due 2025-03-31 and registration by 2025-09-30. Operating without registration is a criminal offence under the Money Laundering Control Act: up to two years imprisonment and NT$5 million, with corporate fines to ten times that. The offshore question is addressed expressly: enterprises established overseas may not provide virtual asset services in Taiwan unless they register a company or branch under the Company Act and complete AML registration, and the FSC's stated intent is that overseas platforms without Taiwan company registration "should not solicit business in Taiwan or from its citizens". The Virtual Asset Service Act passed third reading on 2026-06-30, moving Taiwan to licensing across seven categories with penalties up to seven years and NT$100 million, expected to commence Q1 2027 at the earliest, with twelve months to apply and twenty-one months to be licensed. The transition relief runs to firms already AML-registered. Fluidkey is not, on its own account.

The company's posture, quoted. "Offshore financial platforms have long served Taiwanese users and can operate within the existing framework via referrals, creator partnerships, and localized content. This is the same posture under which offshore brokers and exchanges built their Taiwan user bases."

Why that does not describe what Fluidkey is doing. The defence rests on being a platform Taiwanese users happen to find. The evidence is that Fluidkey actively solicits the market:

• The app and documentation are localised into traditional Chinese with a full parallel zh-TW docs tree, and the founder confirms Chinese-language customer support [39:20].
• Fluidkey publishes Chinese-language step-by-step guides for funding named US brokers (IBKR, Charles Schwab).
• The referral programme defines a qualifying new user by jurisdiction, explicitly Taiwan, Hong Kong, Macau, Singapore and Malaysia, and pays 5 USDC per KYC'd funded user.
• Go-to-market is creator-led by design and the memo says it will now amplify creators "with paid distribution".
• The founder has relocated and hired locally.

Localised product, paid creator distribution, a team on the ground and a jurisdiction-scoped referral bounty is solicitation. It is the opposite of the passive posture the memo relies on. That is not an oversight in the data, it is the stated strategy.

Two mitigations are real. Fluidkey operates no fiat on-ramp or off-ramp in Taiwan; the bank leg is a US or EU account in the user's own name issued through Noah, so the money movement is the user's own cross-border activity rather than a Taiwan payment service. Noah's own restricted-geos page confirms Taiwan is eligible on every rail. But that is precisely the structure the FSC's offshore language is aimed at, so it mitigates the licensing question without resolving the solicitation question. Note also that the memo's market framing rests on Taiwanese residents moving assets offshore for geopolitical hedging, a politically sensitive capital flow narrative in Taiwan that increases rather than decreases the chance of official attention.

Separately, the securities perimeter. Taiwan regulates the offering of foreign securities to residents independently of the VASP question: no person may act as agent for offering offshore funds without FSC approval, private placement may not involve public solicitation, and brokering trades in foreign securities is an activity of licensed securities firms. Taiwan's own STO regime is restrictive, professional investors only, NT$300,000 per project per natural person, trading confined to licensed dealers, proceeds in New Taiwan Dollars. Whether an Ondo note bought by a Taiwanese resident on a foreign app through an aggregator on BNB Smart Chain is caught at all is a genuine open question and no published FSC determination was found either way. What is clear is the direction of travel: Taiwan restricts retail access to tokenized securities and channels foreign securities activity through licensed local firms, and Fluidkey markets 400-plus US equity and ETF exposures to Taiwanese retail in traditional Chinese with a 5 percent APY promotion attached. That is the opposite shape to what the local regime permits domestically.

Two more items nobody raised

Company-funded APY on securities-linked holdings. The Tokenised Stock Sprint pays 5 percent APY, set and funded by Fluid Privacy SA, tiered by a US$30,000 holding, conditioned on maintaining balance, paid in USDC after 15 October. Auto-Earn itself is a genuine pass-through and that architecture is the strong case: the user never transfers title, there is no pooling or rehypothecation, no balance-sheet promise, and on insolvency the customer loses an interface rather than an asset. That distinguishes it cleanly from BlockFi ($100M to the SEC and 32 states), Gemini Earn, Celsius, Vauld and Hodlnaut, all of which promised a rate funded by lending customer assets. The campaign APYs reintroduce exactly the elements the pass-through removes: an advertised rate, set and funded by the company, tiered by deposit size, conditioned on holding for a period. Weakened considerably by genuine self-custody and by being capped and time-boxed, but paying 5 percent APY on holdings of securities-linked notes to retail is the least defensible single practice found here. It is also entirely discretionary and could stop tomorrow.

