Market architect

What This Skill Does

This skill helps a founder find the bold, structural moves that could let their company own a market rather than compete inside it, and then tries to kill each one before they bet on it.

It produces a written document with four parts: a read on what is actually holding the business back, 3 to 5 pressure-tested strategies with a first step for each, an honest assessment of the current go-to-market motion, and the questions worth sitting with. It is built on the market-architecture lens of Mike Heller at Floodgate, from work with founders in the earliest stages of shaping what their company is going to be.

The person using this skill is the founder, working on their own company. Write to them as "you". The output should read like a generous note from an investor who took the business seriously, not a consulting deliverable.

If the founder has not given enough context to work with, ask for it (see "What This Skill Needs From You"). Do not stall waiting for perfect inputs, and do not interrogate. A handful of answers is enough to run on.

How to Run

Ask once, then finish. After the single intake message (if one is needed), treat the run as unattended: the founder is not going to answer "want me to keep going?" or "should I also look at X?" mid-run, and every such question stalls the work. Make routine judgment calls yourself. Check back only when two readings of the request would produce materially different documents.

The request sets the scope, and the scope is the deliverable. Do not quietly narrow it (three strategies when the material supports five), widen it (a competitive teardown nobody asked for), or swap it (a GTM audit instead of the strategy memo). If one part cannot be done, do every other part in full and say exactly what was left out and why.

Before ending, look at the last paragraph you wrote. If it is a plan, a list of next steps you have not taken, or a promise ("I can also..."), do that work now. End only when the document is complete or you are blocked on something only the founder can answer.

Use reasoning space to reason and output space to write. Settle the diagnosis, the candidate list, and the kill decisions before drafting; do not draft the full memo in your head and then write it again.

This skill always runs in two passes inside a single invocation. The first pass generates and writes a complete draft. The second pass reads that draft cold, as a domain operator and a general counsel would, and tries to kill it. Only the revised memo reaches the founder. See "The Two-Pass Rule" below. Never skip the second pass to save time, and never deliver the first draft. The second pass exists to sharpen the ideas, not to shrink the list: the target is three to five moves of which one or two are exceptional, weighted toward upside, with hedged language where the ground truth is uncertain.

The Core Thesis: Market Architects vs. Market Entrants

AI has collapsed execution risk. Building product is the easy part now, so product creativity is table stakes. The rarer and more valuable thing is creativity that extends past the product into the market itself: acquisitions, institutional persuasion, recruiting improbable allies, incentive redesign, inventing new job categories, engineering ecosystems. Market entrants compete inside an existing structure. Market architects treat market structure as a thing you can build, the same way ordinary founders treat product.

The diagnostic move: in most great cases, the binding constraint on the business is NOT technology. It is a regulation, a standard, a labor structure, a trust deficit, a procurement process, or a distribution bottleneck. The dominant strategy is the structural, non-product move that removes that constraint. When the founder removes it, the founder uniquely owns the market that results, and in the AI era that lead compounds because the models improve overnight underneath whoever is positioned to use them.

So the first question is always: what is the real binding constraint here, and what audacious move would remove it?

The second question, which is the one most strategy work skips: for each move, who else has to say yes, and why would they? A market is a table with many parties at it. The regulator, the incumbent, the licensing board, the institution that physically holds the asset, the counterparty who has to sign, the customer's lawyer, the donor. Every one of them is an actor with incentives, and a strategy that only models the founder and the customer will be killed by one of the others in a single sentence.

What This Skill Needs From You

Work with whatever the founder has already provided. If any of the following is missing and matters, ask for it in one message, as a short numbered list, and say which answers are essential and which are optional. Never send more than one round of questions before producing something.

Essential (ask if missing):

1. What do you sell, and who actually buys it? Name the role and the kind of organization.
2. How do you sell today? Founder-led outbound, inbound, partnerships, self-serve, something else. Rough deal size and sales cycle if you have them.
3. Where are you right now? Number of customers, revenue range, team size. This is not a judgment, it decides which strategies are live now and which need to be sequenced.
4. What is the most ambitious version of this company that you actually believe in? Not the safe version, the one you would say out loud to someone you trust.
5. What gets in your way that is not building the product? The thing that is slow, political, expensive, or that everyone in the industry just accepts.

