Portfolio Construction Discipline

When your fund stops looking like the one you raised

A fund declares its construction once, before it has made a single investment, then spends a decade deploying. The gap that opens between the two is the ordinary condition of a fund, not a failure state. What makes it worth watching is that every reading of a fund is a reading of what was declared.

What drift is

A fund declares its construction once, in a document written before it had made a single investment. Then it spends a decade deploying. The gap that opens between the declared setup and what the fund actually became is configuration drift, and it is the ordinary condition of a fund rather than a failure state.

It matters because every reading of a fund, including this model's, is a reading of what was declared. A configuration that has stopped describing the fund produces answers to a question about a fund that no longer exists.

How often it happens

Honestly: nobody knows, and we are not going to invent a number. There is no published dataset on how far declared fund configurations diverge from realized ones, for the straightforward reason that declared configurations mostly live in fundraising documents and private models that never get compared to outcomes at scale.

What can be said is structural. Every input in a construction plan is a target rather than a commitment, none of them is enforced by anything, and each one is set at the moment of least information. A gap between plan and reality is the default expectation, and the useful question is not whether a fund has drifted but in which direction and by how much.

Three ways it happens

  • Drift by accumulation

    No decision was made. Cheques ran slightly larger than planned, or the portfolio filled up faster, and the fund arrived somewhere it never chose.

  • Drift by revision

    A real decision was taken, often a good one, and the declared configuration was never updated to match it.

  • Drift by growth

    The firm changed around the fund. A team doubled, a mandate widened, a partner left. The fund's own numbers are untouched and the firm they describe no longer exists.

The second is the one worth dwelling on, because it is the one that looks like a problem and is not. A firm that decided in year three to concentrate more heavily than planned has made a legitimate call. The only thing wrong is that the declared configuration still says otherwise, which means every reading taken from it, and every LP conversation that references it, is describing the old plan.

What it costs

The immediate cost is that analysis degrades quietly. A configuration reading does not announce that its inputs are stale; it simply answers the question it was asked, about the fund as declared. Nothing looks broken.

The larger cost is external. A fund's declared construction is what LPs were told, and it is what they will measure against at re-up. A firm that has drifted deliberately and never restated the plan has to explain the gap under the least favourable conditions, during diligence, rather than having narrated it as it happened.

There is also a compounding version. The next fund is usually constructed by amending the last one's plan. Drift that is never reconciled gets inherited, so Fund III is built on a description of Fund II that was already inaccurate.

How the platform surfaces it

Two mechanisms, deliberately mild. Every score is recorded as a time-stamped snapshot, captured when an input changes and on a regular schedule, so the record shows the configuration moving rather than only its current state. And a configuration that has not been refreshed while the fund has kept deploying raises a staleness flag, triggered when no fund-level field has changed in 180 days while active investments exist, or when called capital has moved more than 20 points since the last configuration update.

Staleness does not invalidate anything. The scores keep computing against the configuration on file, and the flag simply lets any reader see that the reading may not reflect current operating reality. That is the honest treatment: the model cannot know whether a firm drifted or deliberately changed course, and it does not guess.

The action it asks for is small. Restate the configuration to match what the fund now is. That re-bases every subsequent reading, and it creates a dated record of the change, which is exactly the artifact that makes the LP conversation straightforward later.

Take it to your own fund

Run the model on your own fund.

The platform reads the Colibrí Architecture model against your own firm and funds.