Hide Trail. Not in the memo, not on the call. From docs.fluidkey.com/readme/advanced-privacy: a feature "powered by Houdini Swap, a swap aggregator with a unique private mode that routes funds through two exchanges to remove the link between your existing address and the newly funded one", for incoming transfers of $60 to $100,000. Houdini's own materials describe using Monero as a tunnel between exchanges with swaps "completely untraceable by the recipient, exchanges, or blockchain explorers", usable "without creating an account, going through KYC". Houdini denies being a mixer on the grounds that it never pools or custodies funds and states it runs AML, OFAC screening and geo-blocking. Fluidkey does not operate it, and integrating a third-party tool is weaker conduct than running one. But Fluidkey markets it, names it "Hide Trail", and places it inside a product holding $11.3M of customer balances alongside an unattributed 30 percent cohort. The live precedent is Samourai Wallet: both founders pleaded guilty and were sentenced in November 2025, five years and four years, for conspiracy to operate an unlicensed money transmitting business, on non-custodial software. The counterweight is Van Loon, where the Fifth Circuit held OFAC's Tornado Cash designation unlawful, so sanctioning software as such has been curbed while prosecuting the people who operate or market it has not. A reviewer who sees the non-KYC cohort and Hide Trail on the same page will ask whether unverified users can use Hide Trail. That could not be determined from public sources.

And the customer contract is two and a half years stale. The Terms of Service were last updated 3 January 2024, still say Fluidkey does not "engage in any regulated financial activity", and make no mention of the Bank feature, yield, tokenized stocks, KYC or eligibility restrictions. The Privacy Policy, by contrast, was updated 1 July 2026 and is genuinely sophisticated: biometric KYC on explicit consent, sanctions and PEP screening, named controller allocation between Fluidkey and its regulated partners, a US state privacy notice and a GLBA carve-out. The gap between the two documents is the clearest available evidence that legal effort went into data protection and not into the financial services perimeter. The enforceable contract for $11.3M of customer balances does not describe the product.

Net regulatory assessment

This is a company whose product has outrun its compliance architecture by roughly eighteen months. The pattern is survivable and it is a capability question about the next stage rather than a verdict on this one. The favourable answers were mostly available and nobody assembled them, which for a financial product holding $11.3M of customer money is the concerning tell. Most of the specific gaps are cheap and fast to close and none requires a licence: a counsel opinion, a rewritten ToS, an eligibility gate, a renamed line item. The one that is not cheap is Taiwan, and it is the one that carries the whole company.

What market this actually is

By revenue it is a cross-border payments business: 30bps on large bank transfers, about $120,000 a year, from a handful of large senders. By deposits it is a stablecoin savings business: 75.9 percent of the Taiwan book in yield-earning stablecoins. By origin it is a crypto privacy company, and that business did not grow.

The honest definition: a stablecoin dollar account with automated DeFi yield and resold tokenized US equities, sold to Taiwanese mass-affluent retail who already hold US offshore brokerage accounts, acquired through a cross-border money movement hook and monetised at 30bps on the largest senders. It is not a privacy company and it is not yet a wealth platform. It is a payments wedge with a savings product bolted on and a brokerage product resold.

Who the Taiwanese user actually uses

The strategic problem with the hook is structural, not competitive: a rail that makes the incumbent broker easier to use makes the incumbent stickier. Converting "pipe into my broker" into "replace my broker" is the entire thesis, and the funnel is where it is tested. Today 77.1 percent transact, 41.6 percent ever hold over $1, 17.1 percent hold over $1,000. Three quarters of the people who show up use it as a pipe and leave.

