How the model works.
This is a research-grounded view of how a venture capital firm is built and what every score on the platform actually means, written for the GPs and LPs who use the model in the language they already use.
What the model evaluates.
The Colibrí Architecture model evaluates how a venture capital firm is built. It looks at the firm across three independent dimensions and integrates them into a single firm-level score. A separate companion module supports the follow-on allocation decision when it comes up.
Every fund and firm on the platform is read through the same structured framework, which is calibrated separately to the firm's lifecycle stage.
The Scenario Engine.
The Scenario Engine is a separate decision-support module that helps a GP think through a specific follow-on allocation, including whether the fund should follow on, how much, when, and why. It reads the company's trajectory, the proposed round, the fund's reserve status, and the firm's exposure to the company across every fund.
Three engines, three questions.
The three engines run independently, and no engine's output enters another engine's computation. This separation means that each score can be read on its own terms, and a strong result in one engine cannot mask weakness in another.
Portfolio Efficiency.
Is the fund's deployment of capital configured efficiently for its lifecycle stage?
The engine evaluates five dimensions of how a fund is set up, including whether its check math holds together, whether the ownership it targets is achievable at the stage it intends to invest in, whether its declared industry scope matches the portfolio breadth it wants to run, whether the overall configuration fits the firm's lifecycle position, and whether the investment practice (such as leading rounds and taking board seats) is supportable by the team.
A fund whose choices line up with each other returns a higher score, while a fund whose choices contradict each other returns a lower score. Every output is shown next to the cohort baseline for funds at the same lifecycle stage, so the number always carries the context a GP needs to read it.
Firm Design Congruence.
Do the firm's structural choices fit together?
A firm makes a long list of structural choices when it sets itself up, including how the IC votes, who holds economic rights, whether the firm leads rounds and takes board seats, how large the team is, how broad the geographic mandate is, and how succession is planned. Each of these choices is reasonable on its own. The question this engine asks is whether the choices, taken together, work with each other or against each other.
Eleven independent checks evaluate specific pairs and small groups of structural variables, and each check returns one of three outcomes: no tension, soft tension, or hard tension. The engine aggregates the outcomes into a single score and surfaces every tension it found in a ranked list that the GP can review.
Cross-Fund Concentration.
Is capital distributed across funds in a way that preserves the diversification a multi-fund structure is meant to provide?
A firm with three funds, each diversified within itself but each concentrated in the same sector, has firm-level concentration that no fund-level view can catch. A single portfolio company can also accumulate exposure across funds in a way that was never intended. This engine reads across every fund in the firm and looks at how deployed capital is distributed by sector, by single company, by stage, and by geography.
Sector concentration and single-company exposure are scored, while stage and geographic concentration are surfaced as descriptive analytics rather than flagged outputs. The engine still computes a meaningful view for single-fund firms because the within-fund concentration patterns it surfaces are useful even when there is only one fund to look at.
The eleven firm design checks.
Each check below is a question the engine asks about your firm. The thresholds and rules that decide which outcome each check returns are part of the model's proprietary calibration, but the questions themselves are open to anyone reviewing the methodology. Taken together, they function as a checklist of the structural choices the model considers worth examining.
IC structure and industry scope
Does the investment committee's voting model work given the breadth of the firm's thesis? Consensus voting works well in a narrow thesis where every partner has a defensible view of every deal, but as thesis breadth grows, consensus becomes increasingly difficult to reach.
IC structure and GP rights
Does decision authority align with economic participation? When decision authority sits with one partner but economic rights are fully equal, junior partners bear equal outcomes for decisions they had no weight in.
Board seats and team capacity
Does the firm have enough people to take and serve the board seats its declared practice implies? Board seats are a real post-investment obligation; a team that cannot cover them is structurally overloaded.
Platform scope and team size
Does the firm have the personnel to deliver the platform services it offers portfolio companies? A full platform offering needs dedicated people beyond the investment team.
Geography and team size
Is the geographic mandate supportable by the headcount? A global mandate handled by a two-person team is structurally underresourced regardless of how talented the team is.
