Why the same score means opposite things depending on direction

What gets recorded

Every score the platform produces is written to a time-stamped history: the three engine scores and the Architecture Score, captured when a configuration input changes and again on a regular schedule.

The result is a record with two kinds of point in it. Input-driven snapshots mark the moments a General Partner changed something. Scheduled snapshots mark the passage of time between those moments, which is what makes it possible to see that a score moved because the portfolio moved rather than because the configuration was edited.

The four trend readings

Each score carries a directional marker computed over a 90-day rolling window.

• Improving

  The score has moved up enough over the window to be a direction rather than noise.

• Stable

  The score has moved, as scores always do, but not enough to mean anything. Most readings sit here and that is the correct outcome.

• Declining

  The score has moved down enough to be a direction. Worth reading against what changed in the configuration rather than treated as a verdict.

• New

  Not enough history to say. Shown rather than substituting a guess.

The fourth band is the one worth dwelling on. A score with fewer than thirty days of history, or with too few recorded observations to fit a direction, is marked new rather than stable.

The distinction is not cosmetic. Reporting a firm with a single snapshot as stable would claim knowledge the model does not have, and it would be indistinguishable from a genuine stable reading. Saying nothing yet is the honest output, and it is worth more than a confident-looking one.

Why a snapshot is not enough

The clearest argument for this pillar comes from an adjacent domain. Value Add VC, writing on how emerging managers should benchmark themselves for LPs, puts the central problem plainly: the number one mistake in LP reporting is not bad numbers, it is presenting good numbers with no context.

Their conclusion about re-ups is the part worth carrying across. The managers who raise a second fund are not necessarily the ones with the best numbers. They are the ones who made their LPs feel informed and confident, including through the J-curve, when the numbers themselves were not yet flattering.

That is a statement about reporting rather than performance, and the mechanism is exactly what a trend indicator provides. A figure with a direction and a history attached is a figure an LP can situate. The same figure alone invites the least generous interpretation available, because there is nothing else to go on.

Their own frame of reference is fund performance, benchmarked by vintage against Cambridge Associates, Preqin, Carta, and PitchBook. The scores here are a different object entirely and are not performance benchmarks. What carries over is the structural point: contextualized beats raw, and the context that matters most is time.

How to read a moving score

A declining score is not automatically bad news, and this is the most common misreading. Scores move for three quite different reasons and the trend arrow cannot distinguish them.

The configuration changed, because the firm made a deliberate decision. A firm that raised its target portfolio count will see several readings move, and if the decision was right the movement is simply the cost of it.

The portfolio changed, as capital deployed and positions accumulated. Concentration readings in particular move on their own as a fund deploys, without anyone touching the configuration.

Or the firm crossed into a new lifecycle stage, at which point the same configuration is read against different expectations, and scores can step without anything about the firm having changed at all.

The history is what lets a General Partner tell these apart, because the input-driven snapshots mark exactly where a decision was made.

What the trend is not

It is not a forecast. A score improving over three quarters says nothing about the next one, and the model makes no claim that a direction persists.

It is also not a performance measure. These are readings of configuration coherence over time, not of returns, and a firm whose scores are improving may be doing so while its portfolio does something entirely unrelated.

Sources

• Fund Benchmarking for Emerging Managers: How to Show LPs Where You StandValue Add VC, May 2026On why context rather than raw numbers drives LP confidence, and on what actually separates the managers who raise a second fund.