The crypto-native set, of which the founder named one

He named one of these in 47.6 minutes, and only because Jake named it for him. Combined Harmonic-recorded funding of the five he omitted is $384.2M. Fluidkey is raising $2M.

Local players also beat Fluidkey on the two things that decide a savings relationship, a TWD on and off ramp and local legal recourse: MaiCoin/MAX ($4.8M, AML-registered, first in line for an FSC licence), BitoPro ($10M, TWD deposits, bank partnerships with Far Eastern, KGI and Cathay United), XREX ($42.75M, 109 heads, investors include E.SUN Venture Capital and CDIB Capital, best positioned locally to offer exactly Fluidkey's product with a licence), Rybit ($10M, Chunghwa Telecom and PChome), and in Hong Kong HashKey ($336.6M, IPO December 2025) and OSL ($126.9M).

Tokenized equity access is not scarce

Ondo Global Markets has over $1.17bn TVL, about 222,600 holders, 450+ assets and over $18bn cumulative volume. Fluidkey's $873,000 tokenized stock book is about 0.07 percent of it. Ondo also distributes through MetaMask, which has an order of magnitude more users than Fluidkey will plausibly have this decade. Elsewhere: xStocks/Backed was acquired by Kraken (announced 2025-12-02) and is live on Kraken, Bybit, Bitget Wallet, Gemini and KuCoin; Robinhood Europe holds MiFID II and MiCA licences from the Bank of Lithuania, launched 200+ stock and ETF tokens in June 2025 and put Robinhood Chain on mainnet in July 2026 with 24/7 tokenized stocks; Dinari is the regulated US path with an SEC-registered transfer agent, a FINRA and SIPC member broker-dealer subsidiary, and 724 tokenized stocks launched 2026-08-04. The headline asset feature is a commodity available to anyone who integrates a swap route.

And Robinhood is coming. The literal pitch is "Robinhood for Taiwan". Robinhood bought Bitstamp for $200M, made Singapore its Asian headquarters, and received MAS in-principle approval for a securities brokerage in April 2026. Asked about Robinhood the founder said [27:12] "We don't have any, you know, direct relationships with, with Robinhood specifically." The thesis is a bet that the incumbent stays out. The incumbent is visibly on its way in, with licences.

The graveyard, and what killed each

On cause one, Fluidkey is genuinely different and deserves the credit. The yield is a Morpho vault, on chain and inspectable, the user holds their own keys, and there is no balance sheet promise and no rehypothecation to an opaque desk. That is a structurally better failure mode than every name above. It is not risk-free (curator, oracle and bad-debt risk remain, and a 1 percent fee wrapper does not change who eats a vault loss), but a report that skipped this would be unfair.

On cause three, there is a documented past failure of the core hook. Blocktrend ep306, dated 2025-10-21, asks directly "為什麼 FluidKey 無法入金 Firstrade?", why FluidKey cannot deposit to Firstrade, and describes the pathway as stuck. The same piece warns that US brokers and fintechs may be hostile to crypto, with consequences from returned transfers to permanent account closure. Firstrade is one of the two brokers the founder named [12:11]. Caveats stated plainly: the article is eleven months older than the call, it may have been resolved, and no primary confirmation either way was found since Fluidkey's own site names no supported brokers. Treat it as a documented past failure of the core hook, and one direct question for the founder.

The verdict from this lens

Market attractiveness is genuinely good. The pull was organic and unmanufactured, which is the hardest signal to fake. Taiwan has a high savings rate, an entrenched offshore-broker habit, a dense personal-finance creator culture, a geopolitical hedging motive and enough wealth that a small share of wallet is a real business.

Competitive defensibility is poor and poor in a diagnosable way. Fluidkey rents every component. The yield is a public Morpho vault. The tokenized stocks are Ondo's, resold. The fiat rails are one vendor. The privacy layer is a published Ethereum standard whose reference implementation was built as an unfunded public good by the people who wrote the spec. The one differentiator in the memo that is not rented, the estate tax argument, is an untested tax position the supplier does not assert and that any competing distributor can copy verbatim. The only asset genuinely owned is Taiwan-specific distribution, built by five people, in a market the memo itself describes as "small enough that global players don't prioritize it", and that four better-capitalised companies can enter on a quarter's notice.

