Prague Technology

1.1 Identity and operating context

Prague Technology is described in the runtime brief as a Shanghai-based world-model and embodied-AI startup, but the assigned successful sources do not contain a direct company profile, official website, filing, or customer disclosure for that entity. The available independent material instead places Junyang Lin in the Qwen ecosystem and describes a transition from model training toward agentic systems. Marktechpost reports that Lin was Alibaba's Qwen technical lead, announced his departure on March 3, 2026, and now presents himself as an independent researcher. That is relevant founder-context evidence, not proof of Prague Technology's incorporation, headquarters, product release, or commercial status. A direct identity source remains necessary before later chapters reuse these fields as ground truth.[CO001, CO002, CO003, CO004]

1.2 Leadership and governance

The evidence pool does not identify Prague Technology's full founding team, executives, directors, or governance arrangements. The strongest available leadership evidence concerns Lin's prior technical role rather than a current company appointment: Marktechpost says he was the technical lead of Alibaba's Qwen project and that he now lists himself as an independent researcher. The source also describes a technical agenda spanning reasoning, tool use, environments, and agents, which helps explain potential founder-market fit but cannot establish board control, employment, or key-person concentration inside Prague Technology. Leadership diligence should obtain incorporation records, an authoritative team page, cap-table or shareholder disclosures, and confirmation of any continuing Alibaba relationship.[CO005, CO006, CO007]

1.3 Financing, scale, and milestone evidence

No assigned source verifies Prague Technology's funding round, valuation, revenue, customers, headcount, debt, or cash runway. Forbes documents a broader 2026 pattern in which AI companies can publicize headline valuations while raising in tranches at materially different prices; it reports that the blended valuation is often unreported. That source supports a diligence warning about interpreting any unverified headline valuation, not a Prague Technology financing fact. The China embodied-AI sector source reports 3,025 existing enterprises as of May 2026 and 408 new companies in 2025, while also reporting rapid sales growth; these are market-context metrics and must not be attributed to Prague Technology. The milestone timeline is consequently evidence-limited and should be expanded only with direct company or filing evidence.[CO008, CO009, CO010, CO011, CO012]

1.4 Exhibits

2.1 Market boundary and substitutes

This chapter treats the relevant market as software, models, data and deployment services that enable AI systems to perceive, reason, and act through physical machines, with industrial and research buyers as the primary demand base. It excludes general-purpose cloud AI, consumer robotics sold without an embodied-model or deployment layer, and the full revenue of hardware suppliers unless that revenue is specifically tied to embodied-AI systems. The boundary matters because the assigned evidence mixes model infrastructure with robot bodies and system integration: the State Taxation Administration data reported by China Daily separately identify robot bodies and complete machines, AI algorithms and software integration, and system integration and industry applications. A credible Prague Technology market case should therefore not add those categories into one undifferentiated TAM. An independent-reporting query also returned Prague City Tourism rather than evidence about the startup. That naming collision is a source-discovery false positive, and place-based tourism is explicitly outside the market boundary. Bridging China reports that Lin Junyang's venture is focused on world models and embodied intelligence, but the assigned evidence does not establish which product or deployment layer it will commercialize. The practical substitutes are conventional industrial automation, scripted perception and control stacks, human-operated teleoperation, and general language-model APIs that do not close the loop with a physical environment. The market is strategically attractive because agentic systems need tool servers, sandboxes, environment feedback, and train-serve decoupling, but the company-specific evidence does not establish which layer Prague Technology will monetize. [CM001, CM002, CM003]

2.2 Evidence-constrained sizing lenses

The strongest quantitative lens is China embodied AI rather than a global world-model TAM. China Daily reports 3,025 existing China embodied-AI enterprises as of May 2026, 408 new companies in 2025, and January-May 2026 sector sales growth of 22.4% year over year. It also reports category growth of 30.1% for robot bodies and complete machines, 24.5% for AI algorithms and software integration, and 27.9% for system integration and industry applications. These are reported sales-growth indicators, not a market-size total and not Prague Technology revenue. A second lens is the funding market: Bridging China reports a roughly $2 billion post-money valuation for Lin Junyang's new AI lab; that is a financing signal, not demand or cash available to the company. A third lens is comparable technical supply: The AI Rankings describes Qwen 3.6 open weights as Apache-2.0 and self-hostable, with cited hosted prices of approximately $0.60 input and $3.60 output per million tokens for the 27B model. Open weights and low API prices can compress model-layer pricing while increasing the value of proprietary environments, integration, data, and reliability. No assigned source isolates Prague Technology's obtainable SAM or SOM, so any such figure remains a diligence gap. [CM004, CM005, CM006, CM007, CM016]