Ownership target and lead practice
Is the fund's lead practice consistent with the ownership it intends to acquire? A fund that always leads but targets a small ownership stake is either writing checks below the lead share or following while labeling itself a lead.
Breadth, lead practice, and board capacity
Can the team deliver on the combined demands of portfolio breadth, leading rounds, and board service? A firm that declares many portfolio companies, frequent lead practice, and frequent board service needs a larger team than any one of those commitments alone would suggest.
Stage focus and ownership
Is the ownership stake achievable at the stage the fund is investing in? Earlier-stage rounds support larger ownership stakes at smaller check sizes, while later-stage rounds require larger checks for proportionally smaller stakes.
GP count and IC structure
Does the IC voting model work for the number of partners voting? Consensus with a single GP is degenerate, majority with two GPs is also degenerate, and consensus with many GPs scales poorly because every partner is required to endorse every deal.
GP Load
Is the capital each GP is responsible for deploying per year inside a workable range? When the figure is too low the partnership is underutilized, and when it is too high each partner is writing more checks than they can responsibly source, diligence, and serve.
Lifecycle and succession
Does the firm's succession planning match the stage of its lifecycle? Emerging firms are not expected to have formal succession structures in place, but franchise firms are, because their continued existence depends on transition across partner cohorts.
The Architecture Score.
The Architecture Score is a single firm-level number that integrates the three engines into one composite. It is the headline figure a GP or LP looks at first, and the three engine scores beneath it are the analytical components that an auditor would look at next.
A firm cannot earn a high Architecture Score by being strong on one dimension while weak on another.
The composite uses a method that penalizes imbalance. If a fund is portfolio-efficient but the firm design is incongruent, the Architecture Score reflects the gap rather than averaging it out. The result is a score that means roughly the same thing across every firm on the platform: well-architected on all three dimensions, not on average.
The default view weights the three engines equally. LP-side users can apply custom weights to emphasize a particular dimension in their own views, but every shared and platform-primary view uses the equal-weight default.
The Architecture Score is a diagnostic that tells you how the firm is configured, not whether the firm will succeed.
The model in time.
A score in isolation is a snapshot. The Temporal Layer adds the longitudinal dimension. It records every score the platform produces, tracks the direction of change, and flags configurations that have gone stale relative to operational reality.
Improving, stable, or declining.
Each of the four primary scores (the three engine scores and the Architecture Score) carries a directional marker computed over a rolling ninety-day window. A score with fewer than thirty days of history shows a new indicator instead of a trend, because trends need enough observations to be meaningful.
Suggested, never automatic.
When a firm accumulates enough structural change to warrant a lifecycle transition from emerging to established or from established to franchise, the platform surfaces a suggestion that you can accept, defer, or reject. The platform never auto-transitions a firm between stages on its own.
A time-stamped record.
Every score is snapshotted on input change and on a nightly schedule. Each snapshot carries the lifecycle classification in effect at the time, so historical comparisons across lifecycle transitions remain meaningful instead of looking like sudden discontinuities.
When configuration drifts.
If a fund's configuration hasn't been refreshed in a long time while the fund has continued deploying capital, the platform surfaces a refresh prompt. Staleness doesn't invalidate the scores. It signals that they may not reflect current operating reality.
The Scenario Engine.
The main model evaluates how the firm is built, while the Scenario Engine sits alongside it and supports a different kind of decision. For a specific portfolio company, it helps you think through whether the fund should follow on, how much, when, and why. The engine offers decision-support rather than automation, and you remain the one making the underlying call.
One question, four signals, three minutes.
Every scenario reads the company's current state and trajectory, the proposed round terms, the fund's available reserves, and the firm's existing exposure to the company across every fund. The output is a recommendation, a recommended allocation in dollars and as a share of pro-rata, a timing signal, and a three-part rationale that names what is driving the recommendation. Each run is a point-in-time evaluation against the inputs as they stand on that date.
The engine evaluates follow-on decisions on existing positions only, not new deals. New-deal evaluation is outside the version 1 scope of the platform and will be considered as a future addition rather than as part of the current model.