(1) Market and competition: 4.0

The market is real and the demand was organic, which is the best fact in the deal, and the category ceiling is high (Nubank $45.3bn deposits, Bitso $2.2bn, Felix $1.4bn). Against that, defensibility is close to absent: every component is rented, and the competitive set is ether.fi at 500,000 users, KAST at $600M on 1m users, RedotPay at $194M raised and already in Hong Kong, and Robinhood holding an MAS in-principle approval, none of which the founder raised unprompted. A slice small enough that global players ignore it is also a slice too small to build a fund-returner in unless it generalises, and the generalisation path runs into Ondo's professional-investor gate in exactly the two markets named. Real market, weak position, net median.

(2) Product and wedge: 6.0

This is the strongest part and it is verified rather than asserted. Six ENS resolutions returned six distinct counterfactual addresses. The one custom contract is small, single-file, non-upgradeable (proxy_type: null), verified on Blockscout, audited, and narrow by construction: the config is content-addressed by keccak hash of the whole config list, so the 2-of-3 owner multisig cannot repoint an existing user's auto-earn at a malicious vault. Published technical docs match onchain state byte for byte (config hash 0xd55dfa14...). The Ackee High finding on cross-chain replay is remediated in deployed code (block.chainid present). Recovery without the company is real, deterministic and open source. Two people built all of it and ship at v1.73.0 eleven months after launch.

Against that: no Taiwan on-ramp, so Fluidkey owns the second mile not the first; about 5.5 engineer-days of external review in total, none of it covering the mobile apps, backend, ENS gateway, indexer, banking integration or tokenized-stock routing; the last audit is 19 months old; no bug bounty, no disclosure policy, no security contact; the realistic attack path is the frontend, which Dedaub and an independent Taiwanese reviewer both flagged and which nobody has audited; and single points of failure in one AWS Lambda and two relayer EOAs. Genuinely good, with the unaudited surfaces holding it below 7.

(3) Traction and its quality: 4.0

Deposits genuinely tripled from $4.96M to $11.33M and Taiwan genuinely did 14.6x, and unusually for a seed company the scale is independently corroborated: about $2.19M of measured weekly inflow against a claimed $1M to $2M. That is worth real credit. Everything about the quality pulls the other way. 75.5 percent of the growth is one geography and the rest of the book compounds at a rate indistinguishable from crypto price drift. User retention is 0.39 by month five and the 4.69 dollar retention headline is computed over 24 people on an absolute gain of $80,767. Twenty individuals hold 44.2 percent of the Taiwan book and concentration stopped improving in March. Only 17.1 percent of signups ever hold more than $1,000, and 46 percent of those who transact end with a zero balance. The numbers were produced during $30,000-threshold paid campaigns ending 2026-10-15. Ten App Store ratings worldwide, a flat web traffic series, and a European business that went to zero. Real traction, poor quality, median.

(4) Business model and unit economics: 3.5

The shape is plausible and conventional for a neobank. Almost none of it exists. Live revenue is about $120,000 at 104x the ask, generated by roughly 23 people. The yield share, the largest planned lever, takes 23 percent of a 4.40 percent gross yield the company does not set, was promised for September and is not visible. The only live fee is quoted at 30bps on the call and 10bps in the memo, a 3x disagreement that decides whether the $100M milestone produces $1.29M or $3.58M. The $500 ARPU target is calibrated on the whale, against a distribution where 73.7 percent of users generate 0.1 percent of volume and today's realised figure is about $120. The $50M-at-$1bn case needs card and lending, neither of which exists, to carry half of all revenue. Burn is unknown because nobody asked. Below median.

