Daxos Research: Official List, Rated

1. The five line answer

What this list is. Sixteen recently funded companies in agent security, counter-drone, grid and data-center energy, pharma AI and logistics AI, ranked 9.4 down to 6.7 on a conviction scale whose owner is unknown to us. It is a good sourcing list and a bad pricing list: fifteen of the sixteen are real, well-funded, correctly described businesses, and fifteen of the sixteen are closed, led and priced above anything Daxos can write.

How the sixteen hold up. All sixteen resolved to a real company with high identity confidence, but six of the sixteen Harmonic records handed over in the batch were the wrong company and had to be re-resolved, two of those with a match quality of 3, which is to say the exact-name match was confidently wrong. Nobody on the list is a shell or a fabrication. The mean house COMPANY score is 6.25 against a mean list conviction of 8.04.

The biggest gap between the list and the house. The list is higher than the house score on all sixteen names, by a mean of 1.79 points. The worst three are Niv-AI, 8.6 on the list against 4.25 here, a $12M Glilot seed with zero named customers seven months later and headcount flat at six; Outerlimit, 9.1 against 5.58, $16M before a single nameable customer and placeholder lorem ipsum still live on its news page; and Rig Security, the list's number one at 9.4 against 6.25 here, a genuinely good company whose round closed six days before the list was made. The list rewards round size and investor brand. The house formula docks exactly that.

How many claims failed verification. Of the sixteen claimed rounds, five verified clean, ten verified partly, and one, Avallon Labs, is contradicted outright: the list has the date wrong by six to twelve months and the cumulative total at roughly double the filed figure. Four announcement dates are not close dates and the gap is material in each case. Three claims exist nowhere outside the list: Glacian's Samsung Next lead, Avallon's $9.2M total, and Mara's paid European NATO order.

Is anything here buyable. Two names, and both are at the bottom of the list's own ranking by round size. Glacian has $3,090,000 still unsold on a $5,000,000 Reg D offering with a $50,000 stated minimum, at roughly a $10M post. Palma.ai sits at roughly a $7M post with unsold room on its earlier offering. Everything else is closed. The list's top nine names are uninvestable for a fund of this size at any price we would be offered.

2. The rated table

Sorted by house COMPANY score, descending. Ties broken on FOR DAXOS. COMPANY is the house formula, Base + EV + Geo + Chain. FOR DAXOS is the separate fit number and it is kept deliberately apart from the company score. Delta is list conviction minus house COMPANY, so a positive number means the list is more bullish than we are.

Mean COMPANY 6.25. Mean FOR DAXOS 3.12. Mean list conviction 8.04. Verdicts split four MEET, seven WATCH, five PASS, and no IN. Fourteen of sixteen have a verified technical founder. One name, Outerlimit at $16,000,000, breaches the house hard filter of $15M raised and is marked in the table; it was rated anyway at the partner's instruction and it would have been excluded before scoring under Rule 22.

Where the two rankings disagree most

3. Where the list is right, and where it is wrong

What it gets right

The list is good at finding companies. All sixteen exist, all sixteen raised real money from real investors inside the last eight months, and the one-line theses are mostly accurate because they are mostly lifted from the companies' own words. Outerlimit's "zero trust at the agent action layer" is verbatim from outerlimit.com/platform/enforce. Perceptic's "connective tissue between AI tools and pharma data" is near verbatim from perceptic.com. That is not research, but it does mean the descriptions are not wrong.

It is also honest in one specific place, and that deserves credit. On Surtr it writes "$4.8M pre-seed Jun 26 (Dealroom says $3.11M)" instead of hiding the conflict. Harmonic and the dated press both say $4.8M, so Dealroom is the one that is wrong, and the list flagged a discrepancy it could have buried.

On 5U AI the list is actually better than Harmonic. It says "DHL/GEODIS/DSV/Maersk angels", which matches Emerge Capital's own published memo. Harmonic miscodes those people as corporate investors, listing A.P. Moller-Maersk, DHL, GEODIS and CEVA Logistics as companies, and omits DSV entirely. Senior executives investing personally and their employers investing are very different signals, and the list has it right.

Its single best observation is Glacian, where it says "earliest entry" and means it. That is the most useful line on the whole list, and it is ranked tenth.