2.3 Buyers, adoption path, and constraints

The evidence points to industrial enterprises, system integrators, research organizations, and strategic technology investors as distinct buying or financing audiences. China Daily reports that purchases of embodied-AI robots by industrial enterprises rose 2.3-fold in January-May 2026 and that supporting information-system services revenue rose 1.9-fold, which supports an adoption path from pilot hardware and model evaluation to deployment, integration, and maintenance budgets. Gasgoo's report on TARS provides a concrete adjacent-company pattern: a large financing round was paired with robot hardware, a data-acquisition suite, a model, and industrial assembly milestones. That is not evidence that Prague Technology has equivalent products, but it shows the buyer problem is typically a system sale rather than a standalone model sale. The main constraints are proof of task success, integration with existing factory workflows, compute and data intensity, safety and reliability, switching costs, and the risk that open models make the base capability interchangeable. Forbes' reporting on staged AI financing also cautions that headline valuations can diverge from blended economics, so capital-market enthusiasm should not be used as a proxy for product-market fit. [CM008, CM009, CM010, CM011, CM012]

2.4 Market diligence gaps and downside cases

The market is growing, but the evidence does not resolve whether Prague Technology owns a differentiated technical wedge or can reach paying users. Lin's published discussion, as reproduced by MarkTechPost, describes agentic thinking as sustained action in an environment and identifies tool choice, noisy observations, plan revision, and multi-turn coherence as requirements. It also identifies reward hacking, tool-server quality, and train-serve decoupling as infrastructure risks. Those constraints make deployment quality and environment access potential moats, while simultaneously raising implementation cost and failure risk. Qstheory's concentration data indicate nearly 90% of China's embodied-AI enterprises are concentrated in Guangdong, Beijing, Shanghai, Zhejiang, and Jiangsu; this may help ecosystem access but also implies dense competition for talent, capital, and industrial partners. The assigned pool does not provide verified Prague Technology customer commitments, product pricing, unit economics, deployment reliability, regulatory approvals, or a company-specific market share path. Those are material unresolved questions rather than assumptions to fill with the reported sector growth rates. The tourism result and the inaccessible China Daily allocation also show why search-result relevance and readable source text cannot be assumed merely from a successful HTTP response. [CM002, CM013, CM014, CM015]

2.5 Exhibits

3.1 Competitive landscape and buyer alternatives

The company’s stated direction is associated with world models and embodied intelligence, but the assigned evidence does not provide a public product specification, customer segment, or pricing model for Prague Technology. That makes the most defensible landscape one of overlapping alternatives rather than a clean peer set. Alibaba’s Qwen family represents a scaled, open-weight model platform with self-hosting and multimodal inputs; TARS represents an embodied-intelligence company combining models, data capture, and robots; and frontier AI labs such as the company described by Forbes compete for the same scarce research talent and capital even when their end products differ. The assigned pool therefore supports ecosystem context but not a verified count of all potential substitutes. One executed company-news query returned a Prague travel guide rather than company evidence, showing that namesake noise also limits exhaustive discovery. Status-quo alternatives include extending an incumbent model stack, buying hosted APIs, self-hosting open weights, or building an internal robotics/model team. Prague Technology’s direct comparison therefore remains an evidence gap until its first model, API, robot integration, or customer workflow is disclosed.[CP001, CP003, CP004, CP005, CP015]

3.2 Capability, distribution, and trust comparison

The strongest observable competitive distinction in the evidence is distribution and deployment posture, not benchmark superiority. Qwen 3.6 is described as Apache-2.0 open-weight software that can be downloaded, self-hosted, and used commercially, with text, image, and video inputs; its hosted pricing is reported separately from the free-weight option. TARS is described as a vertically integrated embodied platform with A-series wheeled robots, T-series bipedal robots, a data-acquisition suite, and an open dataset, while Gasgoo also reports purchase intentions from European clients without disclosing signed contracts or revenue. These are materially different buying propositions: software adoption, hosted inference, and physical-system deployment carry different integration, safety, and support burdens. Lin’s discussion of agentic thinking adds a technical constraint relevant to future entrants: agentic systems depend on tool servers, sandboxes, environment feedback, and reward-hacking resistance, rather than only a model checkpoint. Prague Technology cannot yet be scored on these dimensions because the pool contains no official technical documentation, pricing page, security posture, customer reference, or independent benchmark for it.[CP006, CP007, CP008, CP009, CP010]

