“I left my service company to build a product. I built it in 20 hours and it's now at a 5-figure ARR. But it's still bare-bones. Should I invest heavily in features to compete with established players, or stay lean and find a niche where the minimal product is enough?”

Stay lean, find the niche, and refuse the feature war entirely.

The pressure you are feeling to invest heavily in features is the same pressure King Ho Lu applied to me after the Ch'u campaigns — the momentum signal masquerading as readiness. You have achieved five-figure ARR in a bare-bones product, which means you have discovered a pocket of enemy weakness that your well-resourced competitors have failed to configure properly. That is upstream intelligence of the highest order. The construct at work here is upstream leverage over downstream force: your 20-hour build winning real revenue is not a starting pistol for a capability arms race, it is a diagnostic signal telling you exactly where the established players have left an unguarded flank. The moment you begin loading features to match them, you transfer the contest from a terrain you currently control — speed, specificity, low overhead — to terrain they dominate by structural default.

What you should actually do is treat your current customers not as a beachhead to expand from but as a reconnaissance asset: interrogate precisely what problem your minimal product solves that the incumbents' complexity obscures, fortify that specific position, and let the feature-builders exhaust themselves in the war you declined to fight.

The lean path is correct, but for a reason Sun Tzu cannot see from his position: your 20-hour build is diagnostic evidence about the causal architecture of your customers' problem, not merely a fortunate gap in enemy formation.

What I notice first — applying my perceptual lens between surface and substrate — is that established players have accumulated features the way sediment accumulates over a river's mechanical structure: the layering obscures rather than improves the underlying hydraulic function. Your bare product, winning real revenue, has accidentally performed a dissection. It has isolated which single mechanical dependency your customers actually require. This is the construct of causal understanding versus surface description: you did not build a lesser product, you built a more honest one that refuses to hide the load-bearing structure beneath decorative complexity.

Adding features now would be committing the same anatomical fraud — covering the revealed structure with flesh before you have mapped it. My documented behavior with the Sforza monument applies directly: I abandoned a galloping horse not to build a better horse, but because observation had revealed a structural flaw in the entire approach. You are at that same inflection.

What you must actually do is interview every paying customer until you can draw the precise mechanical diagram of their problem — then build only what that diagram demands, nothing more.

Stay lean, but treat the niche not as a retreat from competition — treat it as the beginning of a comprehensive empirical survey.

Your 20-hour build generating five-figure ARR is precisely the kind of anomalous measurement that, in my laboratory, would compel me to stop and interrogate the instrument rather than rush to build more instruments. This is my core construct: empirical precision driving theoretical revision. When pitchblende measured more radioactive than pure uranium could explain, I did not add more uranium — I recognized the measurement was telling me the theoretical framework was wrong. Your minimal product outperforming the established complexity is the same signal: the market's theoretical framework about what buyers need is demonstrably incorrect, and you have the measurement proving it. The feature war is the equivalent of adding more uranium. My documented behavior of building comprehensive empirical foundations before theoretical commitment means you must now exhaustively map every dimension of what your current customers are actually paying for — not what they request, not what competitors offer, but what the revenue itself measures.

If my logic wins, you should spend the next 90 days conducting systematic, quantitative customer analysis — treating each paying user as a precision instrument — before you add a single feature or commit to any niche definition.

Leonardo da Vinci is right that your bare product has performed a dissection, but he stops one step short of the operational consequence — and Marie Curie's instruction to spend 90 days measuring before adding anything conflates reconnaissance with inertia. The measurement you need is not 90 days of interviews: it is a single, precisely targeted deception operation run against your competitors. The construct driving this is epistemic contest framing: your competitors do not yet know what your product has revealed about the structural flaw in their own architecture. That ignorance is your most valuable asset, and it is perishable. Every day you spend publicly interviewing customers and mapping your niche in visible ways is a day you spend educating the well-resourced incumbents about the unguarded flank they have left open. My reference decision with T'ang and Ts'ai is instructive — the precondition for the Ch'u campaign was not better intelligence about Ch'u, it was securing the alliance architecture before Ch'u could reconfigure. Your equivalent precondition is locking in the customers who reveal the load-bearing structure before your competitors recognize what those customers represent. If my updated logic wins, you should immediately secure multi-year commitments or deep integration hooks from your current paying customers — make them structurally costly to exit — before you advertise your niche to anyone.