Finally, on one company the list and the house agree to the decimal: Piq Energy, 7.4 and 7.40. That is coincidence rather than method, but the list had the date to the day and the lead correct, which is more than it managed on four other names.

Where it is wrong, and why it is wrong in one direction

The list is higher than the house score on sixteen out of sixteen. A scoring system that never lands below an independent check is not measuring the same thing. What it appears to measure is round size times investor brand. Rank the sixteen by Harmonic funding and you recover most of the list's ordering: the three $12M-plus names sit at 1, 3 and 6, and the two smallest raises sit at 10 and 16. The house formula pushes in the opposite direction on purpose, because capital already in is dilution already taken and price already set.

Over-rated, worst first.

Niv-AI, 8.6 on the list, 4.25 here, the largest gap in the set. The problem is not the company's idea, which is sound: rack-level GPU power sensing to recover stranded data-center capacity. The problem is that nothing has happened. Harmonic headcount is 6, up from 5 a year ago, with engineering flat at 3 for twelve months, after a $12M seed closed 17 March 2026. No named customer, no named design partner, no counterparty confirmation anywhere. Web traffic flat at 477 visits a month. The list's "sells megawatts" is unsupported by any source. And NVIDIA is attacking the same power-transient problem in silicon, which is the hardest kind of competitor for a sensing startup to outrun. Glilot's own portfolio page confirms the round, which is the strongest single verification in the batch, and the round is all there is.

Outerlimit, 9.1 on the list, 5.58 here. Tony Pepper and Neil Larkins built Egress and sold it, and AlbionVC and Evolution followed them out of it. That is a real signal and it is why the base holds at 5.33. But the list's date is simply wrong. Companies House shows allotments on 4 November 2025, 19 December 2025, 29 December 2025 and 3 June 2026, and the accounts filed to 31 December 2025 already carry £7,638,668 of share premium and £7,582,229 of cash, nine months before the 22 September 2026 announcement. The money did not arrive on the day the list says it did. Against $16M raised there is not one named customer, design partner or pilot in the company's release, on its site, or in any coverage; customers are described as "Fortune 500 and FTSE 100 global brands". One of four product pillars is marked Coming Soon, and unreplaced lorem ipsum placeholder posts are live on the news page. The CTO is a theoretical neuroscience PhD with no prior industry engineering role. At $16M this also breaches the house hard filter.

Rig Security, the list's number one at 9.4, 6.25 here. This is the clearest demonstration of what the list is actually scoring. Everything about Rig checks out. Harmonic confirms $12,000,000, one round, SEED, 2026-09-29, and Axios Pro, SiliconANGLE, SecurityWeek and Calcalist all carry it the same day. Ten Eleven and Brightmind led, CrowdStrike's Falcon Fund is in, Ami Luttwak is an angel. There is one named customer with a named customer-side speaker, which almost nobody else here has. And it is the least actionable name on the list, because the round closed six days before the list was written, at roughly a $60M post, led by two specialist funds and a strategic. A 9.4 that cannot be bought at any price we would see is a 9.4 about somebody else's deal. Rig also told press it has 20 people against Harmonic's 9, and its Fortune 200 accounts are unnamed.

Sable Bio, 6.9 on the list, 4.75 here, and the most uncomfortable finding in the set. The round was announced on 11 February 2026 presenting two co-founders. Companies House shows that on that same day, 11 February 2026, the company filed a TM01 terminating Dr Alex De Giorgio as a director, effective 25 December 2025. De Giorgio was the target-identification domain expert and the founding CEO. He was gone entirely by April 2026. The announcement was therefore true on its face and misleading in substance on the day it was made. Since then headcount has been flat at 14, web traffic has fallen from 3,629 to 1,973 a month, and the company has quietly softened "three top ten global pharma firms" to "Multiple Top-20 Global Pharmaceuticals" on its own site. There is also an earlier round of roughly $1.82M the list omits, which moves the EV tier.

Oktogrid, 7.5 on the list, 5.50 here. Every number the list gives is accurate, and that is the finding. "45+ customers, 1+ GW, 198 MVA unlocked" all come from one place, Elewit's own press release of 3 July 2026, published by the lead investor and supplied by the company. The filed Danish accounts for FY2025 tell a different story: gross profit of DKK 310,082, about USD 48,000, down 79 percent from DKK 1,508,195, against a DKK 7.0M payroll. Revenue is not disclosed at all under the small-company exemption. Long-term debt went from DKK 1.85M to DKK 12.72M during 2025, year-end cash was DKK 3.72M against that DKK 7.0M payroll, and 72 percent of the balance sheet is capitalised development cost. Seven years in, USD 8.9M raised, and the EUR 4M was necessary rather than opportunistic. ABB is simultaneously an investor and a direct competitor through its own TXpert transformer line.