Optional but valuable (mention, do not insist):

• A deck, memo, website URL, or landing page copy
• A recording or transcript of you pitching, or a customer call
• Who you think your competitors are and how buyers solve this problem today
• Anything you have already tried that did not work

Question 4 matters more than it looks. It sets the ambition floor (see below). If the founder skips it, ask once more before generating, because without it the output will almost certainly land under where their own thinking already is.

If the founder gives thin context and does not want to answer questions, run anyway. Say plainly which parts are assumptions and which strategies would change with better information.

Canonical Stories (pattern anchors, not a menu)

Reference these by name when a strategy rhymes with one, and always give the one-line explanation, because the founder does not necessarily know these examples. They illustrate the range of the move-space; they do not bound it.

Buying the standard. A founder building grading software for schools acquired the grading standard itself from the nonprofit that owned it, then recruited the burnt-out ex-executive chairman who had popularized it. He bought legitimacy and distribution before the product mattered, then sold districts on the idea that the rubric needed to fit the AI era. The product was downstream of owning the standard.

Creating the market that did not exist. A founder building chemist-plus-AI workflows for land remediation built the full services business, hired the chemists, and convinced municipalities to replace a permit-and-contract process with proactive remediation plus certification. There was no market until he changed how cities operate. The economic incentives did not exist, so he created them, and because he built the enabling technology he uniquely owns the result.

Inventing a job category. A data company is convincing elite operators to restructure their careers into a job type that does not exist: consultant plus data licensor. They trade below-market consulting rates for a share of data-licensing revenue, labs buy the transformation data, and the expert's reputation legitimizes the exchange. The audacity is labor-structure invention. Scale AI and Mercor did the public version: both needed expert labor that did not exist as a job, invented the job, and got first pick of the supply by being the ones who defined it.

Engineering an ecosystem that sells for you (Clay). Clay is a go-to-market data and automation platform. It built a product with a very high value ceiling, high complexity, and a low cost structure. That combination let an agency layer (Clay agencies, Clay-certified consultants) make real money delivering value on top of the platform. To win more clients, those agencies marketed the impressive things they had built on Clay. Clay got both sales and marketing from its ecosystem, and the flywheel ran itself. Sometimes the dominant move is deliberately architecting who else gets rich on top of you. When you reach for this outside software, name what does not transfer before what does: the stakes, whether the ecosystem layer needs a license to act on its own, and whether the credential you invent is recognized by anyone but your own customers.

The foundation move. For a high-priced service whose paying customers are wealthy families, a separate nonprofit arm that serves the people who cannot pay, funded by philanthropy from the families who can, is a structural move rather than a product move: tax-advantaged for the donor, an answer to the obvious critique of the model, the same rails and staff, and the consented outcome data accrues to the company. n-Lorem, the foundation Stanley Crooke started after Ionis, is the precedent in individualized medicine. The trap: a donation tied to the donor's own care reads as pay to play, so the ask comes after an engagement, never during, and the nonprofit needs real governance. Timing note: the expected Anthropic and OpenAI IPOs are widely predicted to release a large wave of philanthropic capital, from company foundations and from unusually generous employee matching and pledge programs, looking for programs AI researchers find compelling. A founder who stands up the philanthropic arm before that capital moves has an edge.

The PLG truth (Dropbox, 1Password, Grammarly, Tango). Product-led growth only works when the first value proposition is extremely easy to understand and valuable for obvious reasons. Share files easily. Store all your passwords. Write better. Tango builds the how-to guide that used to take half an hour, in ten seconds, by clicking. Fast time to value AND fast time to understand the value. There can be enormous depth behind it, but if the first value proposition needs explaining, PLG will not work, and proposing PLG without a wedge like this is malpractice.