(5) Team: 5.0

Two founders wrote the cryptography themselves and it works; the CEO still commits to the core library in 2026; the Antonio and Federico engineering pair has moved together through five ventures since 2016; the founder has genuine operating seniority across eight years and six African markets ending as Co-CEO and has physically relocated to the market. Against that: five people, one operating hire in two years and nine months, the entire Taiwan organisation is one person about three and a half months into the role with roughly two years of post-graduate work experience and a concurrent outside role, no compliance, finance or commercial leader, and an engineering bench of two people who have never worked apart. The prior outcomes are softer than presented: Pulse was a Ringier subsidiary he joined four years after founding, the $50M is unverified, and Antonio's Ianum sale returned no record found. Above median on technical credibility, below it on bench.

(6) Founder trust, disclosure and regulatory posture: 3.5

Personal candour on the call was genuinely above average and is credited above: six volunteered weaknesses, a flat no on Robinhood, no lead disclosed unprompted, zero closing pressure, a live demo, and the metrics workbook handed over on request. Nothing adverse exists on any sanctions list, court docket, regulator warning list or in any press, and the founders are fully named in the Swiss register with no nominee layer. The score is low anyway, because trust is not only honesty. Nine of nine checked second-order figures run in the same direction. The headline Taiwan number given verbally is 16 to 36 percent above the company's own file. The memo says acquisition is organic while the company's own docs publish a per-user referral bounty and two deposit campaigns. Three material facts were absent from a 47 minute call and from the memo: the July banking migration and forced re-KYC, the $30,000 campaigns running through the months being shown, and a line item that is 30 percent of the balance sheet. The customer contract is two and a half years stale and denies engaging in regulated financial activity. The company holds no licence anywhere and its stated Taiwan posture describes the conduct the FSC has named as the thing not to do. Hong Kong retail appears able to buy instruments the issuer restricts to professional investors, and the company's own eligibility page omits Hong Kong. The favourable answers were mostly available and nobody assembled them.

The adjustment, and why. Minus 0.25 for a cross-component fact the mean treats as six independent scores when it is one bet. 75.5 percent of the growth, essentially all of the thesis, the only distribution asset the company owns, and the single dated regulatory clock all sit on the same square: Taiwan. A mean implicitly assumes some diversification across the parts. There is none here. Remove Taiwan and four of the six scores collapse together rather than one at a time.

A reader who declines the adjustment lands on Base 4.33 and COMPANY 5.33. That reader is not wrong; the argument against the dock is that concentration in one wedge is what a seed company is supposed to look like and docking for it penalises focus. The difference does not change the verdict band or the recommendation.

The Alliance position, which nobody raised

The Notion database holding the memo is titled "ALL17 Memos", placing Fluidkey in Alliance batch ALL17, which started 2026-05-11. The founder confirms graduation [42:01]. Harmonic lists Alliance as the sole recorded investor. Alliance's published standard deal, from alliance.xyz fetched 2026-09-22: "$400k upon admission + $400k follow-on at seed" and "Alliance invests at a $4M post-money valuation via SAFE with a 1:1 token side letter." Fluidkey's executed documents were not seen and no deviation has been disclosed, so treat this as the likely but unconfirmed shape. Three consequences, each a question to put in writing:

1. Alliance sits at a cap 3.1x below what Daxos is offered, set about four months earlier, on a business whose revenue at the time of the call was $10,000 a month. $400k at $4M post is about 10 percent of the company on conversion, the largest outside position ahead of Daxos.
2. Alliance holds a 1:1 token side letter as standard. Fluidkey is a crypto-native product and neither the memo nor the call mentions a token, which does not remove the asymmetry, it only means nobody has priced it. Daxos should either ask for the same side letter or price its absence.
3. The standard follow-on is "$400k at seed", which is 20 percent of this $2M round. If Alliance takes it, the amount available to new investors is $1.6M less whatever the angels hold. The founder did not mention it. Ask whether Alliance has committed, declined or is still deciding, and whether the follow-on comes out of the $2M or sits on top.