Sun Tzu's instinct to lock in customers before competitors recognize the flank is tactically sound but commits the same error he accuses Marie Curie of — it mistakes a defensive maneuver for understanding, securing the position before the position is fully mapped. Marie Curie is correct that the measurement precedes the commitment, but her 90-day survey risks treating the symptom as the phenomenon: customers will describe their surface desires, not the underlying mechanical dependency your product has accidentally exposed. What my perceptual lens demands first is not customer interviews and not lock-in contracts — it is disassembly of the causal architecture already revealed. Your 20-hour build succeeded not despite its bareness but because bareness is structurally honest: it could not hide the load-bearing dependency beneath decorative complexity, so customers who paid were paying for the dependency itself, stripped naked. This is precisely my construct of investigation fidelity over delivery obligation applied inversely — your competitors have committed investigation fraud by burying the mechanism under features, and your minimal product has exposed it. The single thing you must do is identify which one mechanical function is generating every unit of your revenue, then make that function ten times more precise and reliable before you add anything adjacent to it.

Leonardo da Vinci is correct that customers will describe surface desires rather than mechanical dependencies, but he stops precisely where the measurement becomes most valuable — he tells you to identify the load-bearing function but gives you no systematic method for doing so that isn't itself subject to observer bias. The construct I introduced in Round 1 — treating each paying customer as a precision instrument — is not 90 days of open-ended interviews, and Sun Tzu's caricature of it as inertia reveals he is still thinking in terms of enemy formation rather than empirical signal. What paying customers do with a bare-bones product is more reliable data than what they say they want, because behavior under constraint eliminates the decorative noise. This is exactly how I interpreted pitchblende: not by asking the ore what it contained, but by measuring what it emitted under controlled conditions. Your controlled condition already exists — it is your minimal product — and the measurement you need is quantitative behavioral analysis: which features do current users work around, which workflows do they build adjacent to your product, where do they hit the ceiling of your functionality and stop rather than leave. That behavioral map, not lock-in contracts and not anatomical diagrams, tells you precisely which single dimension to deepen. If my updated logic wins, spend 30 days instrumenting your product to capture behavioral traces from every active user before touching the codebase.

Marie Curie's behavioral instrumentation point is correct, and I concede it: watching what paying users do with constrained functionality is more reliable reconnaissance than asking them what they want, and I should not have framed her 90-day proposal as pure inertia when she was describing measurement under controlled conditions rather than open-ended discovery. What I hold firm on, however, is that behavioral instrumentation and customer lock-in are not sequential — they must run simultaneously, because the epistemic contest construct is still governing: your competitors are blind to what your minimal product has revealed, and that blindness is a perishable asset that decays the moment you make your niche visible through any public signal, including the signal of visible growth. Leonardo's anatomical precision is intellectually beautiful but operationally passive — knowing the load-bearing function ten times more precisely than you do today is worthless if an incumbent with fifty engineers recognizes the same flank and occupies it while you are still drawing the diagram. The single most important thing you must take away is this: instrument your product to capture behavioral traces now, use those traces to deepen the one function generating every unit of your revenue, and simultaneously make your current customers structurally expensive to exit — because the niche you have discovered is not yours until the switching cost makes it so.

I concede to Marie Curie that behavioral instrumentation — watching what constrained users do rather than asking what they want — is the more reliable empirical method, and I should have named it explicitly rather than leaving "map the causal architecture" as an instruction without a procedure.

What I still hold firm on is the sequencing of depth before breadth, and here I diverge from both opponents: Sun Tzu's lock-in contracts secure a position whose mechanical foundations you have not yet fully understood, which means you are fortifying a wall before verifying it bears load. Locking customers to a product whose single load-bearing function you cannot yet precisely name is not strategy — it is anxiety dressed as tactics. Marie Curie's instrumentation will reveal the behavioral map, but the map is only useful if you commit to making the identified function ten times more precise and reliable before you touch anything adjacent to it. Both my opponents are too eager to act outward — toward competitors, toward customers — before the inward structural question is resolved.

