The Outbound Investment Inquiry File Standard: Clear, Escalate, or Stop

This standard defines what a finished outbound-investment inquiry file looks like for one target: the AI, semiconductor, or quantum startup you are about to wire into. It is for deal leads, platform and talent partners, and angels with no compliance team behind them. It sets out the eight elements a file must contain, the evidence that closes each one, and grading rules strict enough that a second reviewer reaches the same verdict: Clear, Escalate to counsel, or Stop. It does not replace counsel. It is the file counsel needs to see before signing off.

What does a finished outbound inquiry file contain?

A finished inquiry file has three parts: a regime stamp, eight evidence elements, and a signed grade. Each element is closed with a value, a source, and a date. Treasury never defines the file itself. What it does is list seven factors in 31 CFR 850.104(c) that it weighs when deciding whether a U.S. person made a "reasonable and diligent inquiry." This standard turns those factors into fields.

The factor list is illustrative, and Treasury assesses it on the "totality of relevant facts and circumstances." No public source sets a minimum evidence standard for each factor. What you can control is who closes each one:

The eight elements below are the fields. For each one I give what it proves and what it looks like when it lies.

1. Entity ties. Every legal entity in the group, with its country of incorporation, its headquarters, and where it operates. Proves where the business sits, legally and physically. Lies when a U.S. parent keeps its engineering team in a foreign subsidiary that the registry search never turned up.
2. Subsidiary and affiliate share. The share of revenue, net income, capex, and opex that sits with any country-of-concern entity, shown as computed numbers. Proves whether the entity test's financial-metrics prong (more than 50%) is met. Lies when the target supplies group totals only.
3. Cap-table control. The fully diluted cap table, plus board seats, observer rights, vetoes, and contractual management rights, running in both directions. Proves who directs the company and whom the company directs. Lies when control sits in a side letter or a holding layer the cap table leaves out.
4. Founder and key-person nexus. For each founder, director, officer, and key person: their status under the individual test and their current roles at other entities. Proves whether any individual is a covered foreign person. Lies when status is inferred from a name, ethnicity, or birthplace.
5. Covered-activity mapping. Each activity listed in 850.217 and 850.224, marked present, absent, or unknown, with the technical evidence cited. Proves whether the business does what the rule covers. Lies when the mapping comes from a pitch deck or website copy.
6. List screening. Every entity, owner, and investor screened against the 850.224(m) lists, timestamped, with an ownership check up the chain. Proves whether a covered deal would be prohibited rather than notifiable. Lies when the matching is on names only.
7. Reps sought and obtained. The text you asked for and the text that was signed, side by side. Proves factor 2. Lies when a knowledge-qualified rep is logged as obtained.
8. Red-flag non-answers. Every refusal, evasion, and silence, along with the question it was a response to. Proves that factor 6 was considered. Lies when these are logged as missing data rather than as warning signs.

Which regime and exceptions apply on the inquiry date?

Until the final rule under the COINS Act takes effect, the regime in force is 31 CFR Part 850, which governs the Outbound Investment Security Program (OISP). Name the regime on the file's first line, next to the inquiry date. Treasury has published a proposed rule under the COINS Act, but the existing rules stay in force until the final rule takes effect. Grading against the proposal is a mistake whichever way it cuts.

The two regimes differ on nearly every point that matters to the file:

The proposed rule's six countries are China (including Hong Kong and Macau), Cuba, Iran, North Korea, Russia, and Venezuela under the regime of Nicolás Maduro Moros. It also extends coverage to members of the CCP Central Committee or other political leadership of a country of concern. None of this applies to a file graded under Part 850. All of it will apply when you re-run the file after the final rule takes effect. To check where things stand, read Treasury's outbound investment program FAQs and the Federal Register docket TREAS-DO-2026-0595.

Recording reliance on an exception

The LP exception applies when the U.S. person's committed capital does not exceed $2 million, or when the investment is subject to a specific binding contractual assurance. The proposed rule keeps it. The $2 million is added up across investment and co-investment vehicles, and that is where funds miscount. The statute also lets the Secretary of the Treasury set a de minimis amount. Press reports of a $50,000 general threshold do not appear in Treasury's release, so treat that figure as unverified and don't rely on it. The proposed rule's $50,000 per-entity aggregation floor inside the financial-metrics test is a different number.

Neither the rule nor Treasury's FAQs say how a file should record reliance on an exception. Until they do, my standard has three parts:

• Attach the governing document: the commitment, or the binding contractual assurance.
• Show the aggregation arithmetic across every vehicle, by name.
• Have the fund CFO initial the arithmetic.