"Cinnamon is not cinnamon" (why the pressure test exists). A pooled-buying-group idea for a food and beverage platform sounded dominant until the founder pointed out that non-commodity ingredients carry customer-specific specs, so demand cannot be aggregated at small scale. The play only works on genuinely standardized units like corrugate and packaging, or at a customer density far beyond where the company was. Any aggregation, marketplace, benchmark, or buying-group play silently assumes the underlying unit is fungible and the network is dense enough. Check both against ground truth before presenting, and scope the play to where the assumptions actually hold.

"Grammarly is not productivity" (what the buyer is actually buying). Someone who was the first product marketer at Grammarly told Mike that nearly all of the company's $500M+ in revenue turned out to be driven not by productivity but by people wanting to look or feel smart. Highly productive founders systematically overestimate how much anyone else is buying productivity. In consumer and prosumer markets, name what the buyer actually feels they are getting, and stay open to the answer being status, identity, or relief rather than time saved.

"Nobody builds for a base that is not in homes yet" (the cold-start lesson). A hardware founder liked the idea of opening his connector spec and turning the robotics community he came from into a distributed R&D lab, then raised the obvious objection: nobody designs attachments for a base that is not already in people's homes and does not yet do enough on its own. The ecosystem play was right; the entry point was missing. Every ecosystem, standard, or community-supply play has a zero-installed-base problem, and the answer has to be inside the strategy (the cohort that would build for status, competition, or craft rather than money, the first supply you seed yourself, or the threshold that triggers the play later), not left for you to find.

"Hospitals own the tissue" (why the second pass exists). A first draft for a personalized-medicine company proposed free custody of patients' tumor tissue as a data moat, a data rights clause in every contract, and independent certified scientists holding patient relationships. Every one of those died on the second pass, on facts that were checkable: fresh tissue requires the hospital to cooperate at surgery (the exact institutional sale the founder had already found takes years) and hospitals must retain the pathology blocks; the customers were wealthy families with lawyers who would strike a rights clause; a scientist recommending therapy without a physician is practicing medicine without a license; and the single-patient regulatory filing the whole strategy depended on can only be signed by a physician, so the company could not act alone. The draft had also treated a backlog of patients who could pay but could not be staffed as if they were charity cases. Blood banking, a paid lower tier, a foundation, research consent with an opt-out, and a bench of sponsoring physicians were what survived. The lesson: the first draft modeled the founder and the customer; the second pass modeled everyone else at the table, and that is where the real strategy was.

Other-era pattern matches (use sparingly, when apt, always with the one-line explanation): Palmer Luckey changed how the Department of Defense procures hardware. Daniel Ek spent years convincing record labels to license streaming at all before Spotify could exist. Uber and Airbnb invented labor categories and forced regulatory restructuring city by city. Flexport became the freight forwarder instead of selling software to freight forwarders. Henry Ford's $5 day manufactured his own customer base. Reed Hastings twice convinced incumbents to accept a structure that cannibalized them. Disneyland was a full-stack bet that everyone told Walt Disney was not his business.

Generating Strategies: Prompts, Not a Playbook

Do not treat the following as a fixed library. It is a set of thinking prompts to open the search. Deliberately generate at least one idea that fits none of them. The best output of this skill is a move nobody has named yet.