Every score is calibrated to your lifecycle stage.
Lifecycle is the most important contextual variable in the model. The configuration that fits an emerging manager does not fit a franchise firm, and the configuration that fits a franchise firm does not fit an emerging manager. Every threshold, every band, and every reference point is calibrated separately for each stage, which means that a score of 65 carries one meaning for an emerging firm and a different meaning for a franchise firm.
Emerging
An emerging firm is early in its life, typically operating its first or second fund with a small team and a defined investment thesis. The model expects emerging managers to occupy a frontier that rewards conviction, which means narrower industry scope, smaller portfolio counts at higher ownership, and smaller fund sizes. Concentration of conviction is structurally appropriate at this stage of the firm.
Established
An established firm operates multiple funds, has a working operating cadence in place, and is building a track record across consecutive vintages. The model expects a shift toward measured diversification at this stage, which translates into moderate scope, moderate portfolio counts, fund sizes in the institutional middle band, and the beginning of formal succession work.
Franchise
A franchise firm has scaled into a multi-generational structure with formalized succession and durable operating capacity, typically with multiple funds in active deployment at the same time. The model expects breadth at this stage, which translates into broader scope, larger portfolio counts at moderate ownership, fund sizes in the upper institutional band, and demonstrated transition capacity across partner cohorts.
What the model is, and what it isn't.
The model applies a consistent and principled framework to a firm's configuration, but the value of the output depends on the GP interpreting it in the context of factors the model cannot see. The model is not a substitute for judgment; it is an instrument that supports judgment.
The structured view.
- The architecture of a venture capital firm can be evaluated against a principled framework.
- That framework surfaces tensions and configurations worth examining.
- The outputs are consistent across firms in the same lifecycle stage and support peer comparison.
- The outputs are diagnostic in nature: they tell you something useful about how your firm is set up that you can act on if you choose to.
The boundary.
- The model does not predict fund returns.
- The model does not predict success or failure.
- The model does not produce investment recommendations or investment advice.
- The model does not rate firms as good or bad.
- The model does not capture every factor that determines venture capital outcomes. Market timing, team dynamics, deal pipelines, LP relationships, and many qualitative factors sit outside what the model can read.
Definitions.
This glossary defines every term the model uses in plain language. The italic line under each term is the short version that appears as in-platform hover text, and the paragraph below it is the full definition. Use the in-page anchors or your browser's find function to jump to any term you need.
Scores and outputs
The headline numbers and structured outputs the platform produces for every fund and firm.
How coherently a fund's configuration choices fit together for its lifecycle stage
The Composite Efficiency Index is a 0 to 100 score that measures how coherently a fund's configuration choices fit together for the lifecycle stage the firm has declared. It is produced by the Portfolio Efficiency engine. A higher number means the fund's capital math, ownership targets, industry scope, lifecycle positioning, and team capacity all align with each other, and a lower number means one or more of these dimensions contradicts the others. The score is always shown next to the cohort median and top-quartile for funds at the same lifecycle stage. It is point-in-time and configuration-driven, and it is not a forecast of realized returns.
The fund's expected return shape per unit of dispersion compared to lifecycle peers
Return per Unit of Variance is a ratio that compares the fund's expected return per unit of dispersion against the baseline for funds at the same lifecycle stage. It is produced by the Portfolio Efficiency engine and is displayed as a multiple, for example 1.25x. A value above 1.00x means the fund's configuration is expected to produce more return per unit of dispersion than the lifecycle median, while a value below 1.00x means less. The ratio captures whether the fund is configured to lean into the right-tail outcomes that drive venture returns, and it is not a prediction of realized IRR, TVPI, or any other return metric.
How tightly the fund's expected outcomes cluster around the mean
The Consistency Score is a 0 to 100 score that measures how tightly the fund's expected outcomes are likely to cluster around the mean, given its current configuration. It is produced by the Portfolio Efficiency engine. A higher number indicates a more predictable outcome distribution, and a lower number indicates a configuration where outcomes are more widely dispersed. The score is driven by choices that influence dispersion, including portfolio size, sector breadth, geographic scope, target stage, and reserve allocation. It is read alongside Return per Unit of Variance, and the two together describe the fund's expected return shape.