If the document isn't attached, the file can't rely on the exception.

Is the target a covered foreign person, and does a list screen answer that?

Whether the target is a covered foreign person and whether it appears on a screening list are two separate questions, and a clean screen answers only the second. Status decides whether the rule reaches the deal at all. The 850.224(m) lists decide whether a deal with a covered foreign person engaged in a covered activity is prohibited rather than notifiable.

This matters because Treasury does not intend to publish a list of entities it treats as covered foreign persons. No database will tell you "covered" or "not covered." You establish status from the entity map, the financials, the control rights, and the people. You answer the prohibition question from the lists.

The individual test

An individual is covered if they are a citizen or permanent resident of a country of concern (under Part 850: China, Hong Kong, or Macau), are not a U.S. citizen, and are not a permanent resident of the United States. Every clause counts. A Chinese citizen with U.S. permanent residency does not meet the test. Take the status from documents the target supplies, and record only what the definition requires. This is the part of founder and cap-table diligence that most often turns into guesswork, and guesswork is what the file is meant to replace.

The third-country entity test

Under Part 850, an entity outside a country of concern can still be covered if both of these hold:

• It holds a board seat, a voting interest, or any contractual power to direct the management or policies of a person of a country of concern engaged in a covered activity.
• More than 50% of its revenue, net income, capex, or opex comes from that party.

This is the test that catches an AI startup with a China subsidiary. Delaware incorporation doesn't answer it. Only financials broken down by entity do. The proposed rule adds a 50%-or-more aggregate ownership test and a "direction or control" prong, and keeps the financial-metrics test with a $50,000 per-entity threshold for aggregation.

The lists that make a covered deal prohibited

Under 850.224(m), if a covered foreign person on any of these lists engages in any covered activity, the deal is prohibited, even if the activity would otherwise only be notifiable:

• The BIS Entity List
• The BIS Military End User List
• Treasury's SDN List
• Treasury's Non-SDN Chinese Military-Industrial Complex Companies List
• The "Military Intelligence End-Users" list
• Any party designated as a foreign terrorist organization

Treasury has not published a screening procedure or a refresh frequency for these lists. The Consolidated Screening List does not pick up unlisted affiliates that are 50% or more owned by a listed party, so a name screen on its own leaves a known gap.

What does a missed inquiry cost at seed and angel check sizes?

Civil penalties are the greater of $377,700 per violation or twice the value of the transaction. On any check below $188,850, the fixed figure is the one that applies, so the worst case is more than twice the check. At the smallest checks it is more than four times.

These figures aren't hypothetical. In Treasury's first enforcement case under the program, Amidi, LLC was fined $200,000 for failing to notify Treasury of an investment its controlled foreign subsidiary made into a private Chinese artificial intelligence company. The underlying investment was about $92,478, so the penalty imposed came to 2.16x the check. Notifications are due no later than 30 days after closing. Amidi was penalised for a missed filing, not for a prohibited deal.

Three more mechanics shape the file:

• Liability runs through entities you may not be watching. Amidi was penalised for a controlled subsidiary's investment, and Treasury can penalise a violation even if it was not authorized or even known about.
• Treasury can unwind the deal. It can nullify, void, or otherwise compel divestment of a prohibited transaction. Criminal exposure reaches up to $1 million and, for individuals, imprisonment for up to 20 years.
• Skipping the inquiry doesn't buy ignorance. A U.S. person that skips the inquiry may be held responsible for knowledge it could have acquired, and the statute of limitations is 10 years.

Before quoting the fixed maximum to an LP or investment committee, re-check the current figure in Part 850 on eCFR.

Who should build the file, and what does outsourcing leave with you?

An analyst can build most of the file, a technical partner maps the activities, and counsel grades the result. Outsourcing any part of it doesn't move the duty. On diligence done by a non-U.S. entity, Treasury says U.S. persons "must evaluate whether the diligence conducted by the non-US Person entity is sufficient," which leaves responsibility for sufficiency with the investor.

The practical problem is that venture firms rarely employ the people who do this work. In Refolk's index of professional profiles, I counted U.S. profiles with export compliance, export control, trade compliance, or sanctions compliance titles, then cut that set two ways.

In a 25-profile sample, the most common employers were Honeywell and Boeing (two each), followed by Google, JPMorganChase, Wells Fargo, and Oracle (one each). That is industrial and banking compliance, not funds. Treat 145 as a floor, not a market size. Title matching in the index is strict, and the sample included a "Demand Planning Supervisor" who only partly fit, so the real pool is both bigger and noisier than the count. The direction is still clear. A small fund should plan to borrow this capacity, from fund counsel or a fractional compliance officer, rather than assume it has it in-house.

