Skip to content

Reputation Score or Standing Index

Scoring model — instantiates Reputational Signal Governance

Aggregates a subject's weighted traces into one score, band, or standing index used to sort trust, access, ranking, or scrutiny.

Reputation Score or Standing Index is the compression step of the pipeline. It takes the many weighted traces about a subject — reviews, verified facts, endorsements — and collapses them into a single comparable figure: a score, a percentile, a standing index that others can act on at a glance. Two decisions define it and set it apart from its siblings: it fixes the subject boundary (whose traces roll up into one standing, and where one subject ends and another begins) and it chooses the aggregation function that turns a scatter of signals into one number. It does not capture, verify, or display the underlying traces; it computes standing from them.

Example

A freelance-services marketplace runs a seller-standing index spanning every category a seller works in. Every completed contract, on-time delivery, dispute outcome, and client rating carries a weight and rolls into a single seller standing. The design first settles the subject boundary — standing attaches to the seller account and is aggregated across categories rather than reset per category — then combines the weighted activity into one figure. That figure gates real things: past one threshold a seller unlocks higher-value briefs, past a higher one they surface in the "top-rated" tier buyers filter to. A buyer glancing at a newcomer's standing reads a track record across the whole marketplace in a single number, which is exactly what the index is built to compress.

How it works

What distinguishes it is subject definition plus aggregation math:

  • Fix the subject boundary. Decide the unit standing attaches to — account, verified person, organization, or per-context identity — and how one subject is kept distinct from the next.
  • Aggregate. Combine weighted traces with a chosen function — a sum, a weighted average, a Bayesian shrinkage toward a prior for thin histories — so many signals become one.
  • Express a form. Emit the result as a raw score, a percentile, or a coarse band, depending on how finely the decision downstream needs to sort.

Tuning parameters

  • Subject boundary — account vs. verified person vs. organization vs. per-context. A broad boundary resists whitewash-by-new-account but can unfairly fuse unrelated conduct under one name.
  • Aggregation function — a sum rewards sheer volume, an average rewards quality per interaction, a shrinkage estimator protects thin records from a lucky or unlucky start.
  • Input weighting — how much verified facts count against reviews against endorsements. This is where the model decides what "reputation" is actually made of.
  • Output granularity — a precise number sorts finely but over-signals trivial differences; a coarse band is legible but blunt.
  • Newcomer prior — where a blank-slate subject begins. A generous prior eases the cold-start problem; a stingy one is harder to exploit.

When it helps, and when it misleads

Its strength is that a single comparable figure makes fast sorting, ranking, and gating possible at scale — the whole point of turning history into one number.

That same compression is where it misleads. It erases context: two subjects with identical scores can have wildly different histories, and the number hides which. Once the figure gates anything valuable it becomes a target and is optimized rather than earned — Goodhart's law[1] — so the score drifts from the behavior it was meant to summarize. And thin histories yield confident-looking numbers built on almost no evidence. The classic misuse is reverse-engineering the weights until a favored subject lands on top — tuning the index to ratify a ranking already chosen. The discipline is to publish the aggregation logic, shrink thin data toward a prior, show the score beside its evidence rather than in place of it, and keep the subject-boundary rules stable over time.

How it implements the components

  • aggregation_model — the function that combines a subject's weighted traces into the score, band, or index.
  • reputation_subject_boundary — defines the unit standing attaches to and where one subject ends and another begins; the boundary the aggregation is computed over.

It does NOT capture or display the underlying traces (that's Rating and Review System and Verified Transaction History), apply time-decay (Decay-Weighted Score Update), map the score onto differentiated treatment (Trust-Tier Badging), or defend the subject boundary against fake identities (Sybil, Collusion, and Brigading Detection).

Notes

The index is never better than the traces beneath it, and its single number is seductive precisely because it hides that. It should surface the evidence, not stand in for it — and it deliberately leaves time-decay to a dedicated rule, so that how the past fades stays a separate, reviewable choice rather than a constant buried inside the aggregation.

References

[1] Goodhart's law — "when a measure becomes a target, it ceases to be a good measure." A reputation score that gates real privileges is a standing invitation to optimize the number instead of the behavior, which is why the aggregation logic and its weights need to be governed, not just published once. ↩