Whether the firm's structural choices work with each other or against each other
The Firm Design Congruence Score is a 0 to 100 score that measures whether the firm's structural choices work with each other or against each other. It is produced by the Firm Design Congruence engine. A score of 100 means no tensions surfaced across the eleven checks the engine runs. Each tension surfaced reduces the score by an amount based on severity, with hard tensions reducing more than soft tensions. Most real firms produce a score between 60 and 95.
A specific conflict between two or more of the firm's structural choices
A tension is a specific conflict between two or more of the firm's structural choices, surfaced by the Firm Design Congruence engine. Each tension names the choices in conflict and explains in one sentence why they work against each other. Tensions come in two severities: a soft tension means the choices create friction but are not structurally incompatible, while a hard tension means the choices are structurally incompatible and warrant active attention. The platform displays tensions in a ranked list with hard tensions appearing first.
Whether capital is well-distributed across the firm's funds
The Cross-Fund Concentration Score is a 0 to 100 score that measures whether the firm's capital is well-distributed across its funds. It is produced by the Cross-Fund Concentration engine and reads sector concentration, single-company exposure across every fund in the firm, and the firm's deployed capital footprint. A higher number means a well-distributed firm, and a lower number means concentration patterns worth reviewing. The score falls into five interpretive bands that run from well-distributed at the top to concentrated firm structure at the bottom.
A specific cross-fund concentration pattern worth reviewing
A flag is a specific concentration pattern surfaced by the Cross-Fund Concentration engine. Examples include a sector that holds more of the firm's deployed capital than the firm's declared industry scope would imply, or a single portfolio company whose exposure across every fund in the firm has crossed a threshold worth reviewing. Each flag names the dimension (sector, single company, or other) and the magnitude. Flags come in two severities, soft and hard.
The firm-level composite across efficiency, congruence, and concentration
The Architecture Score is the composite firm-level 0 to 100 score that integrates the Portfolio Efficiency, Firm Design Congruence, and Cross-Fund Concentration outputs into a single headline. It is the first number a GP or LP looks at, and the three engine scores are what an auditor looks at next. The score is calibrated to penalize imbalance, which means that a firm cannot earn a high Architecture Score by being strong on one dimension while weak on another. It supports custom weighting for LP-side users who want to emphasize a particular dimension in their own views, but the default view always uses equal weighting across the three engines.
How strongly the evidence supports a right-tail trajectory for the company
The Outlier Probability Score is a 0 to 100 signal from the Scenario Engine that expresses how strongly the available evidence supports the hypothesis that a specific portfolio company is on a right-tail return trajectory. The score reads the company's current MOIC, its mark history and trajectory, its stage progression since the fund's entry, the speed of valuation step-ups, and its current status. A higher number means a stronger right-tail signal. The score drives the Scenario Engine's follow-on recommendation, and it is not a prediction of the multiple the company will eventually return.
Median and top-quartile reference values for funds at the same lifecycle stage
Cohort context is the median and top-quartile reference values for funds at the same lifecycle stage as the focal fund. Every score the platform displays is shown next to its cohort context so a GP can read whether the value sits above, at, or below the typical fund in their lifecycle bucket. Cohort references are recomputed nightly. When a cohort holds fewer than thirty funds, the platform uses provisional reference values until the cohort grows large enough to compute them empirically.
Whether a score is improving, stable, or declining over the last 90 days
A trend indicator is a directional marker shown next to each of the four primary scores (the three engine scores and the Architecture Score) that indicates whether the score is improving, stable, or declining over a 90-day rolling window. It is produced by the Temporal Layer. A score with fewer than thirty days of history shows a "new" indicator instead of a trend, because trends require enough observations to be meaningful.
Lifecycle stages
The position of the firm in its arc as an institution. The single most important contextual variable in the model.
The firm's stage in its arc as an institution
Lifecycle is the position of the firm in its arc as an institution. It takes one of three values: emerging, established, or franchise. Lifecycle is the most important contextual variable in the Colibrí Architecture model because every threshold, every band, and every reference point is calibrated separately for each stage. The same configuration that fits one lifecycle stage does not fit another.