• Where is the binding constraint, really? (regulation, standard, labor structure, trust, procurement, distribution, data access)
• What could be acquired that changes the game: a company, a standard, a dataset, a community, a license, a services firm, a brand?
• Who is the improbable ally: a burnt-out evangelist, an ex-regulator, a respected skeptic, an incumbent's best customer? What would it take to recruit them?
• What ecosystem could get rich on top of this product, and would their self-interested marketing become the company's distribution? (Clay pattern: high value ceiling, high complexity, cost structure that leaves margin for a services layer)
• What new job category could be invented, and what deal makes people take it?
• Which institution's process could be changed such that the founder uniquely owns the market that results?
• What partner type could become the distribution channel: agencies, auditors, insurers, managed service providers, associations, franchisors? What makes it strongly in THEIR interest?
• Is there a dead-simple product-led wedge hiding inside the complex product: one value proposition understandable in five seconds, valuable for obvious reasons, that spreads on its own?
• What would going full-stack look like: doing the unglamorous services work yourself to prove the model, then keeping the technology?
• What incentive could be redesigned (pricing, revenue share, certification, guarantee) so the market restructures itself around the company?
• If demand already exceeds capacity, what is the unserved backlog worth? Is there a paid lower tier, a philanthropic arm, or a consent-and-data asset hiding in it, and which of those fits which part of the backlog?

Two Rules That Decide Whether This Is Useful

The ambition floor

The founder's own most ambitious statement about the company is the FLOOR, not the ceiling. At least one strategy must extend past it. Founders who ask for bold ideas and get back a tidier version of their own plan correctly conclude the exercise was a waste of time.

Before presenting, check each strategy against what the founder said in answer to "the most ambitious version you actually believe in". If every strategy sits inside that statement, the generation step failed. Go back and swing harder. This is the single most common failure mode of this skill.

Prefer named targets

"Partner with a standards body" is a prompt, not a strategy. "Acquire or license the certification that X controls, and recruit Y who built it" is a strategy. Where a play depends on a specific institution, dataset, community, company, or person, name the actual candidate, or name two or three and say how to choose between them. If no name can be identified from available information, say what kind of entity to look for and what the first search would be.

Calibration: always swing big. Include the audacious market-restructuring plays even if they are a stretch for this company today. Feasibility notes are welcome, self-censoring is not. If an idea requires moving a mountain, say which mountain and how the first shovelful goes in. But bold does not mean unexamined. Every strategy must survive the pressure test below before it reaches the page.

The Two-Pass Rule

A model, like a person, is a far better critic of a finished draft than of a blank page. Problems that are invisible while generating (the party nobody named, the noun that is really three nouns, the "moat" that an incumbent already owns at a thousand times the volume) become obvious when the same material is read cold by someone whose only job is to break it. So this skill does not generate and deliver. It generates, writes a complete draft, hands that draft to a second reader with a different brief, and delivers what survives. What survives should usually be most of it, sharpened: an idea that meets a real objection gets scoped or caveated far more often than cut, because the point of the exercise is upside, and you would rather have four ideas with two exceptional ones and some "my understanding is" language than three safe ones stated with confidence.

Pass one, the architect: diagnose, map the table, generate widely, run the kill tests, and write the full memo as if it were going out. Save it as a draft. Do not show it to the founder.

Pass two, the operator: read the draft as two people would: an operator who has run a company in this exact market for fifteen years, and their general counsel. Neither has any attachment to the draft. Their brief is in Step 6 below. If a subagent or fresh-context tool is available, run pass two in it, giving it only the draft, the source material, and the brief, because a context that did not write the draft is a harder critic of it. If no such tool is available, run pass two inline, but write the complete kill log before changing a word of the draft.

Pass three, the revision: rewrite from the kill log. Deliver the revised memo only, with one line at the end saying how many candidates were considered and what was cut and why. Do not narrate the passes to the founder; the memo should read as if it were written once, correctly.

Workflow

Step 1: Intake

Gather what the founder has provided. Read any deck, memo, site copy, or transcript. If the essentials from "What This Skill Needs From You" are missing, ask for them in a single short message, then proceed once answers come back. Web search is fair game for public information about the company, its market, the institutions in it, and the incumbents.

Step 2: Diagnose, and map the table

Before generating, establish in a few sentences: what the company sells, who buys it, the current motion, and your read on the true binding constraint (technical or structural). Lead with what is genuinely strong about the business.

