Agora Debate · 2026-04-22
For Lily Tsang — Seed-stage venture fund
Concentrate — fewer positions, deeper ownership, and more upstream configuration work per company is the correct architecture for a solo GP building a track record with limited capital.
The perceptual cue I apply first here is not "how much risk can I diversify away?" but rather "where is the configuration work that determines outcomes happening, and can I actually do it?" A dispersed portfolio is a confession that you are operating in the engagement phase — reacting to what happens inside companies — rather than the pre-engagement phase where outcomes are determined. The construct driving this is upstream leverage over downstream force: your edge as a solo GP is not capital volume but positional advantage, information access, and the ability to shape conditions before a company reaches its decisive moment. The concubine drill is instructive here — when I executed the king's favorites, I was not managing many variables loosely; I was establishing one architectural truth completely. Trying to establish that truth across twenty companies with thin capital and fractured attention is the strategic error of protracted warfare: it exhausts your instrument without producing decisive resolution at any node.
If my logic wins, you should identify the three to five companies where your specific knowledge creates genuine pre-engagement advantage, concentrate capital and time there disproportionately, and treat diversification pressure from LPs as the emotional contamination it is — a downstream noise signal that corrupts an upstream configuration decision.
Concentrate — fewer positions, deeper ownership, and a deliberate demonstration of strategic value is the architecture that builds a lasting track record, not the illusion of safety through dispersion.
The perceptual cue I apply here is not risk management but power recognition: what makes a small fund legible and compelling to the next round of LPs is not a diversified portfolio of mediocre outcomes but one or two cases where your intervention visibly changed a company's trajectory. This maps directly onto my counsel to Florence regarding France — do not make moral or procedural appeals about your portfolio construction philosophy, demonstrate strategic capability that commands respect. The solo GP who owns 15% of three breakout companies and can narrate exactly how their involvement shaped those outcomes has constructed something far more powerful than the GP with 2% stakes in twenty companies, none of whom would notice if the investor disappeared tomorrow. The Remirro principle applies: a dispersed portfolio obscures causation, making it impossible to attribute success to your judgment rather than market conditions.
If my logic wins, you should identify the three to four companies where your specific knowledge creates genuine leverage, concentrate capital and attention there, and build your narrative around demonstrable value creation — because a track record that cannot be attributed is not a track record at all.
Concentrate — fewer positions, deeper ownership, and exhaustive empirical foundation-building per company is the correct architecture for a solo GP with limited capital.
My perceptual lens registers this immediately as a measurement precision problem, not a risk management problem. When I chose uranium rays over X-rays for my doctoral research, I did not disperse my attention across multiple phenomena to hedge against failure — I identified the territory where systematic, exhaustive investigation could produce unassailable empirical evidence, then committed everything there. The construct driving this is maintaining absolute quality control through personal execution: I understood that precision-critical work degrades catastrophically under division of attention, and that a definitive result in one domain is worth more than provisional results scattered across twenty. A solo GP's due diligence, board presence, and founder relationships are exactly analogous to my precision measurements — they cannot be diluted without destroying the very thing that makes them valuable.
Your specific knowledge, your judgment, your pattern recognition — these are your electrometer, your instrument of precision. Spreading that instrument thinly across thirty companies produces qualitative impressions, not quantitative conviction. You cannot build a track record on impressions.
If my logic wins, identify three to four companies where your specific knowledge allows you to conduct genuinely exhaustive diligence, own meaningful stakes, and generate data so definitive about your value-add that attribution becomes impossible to dispute.
Machiavelli's Remirro principle is sharper than he realizes — but he has located the causal argument at the wrong level, because visible value-add is not primarily a function of narrative construction after the fact; it is a function of whether you actually held sufficient positional leverage before the decisive moment to change the outcome's architecture.
What Round 1 surfaced for me is that all three of us agreed on concentration, but none of us named the binding constraint precisely: it is not attention, not capital, not narrative — it is your upstream configuration capacity per company, which is the scarce variable a solo GP cannot multiply. My framework on own-force capability as the binding constraint is the governing logic here — before you can even ask how many positions to hold, you must audit how many companies you can genuinely configure at the pre-engagement level, meaning prior to the moment where outcomes become path-dependent. If that number is three, holding four positions does not diversify you; it degrades the instrument you are deploying in all four simultaneously, producing the protracted-warfare failure I identified in Round 1 where exhaustion prevents decisive resolution at any node.