Perceptic, 8.8 on the list, 6.75 here. A good company over-ranked for a structural reason the list does not mention. Tilman Flock came out of Relation Therapeutics and the founder-market fit is close to ideal, CSL is a real named production customer, and Accel led $12M. But the company is selling into pharma against the former employer that owns the accounts, and the list's "targets Phase II attrition" framing appears nowhere in the company's or either investor's material. The geo dock matters here too: UK at +0.25 against US at +1.00 is three quarters of a point, and at $12M the EV bonus is zero.

Beltic, 8.2 on the list, 6.25 here. Three founders with three relevant exits, Norwest leading $7.3M, and nothing sold to anyone. Traction is "partnering with a select group of companies", unnamed, no customer quote, no logo. Meanwhile Cloudflare and the card networks are standardising the layer-one piece of agent identity, which is the part Beltic is selling. The list also garbles an investor: "Oxford SE" is Oxford Seed Fund, the student-run fund at Said Business School where the CEO took an executive diploma, not Oxford Science Enterprises. Those are not comparable institutions and the abbreviation flatters the smaller one. The Form D is stranger: it reports total amount sold of $0 against a first sale date of 17 July 2026, and the CEO is absent from its officer and director list.

Under-rated.

Avallon Labs, dead last on the list at 6.7, eighth here at 6.42, and the best-evidenced company in the set. The list describes it worse than any other name and still ranks it bottom, which is the reverse of its usual error. The claimed date, 29 April 2026, is wrong: the Form D gives a first sale of 12 March 2025, the full $4,662,000 was sold by the 26 August 2025 filing, and the round was announced 6 November 2025. The claimed $9.2M total is contradicted by Harmonic at $5,162,000, which reconciles exactly as the $4,662,000 Form D plus a standard YC cheque of about $500,000. There is one Form D on EDGAR and no press for any further round, and $9.2M is precisely twice $4.6M. Strip the bad numbers away and what is left is the only company here with a real ARR curve: six figures of ARR, a stated tenfold growth since YC Spring 2025, headcount 3 to 12 in twelve months with engineering 0 to 5, and a named customer, Athens Administrators, a real California multiline TPA. It is also at roughly a $14M post, the second lowest in the batch. The founding CTO left after eight months and that is a genuine problem, but on evidence this belongs several places higher than the list puts it.

Piq Energy, twelfth on the list, second here. Two technical founders, a Berkeley-confirmed real bottleneck in interconnection queues, headcount growing 13 to 18 over twelve months, the lead investor's own portfolio page carrying the company, and the date and lead both correct on the list. It is the strongest business in the batch and the list ranks it below Beltic, which has sold nothing.

Glacian, tenth on the list, third here, and first on fit at 7.0. The list's own phrase "earliest entry" is correct and it buried the implication. This is a Penn State spinout with NSF records showing $2,592,318 across eight awards to Wangda Zuo including an I-Corps grant for translating this exact cooling software, a named commercial deployment at the Alerify data center in Harrisburg, and a first paying customer signed 15 September 2026. More to the point for us, it is the only company on the list with an open offering: $3,090,000 unsold of a $5,000,000 Reg D, minimum $50,000, at roughly a $10M post. The list ranked the one purchasable name tenth out of sixteen, which tells you the scale was never built to answer the question Daxos asks.

The structural point

The list and the house formula disagree because they are scoring different things. The list scores the deal's quality as a venture outcome. The house formula scores the deal's quality as a Daxos entry, and it does that by docking capital already raised and by rewarding US domicile. Two names illustrate it cleanly. Rig Security, $12M raised and Israeli, collects 0.00 EV and 0.25 geo, so even a base of 6.00 lands at 6.25. Glacian, $1.91M raised and American, collects 0.75 EV and 1.00 geo, so a base of 5.50 lands at 7.25. The list reads Rig as three points better. On the question of what Daxos can own, Glacian is the only one of the two that exists.