Early in the firm's life, typically the first or second fund
An emerging firm is early in its life, typically operating its first or second fund with a small team and a defined thesis. The Colibrí Architecture model expects emerging managers to occupy a frontier that rewards conviction, which translates into narrower industry scope, smaller portfolio counts at higher ownership, smaller fund sizes (typically below $150M), and more sector or single-company concentration than would be appropriate for a larger firm. Concentration of conviction is structurally appropriate at this stage.
Multiple funds and a working operating cadence
An established firm operates multiple funds, has a working operating cadence, and is building a track record across vintages. It typically operates Fund III, IV, or V, with aggregate capital above $250M and a team large enough to support multiple concurrent deployments. The Colibrí Architecture model expects established firms to shift toward measured diversification, which translates into moderate industry scope, moderate portfolio counts, fund sizes in the institutional middle band (typically $150M to $750M), and the beginning of formal succession work.
A multi-generational structure with formalized succession
A franchise firm has scaled into a multi-generational structure with formalized succession and durable operating capacity. It typically operates Fund V or later, with aggregate capital above $1B, multiple funds in active deployment simultaneously, and a documented succession structure in place. The Colibrí Architecture model expects franchise firms to reward breadth, which translates into broader industry scope, larger portfolio counts at moderate ownership, fund sizes in the upper institutional band ($400M to $1.5B and above), and demonstrated transition capacity across partner cohorts.
A move from one lifecycle stage to the next, suggested by the platform
A lifecycle transition is a move from one lifecycle stage to the next. The Temporal Layer monitors the firm's structural variables and surfaces a suggested transition when the firm has accumulated enough change to warrant it. Triggers include fund sequence (reaching Fund III for the emerging-to-established transition, and Fund V for the established-to-franchise transition), aggregate firm capital, team size, deployment maturity, and the presence of a formal succession structure for the franchise transition. You can accept, defer, or reject the suggestion, and the platform never auto-transitions a firm between lifecycle stages on its own.
Firm and fund structure
The structural variables the model reads when it evaluates your firm and your funds.
A fund is a single pool of committed capital with its own vintage, mandate, target portfolio, investment period, and reserve allocation. A firm operates one or more funds, and each fund is scored independently by the Portfolio Efficiency engine. The Cross-Fund Concentration engine aggregates across all of a firm's funds together.
A firm is the organization that runs one or more funds, sometimes referred to as the GP entity. In the platform, the firm is the organization, and it carries the firm-level structural variables (team size, partnership structure, succession planning) that the Firm Design Congruence engine evaluates.
A GP is a partner at the firm who holds both economic and decision rights, distinct from non-partner investment professionals. The GP count is a structural variable that drives multiple parts of the model, including GP Load and the check that evaluates whether the IC voting structure is workable for the number of partners voting.
Capital each GP carries against LP commitments per year
GP Load is the amount of capital, in millions of dollars, that each GP is responsible for deploying per year of the fund's investment period. It is computed as fund size divided by investment period divided by GP count, and it has a workable range that is bounded on both sides. When the figure is too low, the partnership is underutilized, and when it is too high, each partner is writing more checks per year than they can responsibly source, diligence, and serve.
The IC is the voting body that decides the fund's investments. Voting models include consensus (every partner must agree), majority (most partners must agree), and conviction (decision authority sits with the partner with the highest conviction on the deal). The voting model interacts with industry scope, GP rights, and GP count in the Firm Design Congruence engine.
Industry scope describes how broadly the fund deploys capital across sectors. It takes one of five values, running from narrowest to broadest: deep (single sector), focused (small number of related verticals), thematic (theme-driven across multiple verticals), broad (broad mandate across distinct categories), and generalist (coverage across most major categories).
Lead practice describes how often the fund leads rounds versus follows. It takes one of five values: always, often, mixed, rarely, or never. Lead investors anchor rounds and typically acquire larger ownership stakes than followers. This practice interacts with ownership target, team size, and portfolio breadth across several of the engine's checks.
