Colibrí Architecture methodology

How Architecture tests whether a fund’s design holds together.

Architecture turns a fund plan, the firm behind it and the portfolio that follows into a set of inspectable diagnostics. This reference explains what the platform reads, how each score is formed, what changes as actual data arrives and where partner judgment remains indispensable.

Model
v0.4
Public reference
19 October 2026
Maintained by
Colibrí Strategies
Research foundation
Colibrí Institute

The practical question

Can this firm execute the fund it has described?

Architecture does not start by predicting a return. It starts with the choices a GP can observe and govern: check size, ownership, company count, reserves, pace, scope, decision rights, team capacity and cross-fund exposure. Three independent engines test those choices from different angles. A temporal layer compares the approved plan with the operating fund, and a separate Scenario Engine supports specific follow-on reviews.

InputsFund plan, firm design, investments, follow-ons and manager documents
01Portfolio Efficiency
02Firm Design Congruence
03Cross-Fund Concentration
Integrated viewArchitecture Score, plan-to-actual readings and decision work

Inputs and provenance

The platform distinguishes the manager’s record from outside evidence.

The GP remains the source of record for the fund plan, actual investments, follow-ons, marks, decisions and governance. Architecture can read manager-provided fund, portfolio-company and board materials to reduce repeated entry and to draft context, but it labels those materials as manager-provided evidence. Public research can add financing, market and company context; it does not silently overwrite the structured record.

Declared configuration

The approved fund and firm choices the partnership intends to operate.

Recorded actuals

Investments, follow-ons, cash flows, marks and decisions entered or confirmed by the manager.

Manager documents

Fund and company materials used as attributed context, with page or slide location retained where available.

Public evidence

SEC filings, public LP results, published research and public company information, each used within its stated limits.

The core model

Three engines answer three different questions.

The engines run independently. A strong reading in one cannot erase a weak reading in another.

01Fund level

Portfolio Efficiency

Do the fund’s construction choices fit together?

The Composite Efficiency Index reads five declared choices: capital math, stage and ownership, scope and breadth, lifecycle fit, and team capacity. Early in deployment, each contributes 20%. In mid and late deployment, five percentage points move from capital math to scope and breadth because the observable portfolio becomes more informative.

ComputationCEI = weighted mean of C1, C2, C3, C4 and C5

The score requires the configuration fields used by all five subscores. It is a coherence diagnostic, not an expected-return estimate.

02Fund and firm design

Firm Design Congruence

Do the firm’s operating choices reinforce one another?

Eleven checks compare decision rights, economics, ownership, stage, scope, geography, platform promise, team capacity, lead practice, board load and succession. Each check is visible on its own, so the partnership can address the actual tension instead of reading only a total.

ComputationCGS = max(0, 100 − 10 × hard tensions − 5 × soft tensions)

An aligned check carries no penalty. Check 11 can surface two additional succession tensions when the firm reaches the Continuity stage.

03Firm level

Cross-Fund Concentration

Where is the firm’s deployed capital actually concentrated?

The engine calculates sector distribution and company exposure across the firm’s funds using cumulative deployed capital. Stage and geography remain descriptive. Expected sector ranges depend on declared scope; company exposure thresholds adjust only where the model explicitly provides a lifecycle rule.

ComputationCFCS = max(0, 100 − applicable concentration penalties)

This is exposure analysis. It does not state that concentration is inherently good or bad, and it does not estimate portfolio outcomes.

Inside Firm Design Congruence

Eleven checks make a structural tension specific.

The Configuration Congruence Score is a summary. The check result and its explanation are the decision-useful unit.

01Decision rule × scope

Can the investment committee make decisions at the breadth the strategy requires?

02Decision rule × economics

Do voting rights and economic rights create a workable decision system?

03Boards × team

Can the investing team carry the board workload implied by the plan?

04Platform × team

Is the platform promise supportable by the people assigned to deliver it?