4. The claims audit, consolidated

Every claimed round, date and lead across all sixteen. Status is the verdict on the claim as the list states it, not a judgment on the company.

Claims that exist nowhere outside the list

Three, and one near miss.

Glacian's Samsung Next lead, with Samsung Research America alongside. The round is real and filed to the dollar. The lead is not corroborated anywhere. The Form D does not name investors, Harmonic's investor array is empty, and a full read of Samsung Next's own portfolio page returns zero hits for Glacian. This is the single claim on the list that would most change a decision if it were true, because a Samsung Next lead on an open $5M offering at a $10M post is a different deal from an unled one. Ask the founders directly.

Avallon's "total $9.2M". Contradicted rather than merely unfound. One Form D, no press, and Harmonic at $5,162,000. The number is exactly twice the real round, which is the shape of a transcription error rather than a lie, but it cannot be sourced.

Mara's "paid EU NATO order". Not a round, but the strongest traction claim on the list. It appears in no press release, in none of the nine outlets that covered the round, and in no later coverage. usaspending.gov returns zero prime awards for Mara Solutions. The company's own wording is weaker and different: "early-stage engagement with international military partners in the UK, Czech Republic and Ukraine". A paid European order could exist and be undisclosed, which is normal in defence procurement, but the company itself describes something less than what the list asserts.

The near miss is Mara's earlier round of roughly $750,000, which sits in Harmonic's $7,750,000 total and is reported by nobody. Same for Sable's earlier ~$1.82M and Palma's earlier ~$1.35M. Those are the list's opposite error, omission rather than invention, and all three move the EV tier.

Form D absences against claimed US rounds

A US institutional seed normally leaves a Form D within 15 days of first sale. Where it is missing against a claimed US round, that is worth noting.

• Mara. No Form D six weeks after a $7M round from Khosla, a16z speedrun and Gaingels. EDGAR company browse across registrants beginning "mara" and full-text search for "Mara Solutions" both return nothing, checked 5 Oct 26 with a User-Agent set. The window shut around 10 Sep 26. A real process gap, not evidence of a fake round, which is corroborated many other ways.
• Surtr. No Form D for a claimed $4.8M US round announced 7 July 2026. Robinhood Ventures Fund II's own 10-Q does independently disclose a $250,000 SAFE in Surtr acquired 06/08/2026, so the money is real.
• Piq Energy. No issuer Form D for a $5M US seed. Only the $177,000 Umami SPV.
• Palma.ai. No Form D yet for the announced $1.8M. The earlier $2M offering is filed.
• Beltic. A Form D exists but reports total amount sold of $0, and omits the CEO from its officer list.
• Rig Security and Destro AI. No Form D, and in both cases not yet due.

Filed and exact: Glacian, Fluxnium, Avallon Labs, and Beltic on offering amount.

5. What Daxos could actually buy

Blunt answer: two names, and both sit in the bottom half of the list's own ranking.

Glacian. The only live offering in the set. As of the Form D filed 17 September 2026 there is $3,090,000 unsold on a $5,000,000 Reg D offering, with a stated minimum investment of $50,000, which sits below a Daxos cheque rather than above it. Eight investors are in. Harmonic's dilution estimate puts the post at roughly $10M, which for a Penn State spinout with a licensed IP stack, a named deployment at the Alerify data center in Harrisburg and a first paying customer signed 15 September 2026 is a defensible price. Write $100k to $250k at or near a $10M post, conditional on three answers: who actually led and whether Samsung Next is real, a date by which both founders go full time or a credible full-time engineering hire is funded, and a reference call with Alerify. Both founders currently hold other roles. Zuo is an active Penn State professor with a standing NREL joint appointment and Zhao has been President of P3Global since August 2025. Headcount has been 2 for twelve months.

Palma.ai. Small enough that a cheque is an ordinary allocation. Harmonic puts the post at roughly $7M on the $1.8M pre-seed, and $654,334 was still unsold on the earlier $2M offering as of June. Plug and Play and Deel participating rather than leading is the shape of a round built to take several small cheques. $100k to $500k at or below an $8M post. The case is the founder: Patrick Eden built governance plus cost once, at Replex on Kubernetes, and sold it to Cisco, and is now rebuilding the same thesis on agents, with a genuinely sharp wedge in scanning the arguments of tool calls rather than just the permissions around them. Before wiring, get one named customer that is not an investor, the Form D for the $1.8M, a straight answer on the undisclosed pivot from a ChatGPT prompt-anonymizer browser extension, and a view on what happens when Microsoft ships argument-level policy inside Entra.