There is also a reconciliation problem in the pre-seed. The memo says "$1.25m pre-seed from Alliance, Prelude, and angels". On the call the founder separated the two events: a Prelude-led round "over a year ago" at "about $5 million" average valuation [41:48], [43:24], and separately joining Alliance in the latest batch [42:01], which was May 2026. So the memo's single line appears to bundle a 2024 or early-2025 round at roughly $5M with an Alliance cheque from May 2026 at $4M post. Two events roughly fifteen months apart at two different caps. The word "average" in "the average valuation was about $5 million" implies more than one instrument at more than one price within the pre-seed itself.

And no priced round has ever closed. The Swiss register shows share capital unchanged at CHF 100,000 authorised and CHF 50,000 paid since December 2023, with no capital increase published. A Swiss SAFE or convertible loan does not appear in the register, so this does not prove there was no raise. It does prove everything to date is convertible, and it means half the nominal share capital is still unpaid, a small but real shareholder liability.

FOR DAXOS: 4.5 / 10. WATCH. Do not write at $12.5M today.

Is there a round? Yes, and it is open and unled. $2M on a SAFE at $12.5M, just over $200,000 received from unnamed angels, no lead as of 2026-08-24, no deadline, no pressure. A $100k to $500k Daxos cheque fits mechanically inside a $2M round without difficulty, and the absence of a lead means Daxos could be early rather than filling an allocation. That is the part of the fit that works.

Is the price right? No, on this market's benchmarks. At $12.5M Daxos pays 104x live founder-claimed revenue, 58x a fully-switched-on model of today's book, and 1.10x deposits under management against a conventional 2 to 5 percent of AUM, which would value this book at $340,000 to $567,000. None of that is fatal on its own, because nobody prices a pre-seed on its current book. What matters is the comparison set. Alliance bought in at a $4M post cap roughly four months earlier with a token side letter attached. KAST was priced at $600M on 1m users and $5bn of annualised volume, 48x Fluidkey's cap for something on the order of a thousand times the user base. RedotPay raised $107M at Series B already domiciled in Hong Kong. The $12.5M is not absurd for a seed with real organic pull, but it is a full price for a product surface five months old, measured during paid campaigns, in a jurisdiction with a legislated licensing regime arriving in one to two quarters that the company does not hold.

Can a seed cheque return a fund? In principle, because the category ceiling is genuinely high: Nubank at a $30bn market value, Bitso at $2.2bn, Felix at $1.4bn, KAST at $600M in eighteen months. A 100x from $12.5M means $1.25bn, which is Bitso and Felix territory, and it requires deposits to grow roughly 88x. That path is not arithmetically absurd. It runs straight through a Taiwan licensing perimeter the company does not hold and cannot hold at five people, and then through a Hong Kong and Singapore expansion where the issuer of the retention-anchor feature restricts retail access. So the ceiling exists and the path to it has two named gates on it, both of which have to be cleared by a company with no compliance function.

What Daxos can add: not much where it counts. Daxos brings a US and crypto investor network, and Mark offered introductions to other VCs at [45:40], which Moritz accepted. There is no Taiwan relationship to offer, no regulatory help, no banking or issuer relationship, and no Robinhood connection (the founder confirmed none and Daxos has none to give). The honest assessment is that Daxos would be capital, plus intros, plus a partner who read the workbook carefully.

The Alliance position cuts both ways. It validates the company (admission to a competitive batch) and it sits ahead of Daxos at a 3.1x lower cap with a token claim Daxos would not have. A Daxos SAFE at $12.5M without an equivalent token side letter is structurally worse paper than the paper already on the cap table.

Where 4.5 comes from. The round mechanics fit and the founder is someone worth a relationship. The price is full rather than cheap, the entry is behind a better-priced insider with better terms, Daxos adds little in the only market that matters, and two of the three things that would make this a conviction cheque (post-campaign numbers, a Taiwan counsel opinion) are cheap and will exist within six weeks.

VERDICT: WATCH. This is not a pass on the merits. It is a pass on the price and the timing. Mark's own instinct on the call was right and should be held: [45:10] "once you find a lead, you know, we can. We can circle back with. With commitments, like on. On a timeline." The company is real, the founder is worth knowing, and the two decisive facts arrive on a known schedule.