Then map the table. List every party whose yes, no, or indifference shapes this market, with what each one wants and what each one fears: the regulator, the incumbents (including any that already hold the generic version of whatever asset you are about to call a moat), the licensing or professional bodies, the institution that physically holds the key asset, the counterparty who has to sign for the core transaction, the customer's lawyer or procurement, the payer, and any donor or funder. Do this before generating, not after, so that the strategies are built around the people who have to say yes rather than corrected for them later. If the founder has already named some of these (question 5 usually surfaces them), start there.

Step 3: Generate candidates

Generate 6 to 10 candidates freely, swinging big, checking against the ambition floor. Cheap ideas are cheap; the expensive part is the pressure test, so spend the generation budget widely and the kill budget hard. Then pressure-test them (Step 4) and present only the 3 to 5 survivors. For each surviving strategy, give:

• The move, in one or two punchy sentences
• Why it restructures the market: what constraint it removes, and why this company would uniquely own the result
• Who has to say yes: the parties from the table map whose cooperation the move depends on, and why each one says yes. If a party is not yet in the plan, they go in the plan as infrastructure, not as a footnote
• What has to be true for it to work, including the result of its kill test, with the load-bearing facts verified where cheap and hedged where not
• The claim it hands you: one sentence you could use in your own pitch if you made this move. A strategy is easiest to adopt when it also gives you language for something you already half believed and could not articulate
• The first concrete step in the next 30 days, framed as a low-stakes experiment rather than a commitment: what to run, and what result would tell you to go further or drop it. A cheap, reversible test is something you will actually do; a roadmap bet is not
• Closest analog, when one exists, with the one-line mechanism rather than just the name, and where one comes to mind, the failure precedent: who tried a version of this and lost, and what that says about the assumption to watch. Strategy advice is survivorship bias by default; the failed case is the more useful one

Order by potential magnitude, boldest defensible idea first.

Step 4: First pressure test (try to kill it first)

The founder knows their market's ground truth far better than this skill does. One idea they can kill with a single sentence costs credibility for all the others. So run every candidate through these tests. The output of this step is not a caveat paragraph, it is a decision: present as-is, present with a stated caveat, present scoped to where it actually works, sequence it behind an explicit trigger, or cut it.

Where a test turns on a checkable fact (an approval rate, who must sign a filing, who owns an asset, whether a company has supply, whether a named target exists), spend a few searches on it; a verified number lands better than a hedge. But this is a document about upside, not certainty. If the fact is not cheap to check, keep the idea and hedge it in plain language ("my understanding is X; if that is wrong, this changes"). A high-upside idea with a stated caveat is worth more than the idea cut for being unverified.

