Post-Migration Vault Audit
Prepared 2026-04-12 following Notion import. This document surfaces patterns, flags contradictions, and identifies gaps — it does not draw conclusions. Bliss and Dustin synthesize; this document prepares inputs.
1. State of the Vault
137 files across 6 directories. Covers approximately November 2025 through early April 2026.
| Directory | Files | Coverage |
|---|---|---|
primer/ | 56 | 7 EE 292P lectures; 10 semis actor pages; 22 oil actor pages; 17 research docs |
thesis/ | 24 | Charter, positioning, Q2 goals, 6 Meat+Potatoes drafts, kill criteria |
daily-log/ | 23 | Internal sessions Nov 2025 – Mar 2026; agendas; Bliss’s running notes |
interviews/ | 19 | External meetings with named advisors, investors, experts |
synthesis/ | 10 | Golden Nuggets; competitive landscape; AI-generated selection memos; rubric; research gaps |
prototyping/ | 5 | Ontology architecture; schema v0; demo queries; data asset decomp |
Contacts DB: 14 rows. 2 contacts with last name unknown (Eyck, Milani). 3 referred but not yet met (Singleton, Kumar, Tambi).
What is NOT in the vault:
- No dedicated contact files in
contacts/— all contact data lives incontacts.dbonly - No synthesis doc covering the full advisor network and what each person said
- No synthesis doc on the evolution of the oil-analog thesis
- No customer discovery notes from actual semiconductor operators (all advisors; zero operators)
- No competitive deep-dives on individual players (Altana, Interos, SAP GTS)
- No record of what happened to several November/January action items (Kevin Rudd, Condi Rice contacts; 2022 shortage case study; competitor scrape assigned March 2)
2. Contact Map
Network Topology
The advisor network has real depth across three Stanford-adjacent tiers:
Tier 1 — Core strategic advisors (3+ touchpoints):
- Richard Dasher (Stanford APARC) — most sustained engagement; deepest Asia-Pacific semiconductor context; co-designed design-thinking process; connected team to key contacts
- Steve Blank (Stanford) — product strategy architecture; Lean Startup framing; oil industry template; gave the “build simulation first” directive that shifted March strategy
Tier 2 — Domain specialists (1–2 substantive meetings):
- Ann Miura-Ko (Floodgate) — AI-era product philosophy; “JP Morgan of the industry” framing; “compliance now, derivatives later”
- Dan Iancu (Stanford GSB) — academic grounding in supply chain risk; four opportunity areas that align closely with the product vision; offered intros to Rigetti and Devoret
- H.R. McMaster (independent) — national security framing; defense tech ecosystem map; introduced Sean Singleton
- Steve Weinstein — product discipline (“daily-use vs get-off-the-bus”); compliance-as-daily-wedge recommendation; customer-first critique
- Eyck — most direct challenge to data acquisition thesis; defense-first GTM; competitive intelligence (Wirescreen, Exeger)
- Holland Ferguson — federated architecture validation; cold start framing; downstream-first market selection
Tier 3 — Single meetings, specialist context:
- Milani (Iran/MENA geopolitics), Elsie Wong (HK PE/VC), Cong Le (lab automation)
Gaps in the Contact Map
Missing perspective types per RDI methodology:
- Operators — No one who currently does semiconductor export compliance, supplier onboarding, or UFLPA diligence day-to-day. This is the largest single gap.
- Compliance officers — The named buyer (VP Export Compliance) has never been interviewed. All compliance insight comes from advisors observing compliance pain from outside.
- Tier-2 downstream customers — March 18 notes identify auto tier-2 (Magna, Amphenol, Delphi) as the go-to-market target. None have been contacted.
- Regulators — No one from CBP, BIS, OFAC, or the CHIPS Office. Enforcement intent and current enforcement gaps unknown from primary source.
- Startups in the space — Altana, Interos, and Exeger are named but never spoken to. Their product limitations are inferred, not observed.
- Academics studying supply chain compliance — Dan Iancu is operations research; nobody from trade law, export controls scholarship, or economic geography.