Everything else: nothing, at any price we would be offered. The estimated posts, from Harmonic's standard dilution model, run $13M (5U AI), $14M (Avallon), $19M (Sable), $24M (Surtr), $25M (Piq), $28M (Mara), $35M (Fluxnium), $37M (Beltic), $40M (Destro), $60M (Rig, Perceptic, Niv-AI) and $64M (Outerlimit). Oktogrid was led and priced by a corporate strategic. In each case the round is closed, a named institution led it and will hold its pro rata, and a $100k to $500k ticket is below the size that gets carved out.

Four of those have a next window worth building a relationship for now, which is what the MEET verdicts mean:

• Mara. Thirteen people burning $7.75M on pre-production hardware with a stated goal of ground-mounted deployment to key customers by end of 2026. A seed or Series A inside 9 to 15 months is likely. Price to want: at or under roughly $60M pre on evidence of a first paid order. Take the meeting to be on the list, and ask directly whether a paid European order exists.
• Piq Energy. The strongest business here. A Series A is likely within 12 to 18 months in a market that is heating up, and a relationship built now is the only route a fund this size gets an allocation into a contested A. Press for three callable customers, the ARR, whether Cisco Foundation bought equity or made a grant, and one interconnection timeline measured before and after Piq.
• Avallon Labs. The only real ARR curve in the set, at the second-lowest post. A YC company at six figures of ARR growing tenfold raises an A in 2026 or 2027, and around those there are usually insider extensions and SAFE top-ups at Daxos size. Resolve the ARR number rather than the band, whether Athens Administrators will serve as a reference, and who owns engineering now the founding CTO has gone.
• Surtr. Good founders, real BRAVE1 field validation in Ukraine. The right entry is the next round, after radar and RF are actually integrated, ITAR is granted rather than pending, and there is a signed contract or OTA rather than a battle-lab approval.

Two to avoid even if a window opened. Fluxnium is commodity science risk where the next milestone is a pilot funded by strategics and infrastructure capital, not angels, with revenue most likely seven to ten years out and published lab economics for seawater uranium at roughly 2.5x to 4x the current market price. Niv-AI is wrong on price and wrong on evidence at the same time: roughly $60M post against zero verified traction.

6. Red flags, company by company

Mara. No Form D six weeks after a $7M US institutional round. Harmonic's $7,750,000 total implies an undisclosed earlier round of about $750k that nobody reports. The paid EU NATO order appears nowhere and is contradicted in framing by the company's own "early-stage engagement" wording. The list's 300g interceptor is wrong; the company says 250g. Neither founder has any defence background. They spent six years at Optivolt and left in March 2024 with no exit, and Optivolt has not raised since February 2023.

Piq Energy. No issuer Form D for a claimed $5M US seed; the only EDGAR filing is a $177,000 Umami SPV that closed in November 2025, eight months before the round was announced as oversubscribed. Not one customer is named anywhere and no revenue figure exists. The 1,000 grid models and 10,000 workflows are internal platform counters the company fully controls. Stamatiadis held a full-time role at ESB Networks until August 2024, overlapping his September 2023 co-founder start date. Cisco Foundation participating in a priced seed is unusual for a foundation and may be a grant recorded as participation.

Glacian. Both founders hold current concurrent roles and neither looks full time. Headcount has been 2 for twelve months. The claimed lead investor appears nowhere outside the list. The homepage leads with 53 percent and 74 percent energy savings from DOE research pilots while the company's own commercial claim is 15 to 30 percent, and the CEO volunteers that the 74 percent came partly from finding a setup error that had been there for years. The core modelling base is open source.

Destro AI. No Yusen-side press release for the named customer. Uncited ROI statistics on its own site. The founding-date story is inconsistent across sources. No Form D yet, though not yet due; check after 14 October 2026.

Perceptic. No share allotment filed for the claimed 26 May 2026 round date; the last filed allotment is 11 March 2026 and no US entity exists on EDGAR, so the UK registry is the whole picture. Accounts to 31 December 2025 show only £3.2M share premium and 4 average employees. Lead-investor ambiguity: Air Street names itself first and omits Elder Gull. The list's Phase II attrition framing appears in no company or investor material. No CSL-side confirmation found. Every product metric is company-reported.