Three things that must be true before any cheque

1. A written Taiwanese counsel opinion on whether the current model requires VASP registration, obtained before the round closes rather than after. This is the risk that takes the growth story to zero rather than merely slowing it, and asking for it is a cheap test of how this team responds to being asked. Acceptable alternatives that retire it: a signed agreement with a locally registered VASP, or a filed Taiwan company or branch registration with an AML application in progress.
2. The deposit balance, Taiwan cohort retention and concentration as of 2026-11-01, after the campaigns end on 2026-10-15, split between campaign-eligible and non-eligible accounts. Any diligence treating the August number as an organic run-rate is mispricing the company. This costs the founder one spreadsheet and it is the single highest-value follow-up in the file.
3. A definitive answer, ideally a screen recording of the product, on whether an unverified account can allocate to Earn, buy Ondo tokenized stocks, or use Hide Trail. This one fact determines whether the self-custody defence covers custody only or also distribution, and it is pivotal to four separate risks. One afternoon of work.

Terms that must change

4. A repriced entry. Not above roughly $8M on this evidence. The anchors: Alliance at a $4M post cap four months earlier, a deposit book worth $340,000 to $567,000 on a conventional AUM multiple, and a product surface one quarter old. At $8M with the three conditions above satisfied, a $250,000 cheque is defensible. At $12.5M it is not.
5. A 1:1 token side letter matching Alliance's, or an explicit written statement that no token is contemplated plus an MFN clause that captures one if it appears.
6. Confirmation of whether $12.5M is pre-money, post-money or a cap, plus the discount, the MFN and pro rata rights, none of which were stated.
7. The full cap table by instrument, date, amount, cap and discount, including the Alliance SAFE and its side letter, reconciling Harmonic's $0, the memo's $1.25M and the founder's $1.2M; and whether Alliance's $400k follow-on is inside the $2M or on top.

The questions to send, in order

8. On the call you said Taiwan deposits were "6 to 7 million". The workbook and the memo both say $5.2m for August. Which is right?
9. The memo says acquisition is organic. Your own docs publish a referral programme paying 5 USDC per KYC'd funded user in Chinese-speaking markets and two campaigns requiring a $30,000 balance. What did incentives cost in July, August and September, and what share of Taiwan deposits is campaign-conditional?
10. Who was the banking provider before Noah, why did the relationship end, and what percentage of users failed to re-verify by 2026-07-15? Is the July to August deceleration from +28.3 percent to +9.6 percent attributable to it?
11. "Non-KYC Deposits" is 30 percent of the book and appears nowhere in the memo. What is it, how many wallets, what is the balance distribution, will the 1 percent yield fee be charged on it, and what is its regulatory treatment?
12. Is the transfer fee 30bps (call, docs) or 10bps (memo)? Which is in the plan?
13. Is the $100M milestone six months (call) or twelve (memo)?
14. The M4 dollar retention of 2.93 does not reconcile to the cohort dollars in the same sheet, which give 2.759. What is the calculation? And $330,819, being 93.8 percent of January Taiwan deposits, belongs to no cohort and is gone by February. Which is it: a restated denominator or a near-total loss of the opening book?
15. Ondo restricts Hong Kong and Singapore to professional investors. Hong Kong is 7.7 percent of your deposits and your own eligibility page does not name it. How is eligibility screened, and has Ondo confirmed in writing that your distribution arrangement satisfies its partner representations?
16. Produce the US tax counsel opinion behind the "non-US-situs, shielded from the 40% US estate tax" claim, or remove it from sales material. Ondo does not make this claim and neither do your product docs.
17. What fraction of Taiwan inflow arrives via a local licensed exchange rather than directly, given there is no TWD rail?
18. What happened to the European business? 387 EURC auto-earn deposits totalling 538,761.85 on Base in the first week of February 2026, zero on Base in the week to 2026-09-21. Spain is still shown as 5.6 percent of deposits.
19. Did the Morpho fee wrapper ship in September as promised at [31:48]? Where is the card, which the memo said was expected in Q3?
20. Cash in bank, monthly burn, runway, headcount plan and use of proceeds. None of these were stated.
21. Is the Firstrade funding path working today? Blocktrend ep306 (2025-10-21) described it as stuck.
22. Federico Castelli is a registered board member but is named as a founder nowhere. What is his equity and vesting? Does 110 Ventures hold Fluidkey shares?
23. Pulse. Walk us through your actual role, the transaction, and the basis for the $50M figure.