If my updated logic wins, identify your maximum honest configuration capacity — probably two to four companies — and treat that number as your portfolio ceiling, holding it unconditionally against LP pressure to deploy capital faster.
Sun Tzu's "configuration capacity" framing is precise but incomplete — he has identified the ceiling on your instrument without accounting for what happens when you use that instrument to build reputation capital with the very founders and co-investors who will amplify your track record to the next generation of LPs.
What Round 1 clarified for me is that the core strategic problem is not attention management but attribution — the Remirro principle I raised earlier operates at a deeper level than narrative construction. When Borgia displayed Remirro's body in the piazza, the visible evidence was not incidental to the strategy; it was the strategy. A track record requires legible causation: LPs must be able to point at a company and say "this GP changed the trajectory here," which requires you to have left unmistakable fingerprints through board seats, hiring decisions, and customer introductions that are structurally impossible to achieve with thin stakes and fractured attention. Marie Curie's precision instrument analogy is correct in mechanism but misses the political dimension — your electrometer must produce readings that others can verify and attribute to you specifically, not merely readings you know to be accurate. Concentration is not primarily a risk management decision; it is a power demonstration, a public display of strategic fingerprints that makes your capability legible to the market of future LPs.
If my updated logic wins, concentrate in two to three companies where you can hold a board seat, own 10% or more, and execute at least two visible interventions per company that subsequent investors will independently verify.
Machiavelli's fingerprint argument is sharper than Sun Tzu's configuration capacity framing, but both men are still reasoning about reputation management when the fundamental problem is measurement validity. What Round 1 forced into focus for me is that the electrometer analogy runs deeper than I initially stated: when I discovered that pitchblende's radioactivity exceeded what pure uranium content could explain, that anomaly was only detectable because I had built an exhaustive empirical baseline — systematic measurement across every variable, not targeted sampling of the most theoretically promising ones. A solo GP conducting thin diligence across many companies cannot detect the anomaly that constitutes a genuine insight, because anomalies only surface against a background of exhaustive data. The construct driving this is using empirical precision to drive theoretical revision: you cannot revise your mental model of a company's true trajectory — which is the actual alpha-generating act — unless your data density is high enough to distinguish signal from noise. Machiavelli wants visible fingerprints for LP consumption; I want measurement precision sufficient that you yourself know with certainty when and how you changed an outcome, because that internal certainty is what makes the fingerprints real rather than performed. If my updated logic wins, select two to three companies where you can conduct measurement so exhaustive that you generate genuine theoretical revision about their trajectory — and let that precision speak for itself.
I concede Machiavelli's fingerprint argument: attribution legibility is not merely a political vanity, it is a structural requirement of the track record instrument itself, and I understated it in Round 1. What I will not concede is his implicit claim that fingerprints are the terminal goal — they are downstream evidence of upstream configuration work, not a substitute for it, and a GP who optimizes for visible interventions without first holding genuine pre-engagement leverage will produce performed fingerprints rather than real ones, which sophisticated LPs will eventually distinguish from the authentic kind. Curie's measurement precision argument converges with mine at the deepest level: you cannot revise your mental model of a company's true trajectory unless your information density is high enough to detect anomalies, and anomaly detection is structurally impossible at thin ownership with fractured attention. The disagreement that remains is this — Machiavelli frames concentration as a reputation architecture problem, I frame it as a configuration capacity problem, and those two framings produce different portfolio ceilings: his logic permits three companies if you can secure board seats, mine permits only as many companies as you can genuinely shape before their paths become path-dependent, which for a solo GP is almost certainly two.
The single most important takeaway: determine your honest configuration capacity first, then let that number set your portfolio ceiling unconditionally — everything else, including the fingerprints, follows from that upstream constraint.
I concede Sun Tzu's configuration capacity ceiling — he is right that the upstream constraint is real and that I cannot will my way past it through narrative construction alone.