Finding a fractional officer with venture clients is a people-search problem. Titles vary, client lists sit in profile summaries, and many of these people list themselves as independent. Refolk takes that description in plain English and returns matching people from across LinkedIn and the open web.

Whoever you hire, the file is still yours. Brief them on the eight elements and the grading triggers below, and grade their work exactly as you would grade your own analyst's.

The procedure: one target, eight steps

The procedure runs in eight steps, from scope to signed grade. The timed steps come to roughly two to three and a half working days, plus however long the target takes to answer and the founder step, which has no established duration. Treasury describes its framework as non-prescriptive rather than "one-size-fits-all" and sets no sequence. The order below puts the cheap public work first, so the request to the target can be specific.

A few notes on the steps that most often go wrong:

• Steps 3 and 5 are where public evidence does the most work. Registries show legal entities, but people show where the work actually happens. A list of the target's current employees with their stated locations is a quick cross-check on the entity map: an engineering team in Shanghai with no Shanghai entity on the map is an open question for the target. Refolk returns people by current employer and stated location in a single query, and it also shows each founder's other current roles, which step 5 needs. Neither source can establish citizenship. Status still comes from documents.
• Step 4 depends on the target. If the target sends group-level figures, send the request back and log the delay. A second non-answer goes in element 8.
• Step 7's cadence is my inference. No screening frequency is established. Screening at term sheet, signing, and closing is the minimum I would defend.

Send the request below with the term sheet, so the target's answers arrive while steps 3 and 6 are underway.

Subject: Outbound investment inquiry - information request

As part of standard outbound investment diligence, please answer each item below in writing. "Not applicable" is an acceptable answer; please state it rather than skipping an item.

1. Entities. List every legal entity in the group (parent, subsidiaries, joint ventures, branches) with place of incorporation, headquarters, countries of operation, and the group's ownership percentage.
2. Financial split. For the last full fiscal year and the current year to date, give revenue, net income, capital expenditure, and operating expenditure by legal entity.
3. Cap table and control. Provide the fully diluted cap table and every board seat, observer right, veto, and other contractual management right, with the holder of each.
4. Interests held. State whether the company or any entity in item 1 holds a board seat, voting interest, or contractual power to direct the management or policies of any person of China, Hong Kong, or Macau.
5. People. For each founder, director, officer, and key person, confirm whether they are a citizen or permanent resident of China, Hong Kong, or Macau who is neither a U.S. citizen nor a U.S. permanent resident, and list their current roles at other entities.
6. Technology. Describe the products, chips, models, or quantum systems in development, the compute used, and any published papers or public repositories.
7. Screening. Confirm whether the company, any entity in item 1, or any shareholder appears on, or is majority owned by a party on, the BIS Entity List, the Military End User List, Treasury's SDN or Non-SDN Chinese Military-Industrial Complex Companies Lists, or the Military Intelligence End-User list.

Representations to be requested in the stock purchase agreement, without a knowledge qualifier:
- The Company is not engaged in any "covered activity" under 31 CFR Part 850.
- The Company has no intention of becoming a "person of a country of concern."
- The Company does not hold board seats, voting interests, or contractual management power over any "covered foreign person."

Please flag any representation that cannot be given as drafted, and explain why.

The three representations follow the NVCA model stock purchase agreement, which is the natural place for outbound investment reps and warranties in a priced round.

How do you grade the file: Clear, Escalate, or Stop?

The grade is the most severe outcome triggered by any single element. Elements are never averaged or traded against each other. Each grade has fixed triggers, so two reviewers reading the same cover sheet land on the same grade.

Stop is not a legal finding that the deal is prohibited. It means this file can't support a wire, and the deal doesn't move until counsel resolves whatever triggered it. Escalate is not a soft Clear. It means a question exists that only counsel can answer.

The hardest calls happen where the evidence is thin. The matrix below is the tiebreaker.

To test whether the grade is reproducible, have a second reviewer grade the file blind, from the cover sheet alone. If they disagree, some field is not really closed. Find it and fix it. Don't negotiate the grade.