05Geography × team

Does the operating footprint fit the team available to cover it?

06Ownership × lead practice

Does the ownership target fit how often the firm expects to lead?

07Deal and board load

Can each investor source, lead and support the implied annual workload?

08Ownership × stage

Is the target ownership plausible for the entry stage?

09Partner count × voting

Can the stated voting rule work with the number of decision-makers?

10Capital × partner capacity

Does the annual capital load per partner sit inside a workable operating range?

11Lifecycle × succession

Has the firm built the continuity its current stage requires?

Integrated view

The Architecture Score preserves the cost of a weak dimension.

The score is the geometric mean of the three engine results. CEI and CGS are aggregated across funds using deployed-capital weights; CFCS is already firm-level. More than half of deployed capital must be covered for a firm-level component to be available. Before deployment, the model uses a count-weighted mean across funds with populated scores.

Architecture Score= ∛(CEI × CGS × CFCS)

The geometric mean makes imbalance visible: a very low component reduces the composite even when the other two are strong. A missing component does not become a zero; Architecture shows the available components and explains what information is still required.

80–100

Plan and portfolio aligned. Configuration, design and capital distribution fit together.

65–79

Mostly aligned. A few specific gaps to review.

50–64

Several gaps to review across dimensions.

35–49

Material gaps in one or more dimensions.

Below 35

Foundational gaps to close before the next fund.

Plan to actual

A score is a dated reading, not a permanent label.

The Temporal Layer locks an approved plan, compares operating data with that version and keeps later readings attributable to the model version that produced them. It can surface deployment, ownership, sector, reserve and workload drift without rewriting the original plan.

01

Declare

Complete the fund and firm configuration.

02

Lock

Approve a dated plan version as the comparison point.

03

Operate

Record investments, follow-ons, marks and cash flows.

04

Review

Read changes, open decisions and stale inputs in context.

Companion decision module

The Scenario Engine supports a follow-on review; it does not cast the vote.

For one existing position, Architecture combines the company record and supplied documents with the proposed round, fund reserves, pro rata, current exposure, portfolio alternatives and public research. It returns a decision posture, constrained allocation range, timing signal, reasons, missing evidence and conditions the partnership should resolve.

A

Position signal

Uses recorded marks, trajectory, stage progression, financing recency and status. Missing history remains missing; it is not inferred from a polished deck.

B

Allocation constraint

Starts from pro rata, then applies the stricter of reserve capacity, the offered amount and firm- and fund-level company exposure caps.

C

Timing

Distinguishes act now, wait for a milestone, wait for the round or reassess based on the decision evidence actually supplied.

D

Partner review

States the recommendation, why, the open evidence and the explicit conditions for approval, refusal or another review.

The current policy caps a Conviction-stage allocation at 35% of remaining reserves and other lifecycle stages at 50%, before applying company concentration limits. Those are explicit design rules, not observed optimal allocations. The manager remains responsible for diligence, valuation, conflicts, legal review and the investment decision.

One model, different entry points

Every On Demand report continues into the same operating record.

A one-time report applies a defined portion of the platform to one fund and one current job. It is not a simplified or separate methodology. If the manager continues with a plan, the saved fund, documents and report remain available rather than being requested again.

01

Fund Math Check Report

A sourced report that shows which parts of the fund design cohere, which are stretched and which questions an LP is likely to ask next.

02

Reserves Plan

A company-level view of reserve use, remaining capacity and the design constraints that should shape the next allocation decision.

03

Follow-On Decision Memo

A concise internal memo connecting current exposure and fund capacity with the manager-supplied round, ownership, operating, runway, syndicate and risk evidence for the vote.

05

Fund II Readiness Brief

A readiness brief connecting public next-fund timing evidence with the concrete Fund I progress the manager can show now.

06

LP Quarterly Update Draft

An editable first draft covering portfolio activity, deployment, reserves, cash flows and plan-versus-actual observations for the quarter.

