Skip to main content

IndexFair › Methodology › Source weighting

/methodology · v2.1 · append-only · last updated 2026-01
/methodology/source-weighting

Source weighting & confidence intervals

Each source contributes to a score by a fixed trust weight. The spread of that evidence becomes the ± interval — the score is a point, the interval is the honesty around it. The confidence label shown next to it is a separate signal: it is set from how much evidence has been collected, not from this interval's width.

What it means

Trust weights are fixed per source tier and applied before aggregation — they do not change with a brand's score, cannot be overridden per brand, and are published configuration. A wider confidence interval means less agreement across sources or fewer effective sources, not a worse brand. The confidence label (Normal, Low, Emerging, or Insufficient) is computed separately, from how much evidence has accumulated — it is not derived from this interval, and the two can move independently: a brand can carry a narrower interval than another brand while still showing a less favourable confidence label, or the reverse.

Inputs & data fields

FieldTypeMeaning
source_weightfloatFixed trust weight for this source tier, 0–1.
sub_scorefloatAggregated aspect score from this source, 0–10.
n_efffloatEffective review count after weighting and decay.
sigmafloatStandard deviation of weighted sub-scores across sources.

Process & formula

score = Σ(wᵢ · sᵢ) / Σ wᵢCI = z · σ / √n_eff (z = 1.96, a 95% interval)The confidence label is a separate signal, not a band of this CI — it reflects how much evidence has accumulated.
  1. Multiply each source sub-score by its fixed trust weight and sum.
  2. Divide by the sum of weights to get the blended score point.
  3. Compute σ across the weighted observations, scale by z for a 95% interval, divide by √n_eff to get the interval.

How it affects public pages

What ± means. 7.2 ±0.5 says the index very likely sits in 6.7–7.7. A wider interval means less agreement or less evidence — never a worse brand. Affects: the ± on every score. The confidence chip beside it is a different signal — how much evidence has been collected — computed independently of this interval.

Worked example

Sourceweightsub-score
Source A · Review platforms0.327.4
Source B · App stores0.247.0
Source C · Public forums0.186.9
blend7.2 ±0.5

All figures are illustrative. The interval of ±0.5 reflects the spread across the three sources — not a failing grade.

Related