Vault Bio

• The 46/100 feasibility score is mis-anchored: it inherits therapeutics-sector frontier priors (base/prime editing, organoid screens, 8-12yr binary trials) that are irrelevant to an industrial-enzyme platform. Vault faces no clinical trials — its 'endpoint' is kcat/Km, thermostability (Tm), and titer in a bioreactor, measurable in weeks. On the correct benchmark I'd argue feasibility is closer to 6
• Generative protein design is genuinely at an inflection: AlphaFold2/3, ESMFold and diffusion models (RFdiffusion, ProteinMPNN) now produce de novo binders and enzymes with experimentally validated success rates that have risen from <1% to double-digits for some scaffolds. Thermostabilization (raising Tm 10-20°C via consensus/ML-guided mutation) is one of the most de-risked applications — FoldX/Ros
• Wet-lab-in-the-loop is the correct architecture and the key credibility signal: it means they treat models as hypothesis generators, not oracles, closing the design-build-test-learn gap where pure in-silico plays fail. This directly counters hand-wavy 'AI designs enzymes' claims.
• Traction is real and domain-appropriate: 3 paid discovery contracts with chemical majors, one enzyme at pilot scale, $3.4M committed milestones. For industrial biocatalysis, pilot-scale + repeat majors is a stronger technical proof than any published benchmark — it means an enzyme survived scale-up, the usual failure point.

• Scale-up (the pilot→commercial cliff) is the true technical risk, not design. Enzyme performance at bench (mg) routinely degrades 10-100x in titer/stability under industrial conditions (solvents, high substrate load, continuous flow, 60-90°C). One pilot enzyme is n=1; the platform's repeatability across chemistries is unproven.
• Moat is thin where it matters: enzyme sequences are patentable but easily designed-around, and the real defensibility is proprietary assay/training data — none disclosed. Incumbents (Codexis, Novozymes/Novonesis, Ginkgo) have 15+ yrs of directed-evolution datasets Vault cannot yet match; the 60/100 moat score is generous absent data-asset disclosure.
• Zero quantified plan metrics (0% signal coverage): no titer, Tm improvement, cost-per-kg vs petrochemical incumbent, or hit-rate figures. For a science-driven bet this is a material diligence gap — cost-per-kg parity with petrochemical catalysts is the binary de-risking milestone and it is entirely unstated.

• Market factor 74/100; unit-economics factor 61/100.
• Reference gross margin ~85%; validate against actuals.
• Execution signal 58/100 — commercial validation cited.

• TAM/SAM/SOM and CAC/LTV unconfirmed — require a live data room.
• Sector benchmarks are directional, not company-specific.

• The 74/100 market score is miscategorized. This is industrial biocatalysis, not therapeutics — the relevant TAM is the ~$8-10B industrial enzyme market (~6-7% CAGR), not a $1.77T pharma pool at 14%. That inflates size but the more honest frame lowers growth. Net: I'd mark this to ~58-62, because the specialty-chemical enzyme wedge Vault targets is a subset (~$2-3B addressable), though margin quali
• Positive structural note: mislabeling as 'Biotech/Therapeutics' also imports the wrong risk profile. Vault has NO binary trial risk and NO 8-12yr FDA timeline — it sells enzymes B2B to chemical majors, so revenue is contract- and milestone-gated, not approval-gated. The regulatory (36) and competitive (65) scores penalizing trial risk are largely inapplicable and should be revised upward.
• Demand economics are favorable where a thermostable enzyme genuinely displaces a petrochemical catalyst: buyers face energy/decarbonization pressure and price-inelastic switching once qualified into a process line. That creates high switching costs and a supply-side moat via qualified-supplier lock-in — a more durable rent than the IP patents the model leans on.
• Rents accrue to whoever owns the wet-lab-in-the-loop data flywheel plus per-molecule process IP, not to the generative model itself (which commoditizes). 3 paid contracts + 1 pilot + $3.4M committed milestones is credible early validation, but $3.4M is discovery-stage revenue, not proof of production-scale royalty economics.

• Value-capture asymmetry: chemical majors have the fermentation/scale-up capacity and may license the enzyme then internalize manufacturing, capturing the margin while Vault earns one-time discovery fees. Without disclosed royalty or per-kg economics, the 85% gross-margin assumption is unverified — no LTV/CAC or revenue projection was supplied (0% signal coverage).
• Scale-up cliff: lab thermostability rarely survives to industrial titer/yield economics; a single pilot does not prove the enzyme beats the incumbent petrochemical catalyst on fully-loaded cost. Capital intensity flagged at 95% — $22M Series A likely funds only 2-3 more programs before another raise.
• Competitive incumbency: Novozymes/Novonesis, DSM-Firmenich and Codexis already run enzyme-engineering platforms at scale with decades of process data and customer relationships. Vault's generative edge must translate into faster/cheaper hits than well-capitalized incumbents — an unproven, not disclosed, differentiation.

• The 36/100 imports a 98% 'regulatory intensity' prior anchored to therapeutics (IND→NDA/BLA, GxP, binary trial risk, 8–12yr timelines). That gauntlet does not apply here: Vault sells enzymes for specialty-chemical manufacturing, not medicines. The governing regimes are TSCA (EPA new-chemical/microbial review — enzymes/engineered strains may need a TSCA §5 PMN/MCAN, ~90-day review) and OSHA/EPA pro
• IP posture is the load-bearing asset and the model's 60/100 moat (55% realized) is fair-to-generous: generative-protein sequences face uncertain composition-of-matter patentability (Myriad/Mayo §101 exposure for 'products of nature'-adjacent sequences), so protection likely leans on trade-secret over the AI weights, training data, and wet-lab process — durable but fragile to employee departure. Co
• Deal structure: standard Reg D 506(b) or 506(c) priced Series A — insist on 1x non-participating pref, broad-based weighted-average anti-dilution, pro-rata, board seat (not mere observer given $22M / early stage), and IP-representation warranties with a specific carve-out that all inventor/employee IP is assigned and that customer discovery contracts do not grant the chemical majors background-IP
• The three paid discovery contracts are the strongest signal but also the sharpest legal risk vector: verify whether the chemical-major agreements assign foreground IP to Vault or the customer, whether they contain exclusivity/non-compete field restrictions, and whether the $3.4M milestone payments are contingent on regulatory (TSCA) clearance of the pilot enzyme.

• Customer-contract IP leakage: if the discovery agreements grant the chemical majors ownership or exclusive fields over the co-developed enzymes, Vault's platform value collapses to a fee-for-service CRO — the single biggest undisclosed legal item; 0 quantified fields and no contract terms in the plan means this is unverified.
• §101 patentability gap: engineered protein sequences may be unpatentable as products of nature, forcing reliance on trade secret, which offers no protection against independent re-derivation by better-funded incumbents (Ginkgo, Codexis) — the 'ip patents mature moat' assumption in the model may not hold for this asset class.
• TSCA/biosafety timing: a genetically engineered production strain can trigger an EPA MCAN and potential containment/field-release review, adding 3–12 months per commercial enzyme before scale-up — modest vs. FDA but capable of stalling the milestone-payment schedule and thus the runway on a $22M raise.