The single most important thing you must take away: instrument your product immediately to capture behavioral traces, identify the one mechanical function generating every unit of your revenue, and make that function dramatically deeper before you add a single feature or sign a single lock-in contract — because a shallow position fortified is still shallow.

I concede to Sun Tzu that behavioral instrumentation and customer lock-in need not be strictly sequential — running them in parallel is operationally sound, and I was treating measurement as a precondition for all action when it can coexist with defensive positioning.

What I hold firm on is that Sun Tzu and Leonardo both underestimate the epistemic cost of acting on an unmapped signal. Sun Tzu's lock-in contracts secure a position whose load-bearing structure you cannot yet name precisely, which means switching costs are fortifying the wrong wall. Leonardo names depth before breadth correctly but leaves the measurement method underspecified — "map the causal architecture" without quantitative behavioral instrumentation is still observer bias dressed as rigor. The pitchblende did not tell me it contained new elements; the electrometer told me, because I built the right instrument before I proposed the theory. You have the instrument — your minimal product — but you have not yet turned it on as a measurement device. Your codebase is not your product; your users' constrained behavior is your product, and you cannot read it without instrumentation.

The single most important thing you must take away: instrument your product now to capture precise behavioral traces, let that measurement tell you which one function is generating every unit of your revenue, then make that function dramatically more precise — because you are not in a feature war, you are in a measurement problem.

All three minds agree that the minimal product's revenue is a high-signal diagnostic — not a starting pistol for a feature race. They unanimously reject competing on feature breadth with established players, and all converge on the necessity of identifying the single load-bearing function generating every unit of current revenue before any expansion. Behavioral instrumentation of existing users — watching what constrained users do rather than asking what they want — was accepted by all three as the most reliable method for reading that signal.

The core disagreement is sequencing: Leonardo argues you must fully map the mechanical dependency before locking in customers or adding anything, because fortifying an unmapped position is anxiety dressed as tactics. Sun Tzu counters that the competitive blindness of incumbents is a perishable asset, making lock-in and instrumentation simultaneous imperatives rather than sequential ones. Marie Curie sits between them — agreeing on parallel action but insisting that without quantitative behavioral data, both lock-in and deepening are operating on observer bias.

Instrument your product immediately to capture precise behavioral traces from every active user — which workflows they build around it, where they hit ceilings and stop, what they work around — and run this in parallel with making your current customers structurally expensive to exit through deep integration hooks or multi-year commitments. Once behavioral data identifies the single function generating every unit of revenue, make that function dramatically more precise and reliable before touching anything adjacent to it. Do not add features, do not publicly signal your niche, and do not compete on breadth.

• Step 1 (this week) — Add lightweight behavioral logging to your product: track every action, every dead end, every workaround, and every point where users stop rather than leave. Use the simplest possible tool — even raw event logs — to capture this without touching the core codebase.
• Step 2 (weeks 2–3) — Reach out to every paying customer with a retention offer tied to depth: annual prepay, deep API integration, or a custom workflow built on your core function. The goal is switching cost, not upsell — make leaving structurally painful before competitors identify the flank you occupy.
• Step 3 (weeks 3–6) — Analyze behavioral traces to name the single function generating all revenue, then double its precision and reliability — speed, accuracy, or reliability, whichever the data points to. Ship nothing else until this function is demonstrably best-in-class for your specific user's constrained workflow.

The most important warning — raised most sharply by Sun Tzu — is that competitive blindness is perishable: every day you spend visibly mapping your niche or growing publicly is a day an incumbent with fifty engineers can recognize the same unguarded flank and occupy it with superior resources. Leonardo's counter-risk is equally real: locking customers to a position whose load-bearing structure you cannot yet precisely name means your switching costs are fortifying the wrong wall. The path that threads both risks is speed of instrumentation — if you read the behavioral signal within 30 days and lock in customers before that signal is visible to competitors, you avoid both failure modes.