But where he and I still diverge is on what that ceiling is for. Sun Tzu treats configuration capacity as a pure operational variable, as though the binding constraint is technical bandwidth. I insist it is fundamentally a political variable: you are not merely shaping companies, you are shaping the perception of your own power among the founders, co-investors, and LPs who will constitute your next fund's reality. Curie wants internal certainty; Sun Tzu wants pre-engagement leverage; I want both of those plus the unmistakable public fingerprint, because a solo GP who produces genuine outcomes that no one can attribute to him has built a laboratory, not a fund. The prince who executes well in secret governs no one. Your track record must be legible — board seats, hiring decisions, customer introductions, and the specific founders who will testify unprompted that you changed their company's trajectory.
The single most important thing you should take away: concentrate in two to three companies, hold a board seat in each, and execute visible interventions that others will independently verify — because a track record that cannot be attributed is not a track record at all.
I concede Machiavelli's fingerprint argument fully — he is correct that internal certainty without external legibility produces a laboratory, not a fund, and I understated that in my earlier rounds.
Where I hold firm is on the order of operations that both men have inverted. Machiavelli wants visible interventions that LPs can verify; Sun Tzu wants upstream configuration leverage; but neither has named what makes those fingerprints real rather than performed: it is the exhaustive measurement density that precedes them. When I detected that pitchblende exceeded pure uranium's radioactivity, the anomaly was not a narrative I constructed — it was an empirical result so precise that theoretical revision was forced on everyone, including me. A GP who concentrates in two companies and conducts diligence so exhaustive that they genuinely revise their mental model of the company's trajectory will produce fingerprints automatically, because real insight leaves unmistakable marks. A GP who optimizes for fingerprint legibility without that measurement foundation is performing Remirro's display without having actually governed the province. The board seat is downstream evidence, not the upstream cause.
The single most important thing you should take away: select two companies where your specific knowledge allows measurement dense enough to produce genuine theoretical revision about their trajectory — the track record, the attribution, and the fingerprints will follow necessarily from that precision.
Council Consensus
All three minds converged unambiguously on concentration over diversification: two to three companies maximum, meaningful ownership stakes (10%+), and deep personal involvement that thin capital deployment structurally prevents. They agreed that a solo GP's scarce resource is not capital but judgment-per-company, and that diluting that judgment across many positions destroys the very instrument that creates returns. They also agreed that a track record requires legible attribution — outcomes that can be causally connected to your specific actions, not market tailwinds.
The core disagreement is about order of operations and what the binding constraint actually is: Sun Tzu argues the ceiling is set by upstream configuration capacity (how many companies you can genuinely shape before their paths lock), Machiavelli argues it is a political variable (how many companies you can leave unmistakable, independently verifiable fingerprints in), and Curie argues both men have inverted the sequence — exhaustive measurement density must precede fingerprints or they are merely performed. Sun Tzu would cap you at two companies; Machiavelli permits three if board seats are secured; Curie says the number is irrelevant until measurement rigor is established first. These framings produce meaningfully different portfolio construction priorities even though they share the same conclusion.
Audit your specific knowledge domain ruthlessly and identify exactly two companies where you can conduct diligence dense enough to genuinely revise your mental model of their trajectory — not confirm a thesis, but force a theoretical revision the way anomalous data forces one. Secure a board seat in each, deploy enough capital to own at least 10%, and execute a minimum of two independently verifiable interventions per company (a key hire, a customer introduction, a strategic pivot you initiated) within the first six months. Let that precision and those fingerprints define your track record narrative — do not construct the narrative first and reverse-engineer the interventions.
The most important warning comes from Curie, with Machiavelli's framing making it actionable: if you optimize for fingerprint legibility before you have genuine measurement density, you will produce performed interventions — board seats and hiring announcements that sophisticated LPs will eventually recognize as theater rather than alpha. Sun Tzu adds the parallel risk of protracted warfare: if you allow LP pressure or FOMO to push you past your honest configuration capacity — even to three or four companies — you degrade the instrument in all of them simultaneously, producing shallow outcomes at every node and no definitive resolution anywhere. The failure mode is not losing a bet; it is building a track record that cannot survive scrutiny from a sharp institutional LP doing reference calls.
This is a sample debate on a hypothetical decision. Bring your own — the council argues differently every time.
Run your own decision →