Palma.ai. Undisclosed pivot from a ChatGPT prompt-anonymizer browser extension, visible in Harmonic's name-alias history and confirmed on Wayback. A blog headline frames Silver membership of the Linux Foundation's Agentic AI Foundation as joining OpenAI, Anthropic and Microsoft, when the Foundation's own release lists those names among 97 new members. The San Francisco address is a commercial mail receiving agency while the whole team sits in Germany. No Form D yet for the announced $1.8M. Both named users, D11Z and Plug and Play, are investors.

Surtr. No Form D for a claimed $4.8M US round. No ITAR registration yet, by the company's own statement. Verified sensor fusion is acoustic plus optical only, not the radar and RF the list claims; BRAVE1's own write-up says "data from acoustic and optical sensors" and calls the product prototype-stage. Four people. A duplicate Volkswagen Financial Services Portugal role on Fiore's Harmonic record with dates identical to his Anduril role looks like a parsing artifact rather than a founder misstatement.

Avallon Labs. The list's date is wrong by six to twelve months and its $9.2M total is roughly double the verifiable figure. The founding CTO left after eight months. Four different co-founder start dates appear inside the first year. ARR is given only as a band and growth only as a multiple off an undisclosed base. The navigation advertises a Case Studies page that 404s. Web traffic fell from 5,809 to 3,259 in a quarter. No career claims operator on the founding team.

Beltic. The Form D reports total amount sold of $0 against a first sale date of 17 July 2026. The CEO is absent from the Form D officer and director list. The CTO is absent from Harmonic entirely. No Form D for the $1.5M pre-seed. Headcount is 1 in Harmonic and 2 externally, against at least 5 identifiable named people. "Oxford SE" overstates Oxford Seed Fund. No verified traction of any kind.

Rig Security. Fortune 200 customers unnamed. The 96 percent RICE accuracy figure is self-reported. The CEO's prior "CU (Acquired)" exit has no public trace. Headcount given as 20 to press against 9 in Harmonic. A co-founder CTO and a separately hired CTO are both listed as current with no public explanation.

Fluxnium. The national laboratory behind the adsorption fiber is never named by the company, the lead investor or any outlet, and the characterisation drifts between "licensed from" and "developed with". Published lab economics for seawater uranium run roughly 2.5x to 4x the current market price while the company projects parity with mining. No pilot, no site, no timeline, no uranium-recovered figure. The round closed 5 May 2026 and was presented as news on 21 September 2026. No technical or scientific founder is named.

5U AI. No revenue or customer count has ever been published, by the company or by the lead investor. The one named customer, TCI International Logistics, does not mention 5U on its own site. The site carries unsubstantiated "100% Task Coverage", "15x ROI" and "10x" banners. The lead investor's memo inflates a student thesis into having "learned the industry from inside CEVA Logistics". First-time CEO with no senior logistics operator employed. Only three or four technical staff. The operating entity is a renamed shelf company about four months old.

Outerlimit. The money did not arrive on 22 September 2026; the balance sheet carried £7.58M of cash at 31 December 2025. Customers are described only as "Fortune 500 and FTSE 100 global brands", none named. The "world's first" superlative is unfalsifiable. One of four product pillars is marked Coming Soon. Unreplaced lorem ipsum placeholder posts are live on the news page of a company with $16M. The CTO has no prior industry engineering role. Breaches the $15M hard filter.

Oktogrid. Filed FY2025 accounts show gross profit of DKK 310,082, about USD 48,000, down 79 percent year on year, against a DKK 7.0M payroll. Revenue is not disclosed at all under the small-company exemption. 72 percent of the balance sheet is capitalised development cost, DKK 19.6M of DKK 27.2M. Long-term debt rose from DKK 1.85M to DKK 12.72M during 2025. Year-end cash was DKK 3.72M against that payroll. Interim CFO in post since March 2024. Seven years old and still classified seed. ABB is both an investor and a direct competitor through its TXpert line.