People offered as introductions, never followed up on (as of vault date):
- Chad Rigetti (Rigetti Computing) — via Dan Iancu
- Michel Devoret (Google AI, Nobel) — via Dan Iancu
- Alan Minor (Oracle/Salesforce Japan) — via Richard Dasher; flagged as “when ready for SaaS procurement expertise”
- Nick Santhanam (ex-McKinsey semiconductors) — mentioned Mar 18; no notes
- Tyler Goodspeed (Chief Economist, ExxonMobil) — mentioned Mar 18; oil-analog depth; no notes
- Chris Moran (Lockheed Ventures) — via McMaster; digital twinning context
Partial-name contacts that need resolution:
- “Eyck” — likely Eyck Freymann (Harvard, China policy) but unconfirmed
- “Milani” — possibly Abbas Milani (Hoover Institution, Iran expertise) but unconfirmed; worth confirming before any follow-up
Cross-references not yet linked in the DB:
- Steve Blank appears to have introduced Steve Weinstein (mentioned in debrief notes) — referral chain should be updated
- Dan Iancu and Richard Dasher are both Stanford; worth knowing if they know each other before introducing the team
3. Themes Across Interviews With No Synthesis Doc
The following patterns appear in three or more interview or daily-log files but have no dedicated synthesis document. These are candidates for future synthesis work.
Theme A: The Cold Start / Data Acquisition Problem
Appears in: data-asset-decomp.md, golden-nuggets.md, 2026-02-11-session.md, 2026-02-28 notes, 2026-03-eyck.md, 2026-03-holland-ferguson.md, synthesis/research-gaps-petal.md
Every document that touches data acquisition either asserts that compliance workflows will unlock proprietary supply chain data, or explicitly flags this as the biggest unsolved problem. Golden Nuggets calls for “a 3-hour strategy session just thinking about creative ways to incentivize contribution of data.” Eyck says the most critical data (Applied Materials’ tier-3 suppliers, private Malaysian packaging firms) is precisely the data that won’t flow through compliance workflows. Holland Ferguson’s entire session was structured around the cold start problem. The thesis has not resolved this. It has named it and moved past it.
What a synthesis doc would need to answer: What specific data does the MVP compliance product actually generate? Is that data sufficient to build a useful digital twin, or does it just produce supplier name lists? What’s the minimum viable data pool for the simulation demo to be credible?
Theme B: The Downstream-First Go-to-Market Insight
Appears in: 2026-03-15-session.md (ABNI followup), 2026-03-holland-ferguson.md, 2026-03-eyck.md, Steve Blank advice across multiple sessions
A clear strategic insight emerged late in Q1 but hasn’t been synthesized: don’t start with fabs (TSMC, Samsung, Intel are in a boom cycle, have pricing power, and are difficult to penetrate). Start with downstream electronics — auto tier-2 (Magna, Amphenol), networking OEMs, aerospace — who have the most pain during shortages, who have the budget authority to request supplier data, and who are less strategic about data sensitivity. This completely changes the initial sales motion from “sell to the fab” to “sell to the buyer who got burned.”
What a synthesis doc would need to answer: Which downstream segment has the most acute, solved-with-software pain today? What does a conversation with a VP Supply Chain at a tier-2 auto supplier actually look like?
Theme C: What Advisors Agreed On vs. Where They Diverged
Appears in: Across all 19 interview files
There is strong advisor consensus on: compliance as the wedge; financial products as the long-term prize; AI-native architecture as a genuine differentiator; the oil analogy as a useful frame. But advisors diverged meaningfully on: (a) whether to focus on fabs or downstream (Blank: build simulation for all; Eyck: go defense; Weinstein: pick one vertical); (b) whether to show the shiny demo first or find the customer first (Blank: demo first; Weinstein: customer first); (c) competitive threat level from incumbents (Eyck: Altana and Exeger are serious; rubric materials: organizational inertia is the defense). These divergences have not been synthesized.
What a synthesis doc would need to answer: Which strategic disagreements between advisors actually matter for the next 90 days? Which can be resolved by a single customer conversation?
Theme D: The “AI-Native” Claim Needs Specificity
Appears in: charter.md, golden-nuggets.md, 2026-03-08 strategy session notes, 2026-03-15-session.md
“AI-native from day one” is a recurring claim but the specific meaning drifts across documents. In some places it means “agentic simulation interface.” In others it means “designed to be invoked by AI agents.” In others it means “built for a world increasingly run by mathematical models.” Ann Miura-Ko’s drone-swarm framing (build 10 versions cheaply) is a different interpretation entirely. The patent-worthy technical differentiation question (March 13 and 15 sessions) was never resolved.