Board seat practice describes how often the fund takes board seats in portfolio companies. It takes one of four values: always, often, rarely, or never. Board seats are a significant post-investment obligation, and the practice interacts with team size and portfolio breadth in the engine's checks.
Platform scope describes the breadth of post-investment support the firm offers portfolio companies. It takes one of three values: full (a broad set of dedicated functions including recruiting, business development, talent, marketing, and operations beyond the investment team), narrow (one specialty function), or minimal (capital only).
Reserves are the share of fund capital set aside for follow-on investments rather than initial checks, expressed as a percentage of fund size. Reserves drive the implied initial check size, influence the fund's expected outcome consistency, and define the capacity that is available to the Scenario Engine when it evaluates follow-on decisions.
Pro-rata is the share of a new round that the fund has the right to invest in order to maintain its current ownership in a portfolio company. It serves as the natural anchor for the Scenario Engine's follow-on recommendations, where every recommendation is expressed as a multiplier on the fund's pro-rata entitlement. The multiplier ranges from zero (skip the round) up to roughly 1.5x (super pro-rata, reserved for the highest-conviction situations).
The vintage is the year of the fund's first investment, or the announced vintage year if no investments have been made yet. It is used to compute the fund's year-in-fund value and to anchor the deployment-phase calculation.
Deployment phase describes where the fund sits in its deployment cycle, derived from the share of fund capital that has been called to date. The three values are early (below 30 percent called), mid (30 to 70 percent called), and late (above 70 percent called).
Methodology terms
Concepts used in how the model is constructed and how it operates.
The frontier is a conceptual position on a risk-return surface that depends on the firm's lifecycle stage. The Colibrí Architecture model is built on the principle that emerging managers occupy a different frontier than franchise firms, which means that the same configuration choice can be efficient on one frontier and inefficient on another. The frontier shifts as the firm matures, which is why every threshold in the model is calibrated separately for each lifecycle stage.
A cohort is the set of funds on the platform that share the focal fund's lifecycle stage. Every score the model produces is interpretable only relative to its cohort, which is why a Composite Efficiency Index of 65 means one thing for an emerging manager and a different thing for a franchise firm. Cohort references are recomputed nightly, once the cohort holds at least thirty funds.
A snapshot is a time-stamped record of all of a fund's scores at a specific point in time. Snapshots are captured every time an input changes and on a nightly schedule. Each snapshot carries the lifecycle classification in effect at the moment it was taken, so historical comparisons that cross lifecycle transitions remain meaningful.
Staleness describes a configuration that has not been refreshed in a long time while the fund has continued to deploy capital. The Temporal Layer surfaces a staleness flag when configuration has not been updated in 180 days, or when the called-capital percentage has moved by more than 20 points since the last update. Staleness does not invalidate the scores; the model continues to compute them against the existing configuration, but the flag signals that the numbers may not reflect current operating reality.
A drill-down is the hierarchical path from a headline score back to the specific configuration variables that produced it. The path runs from the Architecture Score to the engine score, then to the engine's sub-components, and finally to the underlying configuration variable. Every score the platform displays can be traced through the drill-down back to its source.
The pro-rata multiplier is the Scenario Engine's recommended allocation expressed as a multiplier on the fund's pro-rata entitlement in a proposed round. It ranges from 0 (skip the round) up to approximately 1.5x (super pro-rata, reserved for the highest-conviction situations). The curve is lifecycle-adjusted, which means that emerging managers see a steeper curve that allows super pro-rata at high conviction, while established and franchise firms see flatter curves that typically cap at pro-rata or just above.
The binding constraint is part of the Scenario Engine's rationale output, and it names which constraint is limiting the recommended allocation in a specific scenario. It identifies whether the limit is the fund's available reserves, the firm's existing exposure to the company across all funds (concentration), or the proposed round's timing relative to upcoming milestones. The binding constraint tells you what would have to change for the recommendation to shift.
The External Reference Guide.
This printable reference covers everything on this page along with the full glossary, language guardrails, and approved phrasing for anyone writing about the model.