OUTBOUND INQUIRY FILE - COVER SHEET
Target:
Inquiry date:
Regime applied: 31 CFR Part 850 (in force) / COINS final rule (only if effective)
Check size and vehicles:
Exception relied on: none / LP exception - aggregate committed capital $____ across ____ vehicles, document attached: yes / no

ELEMENTS (status: closed / open / warning - evidence cited - date)
E1 Entity ties:
E2 Subsidiary and affiliate share (revenue / net income / capex / opex %):
E3 Cap-table control:
E4 Founder and key-person nexus:
E5 Covered-activity mapping (subsections marked present / absent / unknown):
E6 List screening - term sheet: ____ signing: ____ closing: ____ (timestamps)
E7 Reps - requested and signed text attached side by side: yes / no
E8 Non-answers and refusals logged:

Highest-severity trigger fired:
GRADE: Clear / Escalate / Stop
Analyst:            Second reviewer (blind grade):            Counsel:            Date:
Re-run trigger: COINS final rule effective date; any list change before closing; any structural change at the target

How inquiry files go wrong

Most bad files don't fail because an element is missing. They fail because an element is closed on evidence that looks sufficient and isn't. These eight failure modes produce false Clears and false Stops. Each needs a specific check.

1. The Delaware wrapper

• What it looks like: The issuer is a U.S. corporation, the registry search shows nothing abroad, and element 1 reads Clear. The engineering subsidiary is in Shenzhen.
• Why it lies: The Part 850 entity test turns on financial-metric share, not on where the company is incorporated.
• Check: Ask for opex broken down by legal entity, then confirm it against office locations and job postings.

2. Name-only list screening

• What it looks like: The screen comes back clean.
• Why it lies: An unlisted affiliate is majority owned by a listed party, which the Consolidated Screening List won't show. Romanised Chinese names also cause misses.
• Check: Screen the owners going up the chain, and search names in native script.

3. Founder status confused with ethnicity or birthplace

• What it looks like: A reviewer flags a founder for their ethnicity or birthplace.
• Why it lies: The definition turns on citizenship or permanent residency, and it excludes U.S. citizens and U.S. permanent residents. Inference produces false positives and an equal-treatment problem.
• Check: Record status from documents, never from inference.

4. Activity mapped from marketing copy

• What it looks like: A "defense AI" pitch deck triggers Stop, or a bland website triggers Clear.
• Why it lies: Positioning is not a technical capability.
• Check: Cite technical evidence (papers, compute disclosures, repos) against the specific subsection of 850.224 or 850.217.

5. The stale screen

• What it looks like: The screen is dated at term sheet, and a listing lands before closing.
• Why it lies: The lists change on their own schedule, and no refresh frequency is established.
• Check: Re-screen within 48 hours of the wire and timestamp both screens. The 48-hour window is my inference, not an established requirement.

6. The diluted rep

• What it looks like: A rep qualified "to the Company's knowledge," or carrying carve-outs, is logged as obtained.
• Why it lies: Treasury's final rule treats a refusal to give contractual assurances as a warning sign, and a diluted rep is a partial refusal. The rule doesn't say exactly what counts as a warning sign, which is why the file should record the facts rather than a characterisation.
• Check: Put the requested text and the signed text side by side. A refused rep is a factor-6 warning sign, not a blank.

7. The miscounted LP exception

• What it looks like: Each vehicle sits under $2 million, so the exception is marked as met.
• Why it lies: The threshold is aggregated across investment and co-investment vehicles.
• Check: Add up every vehicle by name on the cover sheet.

8. The wrong regime

• What it looks like: The file is graded against the six COINS countries before the final rule exists, or it's never re-run once the rule takes effect.
• Why it lies: Only Part 850 is in force until the final rule takes effect, and the proposal changes both the countries and the entity test.
• Check: Date-stamp the regime, and re-run every open or recent file on the effective date.

A non-answer is not a blank field. It is evidence, and it carries a grade.

Keeping the file current

A grade is a dated finding, not a permanent property of the company. It expires when the regime changes, when a list changes before the wire, or when the target's structure changes. Treat every Clear as "Clear as of the inquiry date, under the regime named on line one."

Re-run the affected elements when any of these happens:

• The COINS Act final rule takes effect. Re-run every file you are still relying on, since a file graded Clear under Part 850 can grade differently under six countries and an ownership-plus-control test.
• Signing and closing. Re-screen every entity and owner and timestamp each pass.
• A structural change at the target. A new subsidiary, a new investor with board rights, or a new founder opens elements 1 to 4 again.
• The BIS Affiliates Rule changes status. The rule extends Entity List restrictions to affiliates that are 50% or more owned by a listed party. It was suspended for one year, and sources disagree on whether it has come back or the suspension was extended. Check the BIS Entity List FAQs (Q.46) before relying on either reading, and record which reading you used.

On retention: a filed notification and its supporting documents must be kept for 10 years. No public source sets a retention period for files on deals that were never notified. I keep those for 10 years too, to match the statute of limitations. That is an inference, and the file should say so.