Fidelity and recency

Fidelity and recency are the two weights the platform uses to combine measurements into a competency profile. Together they produce the per-competency aggregate that drives every Grow surface.

Fidelity

A measurement's fidelity is how reliable the platform considers that kind of measurement. It is set by the assessment type, not by the specific assessment.

The v0.5 fidelity weights:

Higher fidelity means the platform leans on the measurement more when blending into the profile. A simulation result moves the profile noticeably; a CV-derived signal barely moves it.

Recency

A measurement's recency weight decays with time. The full formula in v0.5 uses a half-life of about 12 months — a measurement from a year ago carries roughly half the weight of an equivalent measurement taken yesterday.

Decay is not abrupt. A six-month-old measurement is still meaningful; a five-year-old measurement is nearly invisible in the profile.

The recency decay is what makes the profile current without ever requiring a manual cleanup. Old measurements lose influence automatically; new ones replace them.

How they combine

For each competency, the profile holds a single current level computed as:

level = sum(measurement_level × fidelity × recency_weight) / sum(fidelity × recency_weight)

Where the sum runs over every measurement that has touched the competency.

A new high-fidelity recent measurement can move the profile noticeably. An old low-fidelity measurement can be present in the formula and barely move the result.

What the formula is not

• Not a simple average. A simple average would treat all measurements equally.
• Not a peak. The profile is not the highest level the person has ever measured at; it is a current best estimate.
• Not a decay-only filter. Old measurements still count, just with less weight.

The deliberate balance: fresh, high-fidelity signal dominates, but no single measurement can completely override the body of evidence.

Confidence

In addition to a level, each competency in the profile has a confidence number. Confidence rises when:

• More measurements have touched the competency.
• Recent measurements agree (small variance across recent contributions).
• Higher-fidelity measurements have contributed.

Confidence falls when:

• The competency has only one or two old measurements.
• Recent measurements disagree (high variance).
• Only low-fidelity contributions exist.

The Grow surfaces use confidence to decide how much to lean on a level. A Strong match with low confidence is shown with a small caveat; a Strong match with high confidence is presented without hedging.

Worked example

An employee has the competency Stakeholder management with three measurements:

• A CV-derived self-rating from 18 months ago, level 3 (fidelity 0.4, recency ~0.4).
• A structured interview from 6 months ago, level 2 (fidelity 0.7, recency ~0.7).
• A simulation from last week, level 2 (fidelity 0.9, recency ~1.0).

Numerator: (3 × 0.4 × 0.4) + (2 × 0.7 × 0.7) + (2 × 0.9 × 1.0) = 0.48 + 0.98 + 1.80 = 3.26

Denominator: (0.4 × 0.4) + (0.7 × 0.7) + (0.9 × 1.0) = 0.16 + 0.49 + 0.90 = 1.55

Level = 3.26 / 1.55 ≈ 2.10

The recent simulation and structured interview dominate; the old CV claim contributes weakly. The result rounds to level 2 with reasonable confidence.

Why this matters

When a profile feels off, the diagnosis is usually one of:

• No recent measurement in the competency — the recency decay has eroded what was there.
• A divergent measurement is dragging the level — the recent measurement disagrees with older ones.
• A high-fidelity measurement landed — a recent simulation can shift the level by a level or more.

In all cases, the path to a more accurate profile is a fresh, well-fitted assessment. Manual edits are not the answer; the platform deliberately does not allow them.