3.3 Moat durability and displacement risks

The available evidence points to a difficult moat test. Open-weight releases can reduce software switching costs through self-hosting and permissive licensing, while vertical embodied competitors can bind customers through hardware, data collection, deployment, and maintenance. TARS’s reported financing and product breadth indicate access to substantial capital and an industrial ecosystem, but the source is a trade publication and does not independently verify all performance claims. Forbes provides an adverse market signal: headline valuations for pre-product AI labs can be shaped by tranched financing and may not equal a blended economic valuation, while employees and later investors can bear different risk. That does not establish a problem at Prague Technology, but it makes financing headline comparisons unreliable as a moat proxy. A durable position would need evidence of proprietary data, model performance on a defined workload, distribution or partner access, customer retention, safety certification, or a workflow that is costly to replace. None is documented in the assigned pool, so displacement and commoditization risk remain material diligence questions.[CP011, CP012, CP013, CP014]

3.4 Competitive diligence priorities

The next competitive diligence step is to force comparability around one buyer job and one deployment unit. Obtain Prague Technology’s first public product description, target buyer, pricing or contract model, reference deployment, model or robot capability metrics, and compute and data dependencies. Then compare it with open-weight self-hosting, hosted model APIs, incumbent enterprise suppliers, and embodied platforms on total cost, latency, integration effort, safety obligations, and switching cost. The current evidence can establish that alternatives are active and well funded, but it cannot establish Prague Technology’s relative price-performance, win rate, or customer pull. Any investment conclusion that treats a reported financing round, a benchmark score from another model, or a competitor’s product announcement as proof of Prague Technology differentiation would exceed the source record.[CP003, CP008, CP010, CP013, CP014]

3.5 Exhibits

4.1 Revenue model remains unproven

Public evidence does not establish a current revenue stream, ARR, customer count, realized pricing, or revenue-recognition policy for Prague Technology. The available financing coverage describes the company as a new AI laboratory focused on world models and embodied intelligence, but does not identify a paid product, contract structure, or recurring customer obligation. Industry evidence shows several possible monetization layers rather than one validated model: model access, enterprise integration, deployment and maintenance, and potentially physical systems or partner economics. Those layers have different gross-margin and working-capital profiles. The prudent underwriting position is therefore “pre-revenue or undisclosed,” not zero revenue, and each scenario must keep company claims, third-party reports, and analyst estimates separate.[CI001, CI002, CI003, CI004]

4.2 Commercial path is tied to expensive deployment

The most credible external market evidence points to a deployment-led ecosystem in which software, system integration, deployment, and maintenance are economically distinct. A State Taxation Administration data summary reported that January–May 2026 embodied-AI sales rose 22.4% year over year, while supporting information-system services revenue rose 1.9-fold; this supports a market mechanism, not Prague Technology traction. A separate WAIC account says the industry is moving from demonstrations toward production but that operational uptime and total cost of deployment were not disclosed for highlighted deployments. For Prague Technology, sales-cycle length, compute pass-through, integration labor, support burden, and customer concentration remain open. Until a customer contract or deployment cohort is disclosed, a high-value model API or enterprise license cannot be assumed to have software-like margins.[CI005, CI006, CI007, CI008]

4.3 Financing is a runway input, not proof of cash available

A Sina Finance report, relaying The Information, described a first financing of “数亿美元” and an approximately $2 billion post-money valuation, including reported $100 million investments from GaoRong and Sequoia China and a $20 million Tencent investment. Those are reported transaction terms, not a verified cash balance, unrestricted cash amount, or runway. The company’s compute strategy may require substantial recurring GPU, data, and engineering expenditure before product revenue is established. The Silicon Review also described proposed Chinese restrictions affecting advanced models, training data, algorithms, and semiconductor partnerships; because the measures were still under review in that article, the financial implication is scenario risk rather than a booked cost. Cash, burn, debt, committed compute, tranche conditions, and next-round triggers remain material diligence requests.[CI009, CI010, CI011, CI012]

4.4 Financial verdict and diligence blockers

The financial case is investable only as a high-uncertainty option on future commercialization. Positive evidence includes reported access to substantial initial capital and a market with growing activity in embodied-AI sales and related services. Counterweights are more important for underwriting: no public revenue or margin proof, no disclosed customer cohort, no disclosed cash or burn, and exposure to compute and cross-border technology constraints. The reported valuation should not be translated into a revenue multiple because the denominator is unavailable. A financing headline also should not be treated as cash available without confirmation of tranche terms, closing status, restrictions, and post-close liquidity. The recommendation from this chapter is research-more: request a monthly management pack, capitalization table, signed commercial pipeline, compute commitments, and a bridge from customer activity to gross profit.[CI002, CI008, CI009, CI011, CI012, CI013, CI014]