What a synthesis doc would need to answer: What specifically does “AI-native” mean for the MVP? What can this product do that a Palantir ontology + a GPT wrapper cannot? Is this a product claim or an architectural claim?
Theme E: The Commercial vs. National Security Mission Tension
Appears in: 2026-01-14-session.md, 2026-01-29-session.md, 2026-03-hr-mcmaster.md, multiple Dasher sessions, thesis/charter.md
From November onwards, the team has acknowledged a tension between “build a lucrative business” and “advance US competitiveness in semiconductors.” Steve Blank’s advice was explicit: “Don’t confuse national service mission with business opportunity.” McMaster and Dasher both frame the problem through a national security lens, which shapes the product vision in ways that may or may not align with what a commercial buyer will pay for. The charter resolves this by saying the commercial product happens to produce geopolitical benefit — but whether this is true, or whether the national security framing will create sales and regulatory complications, hasn’t been examined.
What a synthesis doc would need to answer: Is there a version of this product that a purely commercial buyer pays for without any national security framing? And separately: is there a version that a government agency pays for? Do these require different products?
4. Primer Flags
Items in primer/ that appear outdated or contradicted by newer material in interviews/ or synthesis/.
Flag 1: “Semiconductors as the New Oil” Framing — Structurally Complicated
Flagged files: primer/semis-as-new-oil.md, primer/oil-analog-canon.md, primer/global-semis-oil-analog.md
Contradicting file: synthesis/controlled-dominance-test.md
The oil-analog framing pervades the primer and is treated as foundational. But the Controlled Dominance Test — which appears to have been run later — explicitly states: “DRAM is not gasoline and not perfectly fungible.” Batteries score 5/5 on standardized unit clarity; semiconductors score 3/5. The oil analogy works for the strategic narrative (both are critical inputs, both underpin geopolitical leverage) but does not hold for the specific financial infrastructure claim (price indices, derivatives, commodity exchange) because logic chips are custom-engineered, not interchangeable. The distinction between “oil as a strategic frame” and “oil as a financial infrastructure template” is present in later daily-log notes (team settles on cyclical volatility as the pattern, moves away from commodity exchange language) but the primer documents haven’t been updated to reflect this.
Question for founders: Is the oil analogy in the primer documents the current view, or has it been superseded by the digital-twin-and-financial-products framing that doesn’t require commodity fungibility?
Flag 2: “Structural Shortage” Language in Some Primer Docs
Flagged files: primer/semis-volatility-risk.md (mentions “structural shortage for next 40 years” as one possible trajectory)
Contradicting material: daily-log/2026-02-05-agenda.md, daily-log/2026-02-11-session.md
Daily-log notes from February show the team explicitly debating “structural shortage vs. cyclical volatility” and resolving on cyclical. The Feb 5 semis framework test identifies cyclical volatility as the actual pattern. Some primer research documents still carry the structural-shortage framing as a live hypothesis rather than a rejected one.
Flag 3: Historical Semiconductor Benchmarking Failures Not Integrated
Flagged files: primer/hedging-benchmark-layer.md, primer/semis-volatility-risk.md
Source of flag: synthesis/venture-selection-gemini.md
The GEMINI memo explicitly references Enron’s 2001 DRAM forward contracts as a precedent failure for semiconductor financialization. The primer documents on hedging and volatility do not address this. A reader who learns from the primer alone would not know that DRAM futures were tried and failed. The failure-mode analysis should be in the primer.
Flag 4: Competitive Landscape in Primer vs. Synthesis
Flagged file: primer/external-research-references.md
More current file: synthesis/competitive-landscape-2026-03.md
The external research references doc was written earlier and does not reflect the March 2026 competitive landscape update. Altana’s recent fundraise, Cadence’s $213M penalty (July 2025), and the BMI/ICE battery futures launch are not in the primer research references. Anyone using the primer as a competitive reference should use synthesis/competitive-landscape-2026-03.md as the current source instead.