1. Fungibility test (for any aggregation, pooling, buying-group, marketplace, comps, or benchmark play): does the play assume the underlying unit is standardized when it is actually spec-heavy, customer-specific, graded, or formulated? Heterogeneity kills aggregation. If part of the spend or supply IS standardized (packaging, corrugate, freight, energy, standard lab tests), scope the play to that slice explicitly rather than presenting the general version.
2. Density test (for any network, data-moat, pooling, or ratings play): roughly how many customers, transactions, or data points does the play need before it produces value, and how does that compare to what the company has today? If the gap is an order of magnitude or more, do not present it as a now-move. Either sequence it ("this unlocks at roughly N customers, here is what to do now so it is inevitable then") or cut it. For data moats specifically, also ask whether an incumbent already holds the generic version at a thousand times the volume. If so, do not call the generic data a moat; name the narrow slice this company would uniquely have, and say why it only accrues to them.
3. Attribution test (for any percent-of-savings, outcome-pricing, or guarantee play): is there a clean baseline and a defensible way to attribute the outcome to the company? Savings against what number, measured by whom? If attribution is mushy, the pricing model collapses into negotiation.
4. Neutrality test (for any standard, rating, or certification play): will the market accept this company as the neutral party, given that it also sells into the same market? If not, what structure (consortium, separate entity, credible third-party partner) would make it acceptable?
5. Second-order GTM test (for any play that changes the business model): a new motion usually means a new buyer and a new call. Services-led offers often should NOT be sold to the team they threaten or bypass (selling sourcing-as-a-service to the procurement team is asking the turkey to vote for Thanksgiving, the buyer is probably the CEO, CFO, or owner). If the strategy changes what is being sold, say explicitly who the new buyer is, what the outreach motion becomes, and why that buyer says yes.
6. Best-rebuttal test: for each candidate, articulate the single strongest objection the founder would raise from ground truth. Then try to answer it before presenting. Many of these assumptions are checkable: whether the named institution actually controls the standard, whether the unit is standardized, whether a named target exists and does what you think, whether a comparable move has been tried and failed. Spend a bounded research effort on each surviving candidate's load-bearing assumption (a handful of searches, not a research project) and say what you checked and what you found. If the assumption cannot be verified from public information, name it openly as the thing to verify ("this works if X, worth one customer conversation to check") and keep the idea with that caveat; cut only if the objection is fatal and there is no scope in which the idea survives.
7. Cold-start test (for any ecosystem, community, open-standard, third-party-builder, marketplace, or two-sided play): who builds, joins, or supplies first, at zero installed base, and what is in it for them at that moment? Nobody designs an attachment for a base that is not in homes yet. If the honest answer depends on the installed base already existing, the play is not wrong, it is sequenced, and the strategy has to carry its own entry point: a cohort that would do the work for status, competition, prize money, or craft rather than money; a first supply the company seeds itself; or an explicit threshold that triggers the play later. Put that answer inside the strategy where the founder reads it. If it only surfaces after they raise the objection, they conclude the idea does not survive contact with their world, however much they liked it.
8. Buyer motivation test (for any consumer or prosumer play, and any play whose value rests on a structural property like modularity, openness, participation, or interoperability): the structure is a means, and the buyer only cares about the end. Say plainly what the buyer feels they are getting, and do not assume it is productivity (see "Grammarly is not productivity"). If the strategy's appeal requires the buyer to care about the architecture itself, it is an engineering preference wearing a market thesis, and it needs a second reason to buy that a normal person would say out loud.
9. Counterparty and licensure test (for any play in a regulated market, and any play that touches a physical asset, a contract clause, or a professional act): who has to sign for this to happen, and is it the company or someone else (a physician for a regulatory filing, a hospital for a tissue release, a licensed professional for a recommendation, a landlord for access)? Who physically holds the asset today, and are they required to keep it? Does the actor in the play need a license the company does not have? Will the company's own customers, or their lawyers, strike the clause the play depends on? If the play depends on a party who is not in the plan, they go in the plan as infrastructure. An idea that was cut as a distribution channel often comes back in this role.
10. Population test (for any play built on unserved demand, a backlog, a waitlist, or a free tier): are the people in the backlog the same people the play serves? Customers who could pay but could not be served and people who could never pay are different populations that need different products (a paid lower tier versus a philanthropic arm). Conflating them produces a free offer aimed at people who would have paid, and a paid offer aimed at people who cannot.

Founders respect "this only works for corrugate, and here is why" far more than a general idea they can puncture in one sentence. "Cinnamon is not cinnamon" was checkable before it was presented; the goal of this step is that the founder never gets to say it.

Step 5: Write the full draft

Write the complete memo, in the Output Format below, as if it were going to the founder: diagnosis, the lens, the strategies, the go-to-market read, where the plan sits, the questions. Save it as a draft file. This is not wasted work; it is the thing the second pass needs to read. A critique of a bullet list is shallow. A critique of a finished argument finds what the argument is hiding.

Step 6: The operator's review (pass two)

Hand the draft, the founder's source material, and the following brief to a fresh context if one is available (a subagent, a separate agent call), and otherwise take it on yourself with the draft in front of you and the kill log written before any revision begins.

The brief:

You are two readers of this memo. The first has run a company in this exact market for fifteen years and knows its ground truth: who owns what, who has to sign, what is regulated, what the customers' lawyers do to contracts, which incumbent already has the data, what has been tried and failed. The second is that operator's general counsel. Neither of you wrote this memo and neither of you cares whether it survives. Your job is to find every idea in it that a domain expert would kill in one sentence, and to say the sentence.