Lifecycle context

The same choice can mean something different as the firm changes.

Lifecycle changes context; it does not award prestige. Architecture uses three public display labels while preserving the existing stored values underneath.

Stored as emerging

Conviction

The firm is proving a repeatable strategy and should keep the fund design legible enough to execute with a small team.

Stored as established

Cadence

The firm has repeated the work and must make the operating system, ownership model and portfolio practice hold across funds.

Stored as franchise

Continuity

The firm must preserve decision quality while distributing authority, economics, institutional memory and succession.

Interpretation

What Architecture can—and cannot—support.

It can support

A coherent construction review; explicit comparison of plan and actuals; visible structural and concentration tensions; a documented follow-on review; and a sourced account of what information produced the reading.

It cannot establish

Future returns, investment merit, a company valuation, legal or tax conclusions, independent verification of manager materials, or an optimal allocation. A high score means the inputs fit the model’s design rules—not that the fund will outperform.

Validation status

The current model parameters are versioned and test-covered. Several thresholds are principled design rules or provisional reference points rather than outcome-validated optima. The follow-on backtest harness is in place, but the validation result is withheld until at least 200 evaluable historical decisions are available. This release reconciles the public reference with the implemented model; it does not change scoring weights, thresholds or curve parameters without that separate analytical review.

Evidence and limits

Sources are attached to the claim they can support.

Colibrí Institute research informs the questions Architecture asks. Public filings and published studies add context. Manager data produces the fund-specific reading. These sources are not interchangeable.

The Architecture of Returns

White Paper II · 2,142 U.S. venture funds · 2000–2024 vintages

Research context for fund configuration and risk-adjusted efficiency. Historical group comparisons are not return forecasts.

Read the source

Why Emerging Venture Capital Managers Matter

White Paper I · 2,471 U.S. venture funds · 2000–2024 vintages

Research context for lifecycle comparisons. Its sample remains distinct from White Paper II.

Read the source

Colibrí Institute Data and Methods

Research definitions, sources and reuse boundaries

The public record for how the Institute separates research samples, licensed data and reproducible public material.

Read the source

Venture Fund Benchmarks

SEC filings and public LP results, maintained as separate evidence streams

Public context for fund formation, pacing and reported results. A comparison is shown only with its population and source.

Read the source

Form D

U.S. Securities and Exchange Commission

A notice of an exempt securities offering used for raise context. It is not treated as a final-close record or performance result.

Read the source

The Risk and Return of Venture Capital

John H. Cochrane · NBER Working Paper 8066

A research input for the deterministic reserve simulation’s venture-outcome distribution assumptions, not a calibration claim about a specific fund.

Read the source

White Paper I and White Paper II use distinct samples—2,471 and 2,142 U.S. venture funds respectively—and are not combined. SEC Form D is a notice of an exempt offering, not evidence of a final close or investment outcome. Public LP results are kept separate from filing-derived raise data. Proprietary data that cannot be republished is used only within its license and is not presented as a public dataset.

Definitions

The terms used across the platform.

CEI
Composite Efficiency Index: the fund-level coherence of five construction dimensions.
CGS
Configuration Congruence Score: the result of the eleven structural checks.
CFCS
Cross-Fund Concentration Score: the firm-level reading of sector and company exposure.
Architecture Score
The geometric mean of CEI, CGS and CFCS, subject to component availability and coverage.
Plan
A dated, locked version of the construction choices approved by the manager.
Actual
The operating record confirmed in Architecture: investments, follow-ons, cash flows, marks and decisions.
Benchmark
A public comparison with a named population, period and source. It is not automatically a target.
Design rule
A model parameter chosen to make a tension or operating constraint explicit. It is not described as an observed optimum unless validation supports that claim.

Apply the methodology

Begin with the fund you are operating now.

Model one fund free, see which inputs drive the first reading and decide whether the continuing platform or one focused report fits the work in front of you.

Design one fund freeSee the six one-time reports