5. Thesis Pressure Points
Claims in thesis/ that are unsupported by evidence in interviews/ or contradicted by evidence in synthesis/. This is not a judgment about whether these claims are wrong — it is an inventory of what needs validation.
Pressure Point 1: Companies Will Share Proprietary Supply Chain Data
Thesis claim: Charter, V3-V6, golden-nuggets — “every compliance check, every supplier onboarded, every shipment verified adds a unique node/edge to the ontology.” The model assumes compliance automation generates data exhaust that compounds into a proprietary digital twin.
What interviews say: Eyck (March 9): “Most critical data is proprietary. Applied Materials doesn’t know its own tier-3 suppliers, let alone share them. Packaging/testing companies are largely private Malaysian firms.” Data-asset-decomp flags the incentive problem as unsolved. Research-gaps-petal calls it an open question. No interview note contains a person saying “yes, I would share our supplier data in exchange for a compliance tool.”
Status: This is the load-bearing assumption of the entire business model. It has not been validated with a single potential customer.
Pressure Point 2: The Oil Analogy Supports Financial Products
Thesis claim: Charter and all Meat+Potatoes versions — “if semiconductors are the new oil, financial infrastructure will follow: price indices, hedging, parametric insurance, capacity intermediation.”
What synthesis says: Controlled Dominance Test rates semiconductors 3/5 on benchmarkability and standardized unit clarity, noting “DRAM is not gasoline.” Oil-analog scan acknowledges “semiconductors may be structurally unsuitable for commodity-style infrastructure.” Prior attempts (DRAM futures, Enron 2001) failed. Battery Passport analysis rates batteries as structurally superior for this path (5/5 on most financial infrastructure criteria).
The distinction that matters: The oil analogy may be valid for the intelligence and risk products (scenario modeling, disruption alerting, risk scoring) without requiring commodity fungibility. It likely does not support the commodity exchange path. The charter has evolved away from the exchange framing, but it’s worth being explicit about which parts of the analogy still hold.
Pressure Point 3: Regulatory Tailwind Is Durable
Thesis claim: Rubric scores regulatory tailwind 5/5 and claims “trend toward more regulation spans both parties and allied governments.” Charter says “compliance is mandatory; firms must perform it regardless of administration.”
What interviews raise: No advisor directly contradicts the regulatory tailwind claim, but none has validated it either. H.R. McMaster’s session (March 18) deals with current geopolitics, not regulatory durability. The rubric’s own risk section flags “geopolitical assumption dependency” and “regulatory reversibility” as yellow flags. Export controls may be bipartisan, but UFLPA enforcement intensity and EU Battery Passport timelines are both subject to political and trade negotiation pressure.
What’s missing: A single conversation with a BIS or CBP official, or an export controls attorney, about whether current enforcement priorities are durable.
Pressure Point 4: Founder-Market Fit Is Strong
Thesis claim: Rubric scores “Founder-Market Fit” at 5/5 and cites GSB training, Palantir network, and academic sponsorship (Prof. Berk).
Internal contradiction: The rubric’s own definition of 5/5 is “direct operational experience in the target industry.” The rubric simultaneously notes “founders lack deep semiconductor expertise.” These statements are in the same document. The 5/5 score reflects conviction, not the stated scoring criteria.
What this means for next steps: Semiconductor domain credibility is a gap that advisors can partially fill, but at some point an operator (a VP Export Compliance, a chief supply chain officer) either validates or challenges the product vision in ways no advisor can substitute for.
Pressure Point 5: Altana’s Organizational Inertia Is a Sufficient Moat
Thesis claim: Competitive-landscape doc identifies Altana ($1B, $322M raised) as “most dangerous competitor” but argues organizational inertia and multi-industry focus prevent them from winning semiconductor compliance. Rubric scores incumbent displacement at 3–4/5.
What’s missing: Any analysis of what Altana’s product actually does, who their semiconductor customers are (they serve CBP and defense agencies per competitive landscape), and what specifically they cannot do. The defense is “organizational inertia” — but $322M buys a lot of focus if the opportunity becomes clear. No one has talked to a former Altana employee or customer.
Pressure Point 6: The MVP Directly Feeds the Digital Twin
Thesis claim: Charter and Intake Brief — “compliance is the Trojan horse; every workflow feeds the ontology; no wasted motion.”