4.5 Exhibits

5.1 Product definition and maturity

The available company description characterizes Prague Technology as a world-model and embodied-AI startup, but the assigned successful fetches do not include a company homepage, product page, API, model card, repository, benchmark, customer deployment, or technical paper attributable to Prague Technology. That is a material diligence limitation rather than evidence that no product exists. Public comparators illustrate the range of what a mature technical disclosure looks like: Google DeepMind describes Genie 2 as a foundation world model that generates action-controllable, playable 3D environments from a single prompt image, while OpenAI's Sora system card specifies multimodal inputs, output limits, a diffusion/transformer design, training-data categories, evaluations, and deployment mitigations. Prague's customer workflow, product surface, supported modalities, model access, and measurable benefit cannot be verified against that baseline.[CE001, CE002, CE003, CE004]

5.2 Architecture, operating model, and differentiation

The technical evidence points to a demanding architecture problem. Genie 2 describes autoregressive latent diffusion over video-derived latent frames, a transformer dynamics model, action conditioning, and a need to maintain world consistency across time. The embodied-intelligence survey frames physical simulators and world models as complementary components: simulation provides controlled training and evaluation, while world models support predictive planning and adaptation. Junyang Lin's discussion of agentic systems adds an operating-model constraint: the model is trained and evaluated with its environment or harness, including tools, sandboxes, feedback, and safeguards against reward hacking. None of the assigned sources establishes that Prague has built any particular stack, data pipeline, simulator, inference service, or proprietary feedback loop. Differentiation should consequently be underwritten only after primary technical materials and reproducible evaluations are obtained.[CE005, CE006, CE007, CE008]

5.3 Deployment, reliability, and trust controls

Product risk is elevated by the gap between compelling demonstrations and robust deployment. The survey identifies the sim-to-real and generalization problem as a central challenge for embodied intelligence. Sora's system card provides a concrete comparator for controls that a production generative system may need: external red teaming, content filtering, age gating, restrictions on likeness uploads, watermarking, reverse-search tooling, and iterative monitoring. These controls are not evidence that Prague faces the same exact misuse cases, but they show the level of operational detail required to assess a foundation model product. Public evidence in the assigned pool does not disclose Prague's reliability targets, incident history, privacy posture, security testing, export-control operating plan, evaluation suite, or support model. The investment case should therefore treat technical maturity and deployment readiness as open questions, not as inferred strengths from founder pedigree or sector momentum.[CE009, CE010, CE011, CE012]

5.4 Exhibits

6.1 Buyer and user segments

Public reporting describes embodied AI as serving industrial production, software and information-technology services, scientific research, robotics, and prospective healthcare or industrial applications. China-wide data show that industrial enterprises increased purchases of embodied-AI robots 2.3-fold in January–May 2026, while supporting information-system services revenue rose 1.9-fold; those are sector observations, not Prague Technology results. World-model coverage similarly describes the category as potentially useful for robotics, delivery, healthcare, industrial process control, and wearables, but does not name Prague Technology as a supplier or customer-facing vendor. The investable segmentation therefore remains a hypothesis: industrial and robotics buyers may be the eventual payer, technical teams may be the user, and integration or deployment partners may be the channel. No public source establishes which segment Prague Technology has converted. [CU001, CU002, CU003, CU004]

6.2 Adoption trajectory and durability

The available evidence supports category demand, not company adoption. State Taxation Administration data reported by China Daily show 3,025 embodied-AI-related enterprises as of May 2026, 22.4% year-over-year sector sales growth in January–May, and 2.3-fold growth in industrial-enterprise robot purchases. Those denominators cover the sector and cannot be converted into Prague Technology accounts, active users, utilization, revenue, retention, or renewal. Independent world-model reporting also warns that research-first companies can take years to reach commercial applications; TechCrunch quotes AMI Labs' CEO saying world models are not a typical startup that releases in three months and reaches revenue in six months. Prague Technology has no public cohort, churn, NRR, GRR, contract-length, or customer-count disclosure in this pool.[CU005, CU006, CU007, CU008]