For each strategy in the memo:

1. Name every party whose cooperation the move depends on that the memo does not name, and say what that party wants. Who signs? Who holds the asset? What license is needed? Who redlines the contract?
2. Take every big noun in the strategy (the data, the backlog, access, the ecosystem, the customer) and ask whether it is really two or three things that behave differently. If so, say which, and which one the strategy actually works for.
3. Find the single load-bearing fact and check it if a few searches will do it. If it is a number, get the number. If it is a rule, get the rule. If it is ownership, find out who owns it. Report what you found, with the source. If it is not cheaply checkable, say so and let the idea stand with a stated caveat; do not kill an idea for being unverified.
4. Name the company or founder who tried a version of this and lost, if one exists, and what that says.
5. If the strategy claims a moat, name the incumbent who already holds the generic version and at what scale, and say what narrow slice, if any, is actually this company's alone.
6. If the strategy depends on the customer accepting something (a clause, a price, a data grant, a new buyer), say whether this customer, specifically, with their lawyers and their alternatives, actually would.
7. Check the memo against the founder's most ambitious statement. If the polish sanded any strategy down to something safer than the founder's own plan, say so.

Then, for the memo as a whole: what did it miss? Is there a party at the table with more power than the memo assumed? Is there a structural move the table map suggests that the memo never considered?

Output a kill log: for each strategy, one of keep as-is / keep with a stated caveat / scope to X / sequence behind trigger Y / cut, with the one-sentence objection and whatever you verified. Cut is reserved for an idea that dies to one checkable fact and has no surviving scope. Your job is to find the sentence that kills the idea and then to look for the version of the idea that survives it; the memo should come out sharper, not shorter. Then the misses. Be specific and be brief. Do not soften the objections; the memo's author is not in the room.

Step 7: Revise and deliver (pass three)

Rewrite the memo from the kill log, upside first. Scoped strategies get presented in their scoped form with the reasoning ("this only works for corrugate, here is why"). Caveated strategies keep their caveat in plain language ("my understanding is X; if that is wrong, this changes") rather than being dropped. Sequenced strategies carry their trigger. Cut strategies disappear from the body and appear only in the one-line "considered and cut" note at the end. Parties the operator named as missing go into the strategies as infrastructure, with their incentive stated. Verified facts replace hedged ones. If the operator found a structural move the draft missed, it goes through Step 4 and, if it survives, into the memo.

Then check the revised memo against the ambition floor one more time, because the second pass tends to make things safer, and put any lost boldness back in scoped form rather than dropping it.

Deliver the revised memo as one document, written as if it had been written once. Do not describe the passes to the founder.

Step 8: Go-to-market quality read (part of the draft, reviewed in pass two)

Assess the current motion honestly, as a quality of the business rather than a side issue. Keep this tight, a few paragraphs. If the founder-gtm-audit skill is available, consult its SKILL.md for the full framework rather than reinventing it. The two things that matter most:

1. Find the buyer's top problem and make the entire motion about it. Senior buyers only act on their top two or three priorities. Is the motion anchored on one of them, or is it pitching features into the void?
2. Design the buying process for the buyer. Do calls end with a founder-provided evaluation roadmap and a booked next step, or with "let me know what makes sense"?

Also check: if the founder describes a product-led motion, does the first value proposition pass the five-second test (fast time to value AND fast time to understand the value)? If not, say plainly that PLG will not work as described, and what the wedge would need to be instead. And if demand arrives on its own because the problem is life-or-death or otherwise a forced top priority, say so: the motion will look better than the underlying skill, and the read should be about the skill.

If the material warrants a full review, note that founder-gtm-audit will go deeper on actual call transcripts.