What the prototyping docs reveal: The ontology v0 seed universe is 5 firms, 10-15 products, 3 regulations — deliberately thin. Demo queries v0 include scenario analysis that assumes data not yet in the graph (tier-2/3 networks, beneficial ownership chains, financial impact modeling). The Q2 roadmap specifies an MVP “on public data.” If the MVP runs on public data and the digital twin requires proprietary data, the pathway from MVP to digital twin is not the “straight line” the charter implies.
6. Gaps and Suggested Next Interviews / Research
Ranked by impact on de-risking the core thesis.
Gap 1 — No operator interviews [HIGHEST PRIORITY]
Every interview in the vault is with an advisor, investor, academic, or policy figure. Zero interviews with someone who currently does semiconductor export compliance, UFLPA diligence, supplier onboarding, or supply chain risk management as their job.
Suggested interviews:
- VP Export Compliance at a mid-size semiconductor OEM (not a top-5 fab — they have armies of lawyers; a $2-5B chipmaker is a more realistic first customer)
- Customs broker or freight forwarder who handles UFLPA detentions (they see the pain at ground level)
- Supply chain risk manager at an auto tier-2 supplier (Magna, Aptiv, Amphenol) — validates the downstream-first GTM
- Export controls attorney (validates regulatory durability and specific legal use cases)
What to ask: Not “do you like our idea?” — instead: “Walk me through the last time a shipment was detained. What did you do? What did you use? What did you wish you had?” and “How do you currently track your tier-2 suppliers?”
Gap 2 — No analysis of why prior semiconductor benchmarking failed
The GEMINI memo flags Enron’s 2001 DRAM forward contracts as a precedent failure. Prior to committing to the financial products roadmap, understanding specifically why these failed matters enormously. Was it lack of liquidity? Non-fungibility? Regulatory barriers? Counterparty risk? The answer shapes how the financial products arc should be sequenced or whether it should be attempted at all.
Suggested research:
- Academic literature on commodity financialization failures (agricultural futures, bandwidth trading, etc.)
- Conversations with anyone who was at CBOT, CME, or an energy trading desk when semiconductor derivatives were attempted
- The BMI trajectory from price reporting (2014) to ICE futures (2025-26) — what made batteries work when chips didn’t?
Gap 3 — No competitive intelligence on Altana from the inside
Altana is named as the most dangerous competitor in the competitive landscape doc but has never been analyzed from primary source. What does their product actually do? What are its known limitations? Who are their semiconductor customers? Why have they not moved into semiconductor compliance specifically?
Suggested approach:
- Talk to a former Altana employee or a company that evaluated Altana and chose not to buy
- Talk to a customs broker or compliance consultant who has used Altana in a semiconductor context
- The competitive landscape doc notes Altana “serves CBP and defense agencies” — understanding what that relationship looks like shapes the risk level
Gap 4 — Legal validation of the multi-tenant data model
The data-asset-decomp and research-gaps-petal note that US trade manifest confidentiality rules prevent publishing shipment details, and EU Battery Passport regulations (2023/1542) restrict data reuse in ways that may undermine multi-tenant aggregation. No lawyer has looked at the specific architecture.
Suggested research:
- One conversation with an export controls attorney about what data TBD can legally aggregate across customers
- One conversation with an EU counsel about Battery Passport data residency requirements
- Clarity on whether differential privacy or federated learning satisfies the legal constraints or just the ethical ones
Gap 5 — Customer willingness-to-pay research
The CLAUDE memo asserts $200–500K ACV for compliance SaaS with 5–10 logos in Year 1. This has never been validated with a potential buyer. Enterprise compliance sales cycles are 6–18 months. The math may not work without external funding.
Suggested approach:
- Ask three or four VP Export Compliance or VP Supply Chain contacts for a “pretend you were going to buy this” pricing conversation
- Research what SAP GTS and Oracle GTS actually charge for trade compliance modules
- Look at what Exeger and Interos publicly disclose about pricing
Gap 6 — The synthesis docs need a “surprises” section
Per RDI methodology, every synthesis memo should have an explicit surprises or contradictions section. Looking across synthesis/, the AI-generated memos (CLAUDE, GEMINI) are well-structured but were produced without primary research. The rubric and decision matrix are frameworks, not syntheses. Only Golden Nuggets and Petal Notes feel like genuine distillations of learning. The vault would benefit from a synthesis doc written after the first five operator interviews — one that specifically tracks what surprised the founders vs. what confirmed prior assumptions.