6.3 Named proof, expansion, and concentration

No assigned source names a Prague Technology customer, describes a Prague Technology pilot, or reports a production deployment, outcome, renewal, expansion, or concentration statistic. Comparable evidence shows what stronger proof would look like: Gasgoo reports that TARS's A1 robot received clear purchase intentions from clients across industries and European countries, while TechCrunch identifies Nabla as AMI Labs' first disclosed partner and says early models would be evaluated with real-world data. These are competitor examples and cannot be attributed to Prague Technology. The current customer case is consequently asymmetric: the market may be forming, but there is no company-specific proof of conversion or durability. Management should provide named references, signed contract status, deployment dates, usage or outcome metrics, and revenue concentration. [CU009, CU010, CU011, CU012]

6.4 Commercial diligence implications

The strongest positive signal is an expanding Chinese embodied-AI ecosystem with measurable industrial purchasing activity. The strongest negative signal is evidentiary: the assigned pool contains no Prague Technology customer proof, and the company-specific customer funnel cannot be separated into prospect, pilot, production, paid deployment, repeat purchase, or expansion. Competitive world-model coverage also indicates long commercialization timelines and high dependence on real-world evaluation. Until management provides a customer schedule and primary evidence, customer quality should not support a revenue estimate, a retention assumption, or a valuation premium. The key diligence test is whether at least three references can confirm paid production use, measurable outcomes, renewal or expansion intent, and the share of revenue represented by the largest accounts. [CU013, CU014, CU015]

6.5 Exhibits

7.1 Risk ranking and thesis-break triggers

The highest-consequence risk is not a documented incident at Prague Technology; it is the absence of public evidence that a pre-product world-model company can convert founder credentials and sector momentum into repeatable deployments. Sector evidence shows both intense capital formation and an early-stage competitive field. PanDaily reports 503 China embodied-intelligence funding rounds totaling more than RMB 96 billion between July 2025 and June 2026, while also noting that even a highly patented competitor had only 20 core invention patents out of 262. The combination raises the bar for technical moat, data access, and customer proof. A thesis break would be a failure to demonstrate a production deployment with measurable outcomes, or a financing event that leaves compute capacity and runway materially below the plan. Monitoring should therefore emphasize signed deployments, repeat usage, model-evaluation deltas, cash runway, and funding terms rather than headline round size.[CR001, CR002, CR003, CR004]

7.2 Legal, regulatory, and governance exposure

Publicly available evidence in the assigned pool does not confirm Prague Technology's licenses, litigation status, privacy controls, export-control permissions, or incident history. That is a material diligence gap, not evidence that no issue exists. The company-registry URL in the pool resolves to a login page, so ownership, registered capital, legal representatives, and court history remain unverified from that source. The sector context also creates cross-border compliance exposure: an embodied-AI platform may combine model weights, robotics hardware, customer operational data, and cloud or GPU dependencies. Before investment, counsel should verify the Chinese entity and affiliates, IP assignments, data-processing basis, model and robotics safety obligations, export-control screening, and any restrictions affecting overseas customers or suppliers. A thesis break is a regulator, court, or counterparty finding that blocks core data, compute, or deployment access.[CR005, CR006, CR007]

7.3 Operational, technical, partner, and financial dependencies

The operating model is exposed to three linked dependencies. First, agentic systems need reliable environments and tool access: the reviewed technical discussion identifies reward hacking as a central risk once models gain real tool access and says that training and inference must be cleanly decoupled to avoid rollout bottlenecks. Second, large incumbents are entering embodied intelligence as infrastructure providers, and automotive supply chains can transfer sensors, actuators, decision systems, and data pipelines into robotics. This can accelerate the market while compressing differentiation. Third, early-stage AI financing can obscure effective valuation and employee or investor downside: Forbes describes tranched rounds in which the headline valuation can differ materially from the blended price. Prague Technology's public financing, runway, customer concentration, supplier commitments, and model-cost profile are not established by this pool. The practical kill criteria are failed safety or reliability evaluations, inability to secure compute on acceptable terms, loss of a critical partner, or a financing requirement before repeatable commercial evidence.[CR008, CR009, CR010, CR011, CR012]