What would make this an IN

A repriced entry at or below $8M, with a Taiwanese counsel opinion in hand, a 2026-11-01 balance that holds above roughly $10M post-campaign with Taiwan cohort retention no worse than the current curve, a written Ondo confirmation on jurisdiction controls, and a lead investor who has done their own regulatory work. On that package a $250,000 cheque is a good risk. Absent the counsel opinion, no price makes it a good risk, because the failure mode is not a slow grind, it is the loss of the only market the company has.

• Whether an unverified account can use Earn, buy tokenized stocks, or use Hide Trail. Pivotal to four risks. Not determinable from public sources. Requires a product walkthrough or a direct written answer.
• An independent AUM figure. The onchain work measured flow, not stock. Enumerating every Safe ever created would allow summing balances, but the log API rate-limited out after more than 8,000 ModuleInitialized events. The $11.3M balance is corroborated only by order-of-magnitude flow.
• The Pulse transaction price. No public figure found (press searches, Ringier corporate disclosures, PitchBook via search snippets; Crunchbase returned 403 and PitchBook is paywalled).
• Antonio's Ianum sale. No acquirer, price or date found anywhere. A small Italian identity-software sale would plausibly go unreported, so this is not a red flag, but Daxos should not carry it as an exit.
• The composition and geography of the non-KYC cohort. No public signal exists.
• Noah's specific licences and registrations. The "licenses in the U.S., Canada and Europe" claim at [26:16] is founder-claimed; no licence numbers were found on Noah's own site or verified against a regulator register. The SSB Bank and Lead Bank allocation by geography is founder-claimed with no public corroboration.
• The Italian company registry for ODF Data, Ianum, tProof, Daolize, Above Tech Inc. and ABOVE S.r.l. requires a paid account. Three of the four founding board members are Italian and their prior ventures were Italian, so this is a real gap. The extracts are inexpensive and should be bought if Daxos proceeds.
• The Swiss Federal Supreme Court judgment search refused connections from this host.
• Android install and rating counts. The Play listing for com.fluidkey returns HTTP 200 but is JavaScript-rendered; only the 1K+ download band was obtainable.
• Taiwan FSC and MOEA registries were not searched for a Fluidkey entity.
• Whether any written legal opinion exists on the Taiwan position, the securities position or the Swiss perimeter. None was offered, and there is no data room.
• Whether the CHF 50,000 unpaid share capital has since been called.
• The Taiwan market-size claims. "$100b+ in US securities held by Taiwanese individuals, growing 20% a year" and "$30b with US offshore brokers": no record found, across US Treasury TIC framing searches, LSEG Lipper April 2026, AmCham Taiwan Business TOPICS September 2026, and IBKR and Firstrade press pages. TIC does not separate individual from institutional holdings and neither broker discloses Taiwan client assets. The wealth distribution claims are attributed to UBS charts that could not be fetched.
• Whether the Morpho fee wrapper shipped in September. The docs still say auto-earn is free. Checkable today by inspecting the vault configuration for a fee recipient.
• A February baseline on Ethereum mainnet, so the onchain flow trend is a single-week snapshot plus one like-for-like Base comparison, not a trend line. Free RPCs archive-gate blocks that old.

What the table says. Fluidkey lands above every house crypto-consumer wallet deal rated this year and below every crypto deal with either a real organisation or real defensibility. That is the right place. Under the tiering heuristic, 5.0 to 5.9 is "pass, or micro check on relationship", which is consistent with the WATCH verdict and with declining the current price.