7. Proposed Reading Order
For a new collaborator (investor, technical partner, domain expert) who needs to get up to speed. Sequenced to build understanding in layers, not to impress.
Layer 1 — The Hypothesis (30 minutes)
thesis/charter.md— What we’re building and whythesis/kill-criteria.md— Intellectual honesty document; the conditions under which this doesn’t workthesis/one-pager.md— The compressed version
Layer 2 — The Industry Context (1–2 hours)
primer/bliss-primer-2026-04-09.md— Comprehensive primer; read this before anything else in primer/primer/oil-analog-canon.md— The theoretical foundation; what the oil market built over 100 yearsprimer/three-insights-semis.md— Key framing: complexity ≠ resilience; resilience is a public good; financial infrastructure is missingsynthesis/competitive-landscape-2026-03.md— Who else is working on this; where the whitespace is
Layer 3 — What We’ve Heard (1–2 hours)
interviews/2026-03-eyck.md— Most direct challenge to the thesis; read this skepticallyinterviews/2026-03-ann-miura-ko.md— Best strategic framing of what to build and in what orderinterviews/2026-03-steve-weinstein.md— Sharpest product discipline; the daily-use testinterviews/2026-01-dan-iancu.md— Academic grounding for the data asset thesissynthesis/golden-nuggets.md— How the thinking evolved over Jan–Feb 2026
Layer 4 — The Scoring and Gaps (45 minutes)
synthesis/tbd-rubric-draft.md— How the founders score this opportunity (read critically)synthesis/research-gaps-petal.md— What’s still unknown and contestedsynthesis/controlled-dominance-test.md— The one document that pushes hardest against the thesis
Layer 5 — The Product (30 minutes)
prototyping/ontology-architecture.md— What the product is technicallyprototyping/demo-queries-v0.md— What it would need to demonstratethesis/2026-q2-goals.md— What Q2 looks like with different confidence levels
8. Surprises and Contradictions
Per RDI methodology, this section is the most important. These are not noise — they are where the learning is.
Surprise 1: The Controlled Dominance Test rates batteries as structurally superior to semiconductors for the financial infrastructure thesis. The team committed to semiconductors anyway — partly because of the advisor network, partly because of the compliance enforcement severity. Whether batteries were the right answer to abandon deserves explicit revisiting, not just “we chose semis.”
Surprise 2: Zero operator interviews in 5 months of research. The RDI methodology calls for operators first. The vault is full of advisors, investors, academics, and policy figures — all high-quality, all insightful — but none of them are people who currently run a compliance desk or manage semiconductor supply chain risk as their primary job. This is the loudest silence in the vault.
Surprise 3: The March 18 Steve Blank advice — “build a macroeconomic semiconductor simulation first; use it to start conversations with downstream electronics companies” — is a significant strategic shift that appears in only one daily-log note and has not been synthesized into the thesis documents. The charter still describes starting with compliance. The March 18 insight may be better than the charter, or it may be in tension with it. It needs a decision.
Surprise 4: The financial products arc (“compliance → intelligence → hedging → insurance → marketplace”) is described as “sequenced and inevitable” in the charter and rubric. But BMI (battery benchmarking) took roughly 10–12 years to move from price reporting to regulated benchmark to ICE futures. The charter implies this can be compressed dramatically by AI and regulatory forcing. That compression claim has no evidence attached to it.
Surprise 5: Holland Ferguson’s meeting (March 13) surfaced a practical product insight that doesn’t appear in any thesis document: the federated architecture where the digital twin “sits between participating firms” and preserves individual data silos while enabling aggregated insights. This is architecturally different from the multi-tenant ontology described in the charter. Either they’re the same thing described differently, or there’s a genuine design tension that should be resolved.
This document prepares inputs for synthesis. It does not recommend a strategy, validate the thesis, or reach conclusions. Those belong to Bliss and Dustin. If there are none to two surprises in a given reading of this document, we’re not learning fast enough — revisit the operator interview gap.