7.4 Required monitoring and diligence actions

The risk posture should remain high until management supplies primary evidence. The minimum package is a current cap table and financing instruments; a 24-month cash, compute, and hiring plan; named pilot and production customers with outcome metrics; a model and robotics safety case; incident and vulnerability disclosures; IP ownership records; and a map of cloud, GPU, data, manufacturing, and distribution dependencies. External monitoring should track new entrants and financing intensity, benchmark reproducibility, regulator or court records, security incidents, export-control changes, and any down-round or tranche disclosures. Because sector valuation sentiment can reverse quickly, an investment committee should not treat any headline valuation or funding total as cash available or fair value without primary documentation. The decision should be revisited only when the gaps are closed with dated, attributable evidence.[CR002, CR006, CR010, CR012]

7.5 Exhibits

8.1 Recommendation and entry discipline

Prague Technology should remain a track rather than a buy on the evidence available for this refresh. The company is associated with a high-credential world-model and embodied-AI thesis, but the pool does not provide a directly usable financing announcement, term sheet, audited financial statement, or cap-table disclosure for Prague Technology itself. That makes a price-sensitive recommendation impossible without false precision. The market context is supportive: China Daily reporting based on State Taxation Administration data says embodied-AI enterprise sales rose 22.4% year on year in January-May 2026 and industrial purchases of embodied-AI robots increased 2.3-fold, while the same report says 3,025 related enterprises existed as of May 2026. Those figures support market opportunity, not Prague Technology's share, margins, or conversion into cash flows. Marktechpost's account of Lin Junyang's transition from Qwen technical leadership toward agentic systems supports founder credibility and strategic relevance, but not enterprise value. The appropriate entry discipline is to require verified round size, price per share, preference stack, ownership, runway, product milestones, and evidence of paid deployment before underwriting a valuation or target return.[CV001, CV002, CV003, CV004]

8.2 Thesis, anti-thesis, and financing overhang

The investment thesis has two evidence-backed pillars and three material counters. First, the sector is expanding, with reported growth in Chinese embodied-AI sales and industrial adoption. Second, the founder's prior Qwen work is connected to a credible shift from static model reasoning toward agentic systems, including tool use and environment feedback. The anti-thesis is that market growth can coexist with severe selection and financing risk: Gasgoo reports a $455 million Pre-A round for competitor TARS, and says the capital is being used to develop proprietary models and recruit talent, illustrating the scale of capital required and the intensity of the comparator set. TheAI Rankings describes Qwen3.6 as an open-weight model under Apache 2.0 with free self-hosting, creating potential pricing and differentiation pressure for model-centric businesses. Forbes' reporting on tranched AI financings shows why a headline valuation can materially overstate a blended economic price and why employee or later-investor pricing can diverge from the first tranche. For Prague Technology, the missing financing documents are therefore not a clerical gap: they determine dilution, liquidation preference, cash available, and whether the reported or implied price is investable.[CV005, CV006, CV007, CV008, CV009]

8.3 Scenario framework and diligence triggers

A scenario range is more defensible than a point estimate because the assigned evidence does not establish revenue, ARR, customer count, gross margin, burn, or a confirmed round price. The bull case requires proof that Prague Technology converts founder and research credibility into a differentiated product, secures paid industrial or platform deployments, and retains enough compute access to iterate under export-control constraints. The base case assumes continued technical development and sector growth but delayed monetization, with a valuation that must be negotiated against milestone evidence rather than narrative scarcity. The bear case combines an unverified or preference-heavy financing price with weak commercialization, open-model substitution, compute constraints, or talent dilution; in that case the last round would not be a reliable mark and a down-round or recapitalization risk would be material. The decision should move from track to invest only after primary documents answer the financing questions and operating evidence supports a model-appropriate revenue or milestone framework. Exit readiness is currently low because no public evidence in this pool establishes liquidity, a strategic buyer process, or a credible exit window.[CV010, CV011, CV012, CV013]

8.4 Final diligence asks and thesis-break triggers

The next diligence packet should include the signed financing documents, capitalization table on a fully diluted basis, option-pool treatment, liquidation and participation preferences, pro-rata rights, investor concentration, debt, and unrestricted cash. It should also include monthly management accounts, runway and burn reconciliation, revenue contracts, paid-pilot conversion, deployment reliability, compute procurement, export-control counsel, model evaluations against relevant open and closed alternatives, and a hiring and retention plan. The thesis breaks if the company cannot substantiate the claimed financing or valuation, if material proceeds are unavailable because of preference or escrow terms, if technical milestones repeatedly slip without paying customers, or if compute access prevents reproducible training and deployment. A valuation committee should use a staged entry with milestone-based release rather than relying on a headline mark. The public evidence supports the diligence path and a track recommendation; it does not support a precise target return, exit multiple, or investment price.[CV014, CV015, CV016]

8.5 Exhibits