Sable Bio. The founding CEO's resignation, effective 25 December 2025, was filed at Companies House on 11 February 2026, the same day the round was announced, and the announcement still presented him as a co-founder. He left entirely by April 2026, and he was the target-identification domain expert. No share allotment was filed on or near the publicised date; the last SH01 is for 10 September 2025. The customer claim has been quietly downgraded from top-ten to Top-20 on the company's own site. Web traffic is falling. Harmonic's legal name for the company does not exist at Companies House.

Niv-AI. Headcount flat for twelve months after a $12M raise, with engineering flat at 3. The press-claimed 10 employees is contradicted by Harmonic's 6. The promised US deployment window has elapsed with no public evidence. "Sells megawatts" is unsupported by any source. NVIDIA is attacking the same power-transient problem in silicon. No named customer, no named design partner, no counterparty confirmation anywhere. Web traffic flat at 477 visits a month.

7. Method and limits

What was run. Each of the sixteen names was re-identified first, then rated, then its claims audited. Identity came from Harmonic enrichment by domain, Harmonic keyword search when the domain was unknown, and primary registries. Six of the sixteen Harmonic records supplied in the batch were the wrong company: Mara pointed at a parked Spanish domain, Destro at an unrelated shell, Beltic at a two-person Cordoba consultancy, Glacian at an Indian bottled-water brand, Avallon at a New York consultancy with a colliding name, and Niv-AI had no match at all. Two of those six carried a match quality of 3, meaning an exact name string match that was confidently wrong. All sixteen were then resolved to high identity confidence against a filing or a registry.

Scoring. Final = Base + EV + Geo + Chain, per house Rule 22, with base anchored on the Daxos benchmark distribution. EV is the 5-tier funding bonus of Rule 21. Geo is additive. No company here is Berachain-primary, so Chain is 0.00 throughout. FOR DAXOS is a separate fit number and is not part of the formula. One name, Outerlimit at $16,000,000, breaches the $15M hard filter and would normally have been excluded before scoring; it was rated at the partner's instruction and is marked in the table.

Sources used. Harmonic API for all funding, stage, headcount, investor and round-count figures, pulled 5 October 2026. SEC EDGAR full text and browse-edgar with a User-Agent set. UK Companies House. Danish CVR. German commercial register via Northdata. Israeli Registrar of Companies via data.gov.il. NSF award API. usaspending.gov and SBIR for the defence and energy claims. Wayback. Lead investors' own portfolio pages, which turned out to be the single most useful confirmation route: Glilot, Active Impact, Congruent and Samsung Next were all checked directly, and the one absence found there, Glacian on Samsung Next, is a real finding.

Limits, stated plainly.

Harmonic was independently contradicted in at least six places in this set. It had Glacian's total at $2,060,000 when the Form D shows $1,910,000 including $150,000 of converted SAFEs. It carries Convective as a Piq investor when the company's own release does not. It miscodes the 5U AI angels as corporate investors and omits DSV. It gives Sable Bio a legal name that does not exist at Companies House. It records only two Perceptic founders where Companies House shows three directors and three persons with significant control appointed on the day of incorporation. Its Oktogrid headcount of 18 sits against an audited average of 10 in the filed accounts, and its address for the company is stale. Where Harmonic and a filing disagree, the filing is used and both are reported.

Posts are estimates, not facts. Every post-money figure in section 5 comes from Harmonic's standard-dilution model applied to a disclosed round size. None is a confirmed valuation. Treat them as a range and nothing more.

Almost no revenue here is verified. Oktogrid's gross profit and Avallon's exact round size are the only hard financial numbers in the set, and the former is a filed figure while the latter is a filing on capital rather than revenue. Sable's $0.5M and 5U's "paying customers since late 2025" are company statements. Everything else is a platform counter the company controls or an unnamed logo.

Announcement dates are not closing dates, and on this list the difference is routinely four to twelve months. Four were material enough to mark as contradicted. If a date matters to a decision, pull the filing.

CourtListener could not be queried. The anonymous API returned a daily rate-limit error on 5 October 2026, so litigation dockets are unchecked across all sixteen. No adverse litigation surfaced in press or search for any of them, which is weaker evidence than a docket search would have been.

Three things could not be settled at all and are named as open rather than resolved: Glacian's lead investor, the identity of Fluxnium's national laboratory, and whether Mara's paid European order exists.

Built 5 October 2026 from 16 deep company cards, the original 16-entry claim list, the Harmonic API, and the filings named above. All funding data from Harmonic or a primary filing. Internal and confidential.