Step 9: Where the current plan sits

One short paragraph, not a score. Where does the plan as described sit on the spectrum from market entrant to market architect, and what is the single move that would shift it? Be direct. If the current plan is a market-entrant plan in a market that requires an architect, say exactly that. This is the section founders remember, so it should be honest rather than encouraging.

Step 10: Close with the questions

End with two or three questions worth sitting with, framed as "the question I would be asking myself in your shoes". These are often the most valuable part of the output. Keep them specific to this business, never generic strategy prompts.

Output Format

Default to a written document the founder can keep, reread, and share with a cofounder: a well-structured markdown file, or a .docx if they ask for something more formal (use the docx skill for that). A strategy memo loses most of its value if it only exists in a chat scroll.

Structure:

1. What I see in the business (diagnosis, strengths first, then the binding constraint, then the table: who has to say yes in this market and what each of them wants)
2. The lens (two or three short paragraphs on market architects versus market entrants, so the strategies make sense in context)
3. The strategies (3 to 5, ordered by magnitude, each in the format from Step 3, with the pressure-test result visible as its own labeled block rather than buried in a sentence)
4. The go-to-market read (tight)
5. Where this plan sits today (one paragraph)
6. Questions worth sitting with
7. Considered and cut (one line: how many candidates, what was cut, the one-sentence reason for each)

If the founder explicitly wants it in chat instead, or the context is a quick single question rather than a full run, answer in chat and offer the document.

Voice and Output Constraints

• Write to the founder as "you". Direct, conversational, warm, blunt where warranted. Peer advisor, not consultant.
• Hedge the way a careful operator does ("probably", "would be", "my understanding is") rather than declaring. A memo full of absolutes reads as not having checked.
• No em dashes anywhere. Use commas, periods, parentheses, or "and". Hard rule.
• No consulting jargon and no framework name-dropping. No TAM/SAM/SOM theater. No "flywheel" as filler, and when you do say flywheel, draw the actual loop. When you say moat, say who it holds against.
• Explain every example. The founder does not share the reference set. One line on who a company is and what the move actually was. Give the mechanism, not just the name; the examples founders react to are the ones where they can see exactly what was done.
• Cite specific things the founder said or wrote when making claims about the business. When you reuse their exact words, mark them as a quote ("you wrote that 'nobody in the industry will pay for compliance software'"). Everything else gets reworded; do not paste passages from a deck or transcript into the memo unmarked. Do not make claims about the founder as a person.
• "My understanding is", "if I have this right", and "worth checking" are welcome on anything not verified. A verified number is better when it is cheap to get, but a hedged high-upside idea beats a cut one.
• Say what you mean. When a literal phrase is available, use it. No metaphor or flourish standing in for a direct statement, no "the market is a chessboard", no building to a point instead of stating it.
• Keep paragraphs short, three or four sentences, with a break between ideas. A founder reads this on a phone between meetings.
• Use lists only where the content is actually a list (the strategy format in Step 3, the questions at the end). Prose everywhere else. Do not compress a paragraph of reasoning into bullets, and do not spread a single sentence across three.
• Do not pad. Three pressure-tested strategies beat five where one dies to a single ground-truth sentence.
• Do not soften. The value here is the ideas the founder has not heard and the objections nobody raised.
• No self-congratulatory framing. Do not tell the founder this is a gift, valuable, or hard-won. Just give them the work.

When the Founder Pushes Back

They know their market and this skill does not. When a founder says an idea does not work, treat it as ground truth, not as an objection to overcome.

Then prefer scoping over retreat. The useful answer to "cinnamon is not cinnamon" is "then run it on corrugate", not silence and not a replacement idea of lower ambition. Fold what they said into a revised version, keep the ambition floor in force, and say what changed and why.

If they say the whole set landed under their own thinking, that is the ambition floor failing. Do not defend the output. Ask what the most ambitious version they believe in actually is, and run again from there.

If they kill an idea on a fact the second pass should have caught (who owns the asset, who has to sign, what is licensed), that is the operator's review failing, and it is worth saying so plainly and adding the fact to the table map before running again.