Decay-Weighted Score Update¶

Calibration process — instantiates Reputational Signal Governance

Discounts old evidence on a schedule so standing tracks who a subject is now, not who they were years ago.

A Decay-Weighted Score Update is the rule for how the past fades. It re-computes standing on a cadence, weighting each piece of evidence by its age so that recent conduct dominates and stale conduct gently loses its grip. Its defining move is treating time as a first-class weight rather than a filter: nothing is deleted, but a two-year-old rating counts for a fraction of a two-week-old one. This is what lets a reputation breathe — it builds in the possibility of change, so a subject who has genuinely turned around is not held forever to their worst stretch, and a subject coasting on old glory sees it quietly erode.

Example¶

A freelance marketplace shows each seller a rating. Early on it was a lifetime average, and it lied in both directions: a designer who was mediocre for her first ten jobs but excellent for her last fifty still looked middling, while a onetime star who had gone stale still looked great. The platform switches to a decay-weighted update. Now each new review enters at full weight and older ones fade on a half-life — roughly, work from the last few months carries most of the signal, and reviews past a couple of years contribute little.

The effect is immediate and fair in the right direction. The improving designer's score climbs to match her recent work within a quarter; the coasting star's drifts down toward his current, thinner output. Serious, verified violations are set to decay far more slowly than an ordinary late delivery, so "recency" doesn't become a loophole for waiting out a real harm. Sellers can see the rule — the half-life is published — so nobody imagines the number is a fixed lifetime verdict.

How it works¶

Age-weight every piece of evidence. Each contributing event gets a weight that declines with its age (commonly an exponential decay with a stated half-life), so freshness, not just count, drives the result.
Refresh on a cadence. Standing is recomputed on schedule or on each new event, so it reflects the present rather than a frozen snapshot.
Weight by more than age. The same update applies quality weights — verified evidence and higher-credibility sources count more than thin or anonymous ones — so decay works alongside, not against, evidence quality.
Decay asymmetrically where it matters. Severe, verified negatives are given longer half-lives than routine ones, so the rule enables rehabilitation without amnesia.

Tuning parameters¶

Half-life / decay rate — how fast old evidence loses weight. Short half-lives track change quickly but make standing jumpy and easy to reset by waiting.
Positive/negative asymmetry — whether negatives fade slower (or faster) than positives. Slower-fading serious negatives prevent "wait it out" gaming; too slow denies genuine rehabilitation.
Refresh trigger — scheduled recompute versus event-driven. Event-driven is responsive; scheduled is predictable and cheaper.
Decay floor — how much weight the oldest evidence retains rather than reaching zero. A nonzero floor preserves long memory of patterns; a floor of zero maximizes fresh-start.
Cold-start handling — how a thin or new record is scored before enough recent evidence exists, and how heavily to hedge it.

When it helps, and when it misleads¶

Its strength is that it makes reputation current and improvable — it tracks who a subject is now, prevents ancient evidence from dominating, and gives concrete institutional form to the idea that people can change. Recency-weighting is the standard, well-understood way to do this.^[1]

It misleads when the half-life is set without regard to what kind of memory should be short. A uniform fast decay erases warranted long memory — a serious harm should not fade at the pace of a late shipment — and it hands strategic actors a "reset by dormancy" move: go quiet, let the bad stretch age out, return with a laundered score. Its classic misuse is tuning the rate to bury one specific bad period. The discipline that guards against it is asymmetric decay for severe verified negatives, a published rate so the aging is legible rather than covert, and never letting decay substitute for the appeal path when an entry is wrong rather than merely old.

How it implements the components¶

decay_and_rehabilitation_rule — its core: the schedule by which old evidence loses weight, which is what operationalizes rehabilitation-through-time.
evidence_quality_weighting_rule — the same recompute weights evidence by quality and source credibility, not age alone, so fresh-but-flimsy evidence doesn't swamp older solid evidence.

It deliberately does not build the underlying record or define the base aggregate it re-weights — capturing conduct is Contribution Ledger's job, and the base score it updates belongs to the score index (Reputation Score or Standing Index); this process ages and re-weights, it does not originate.

Instantiates: Reputational Signal Governance — it is the time-calibration that keeps standing current and makes change possible.
Consumes: Contribution Ledger and the base score index supply the dated evidence this update re-weights.
Sibling mechanisms: Contribution Ledger · Moderation Record with Reentry · Reputation Score or Standing Index · Appeal and Correction Workflow · Complaint and Resolution Record · Attested Credential Registry · Peer Reference or Vouching · Rating and Review System · Sybil, Collusion, and Brigading Detection · Trust-Tier Badging · Reputation Portability Protocol · Verified Transaction History

Notes¶

Decay and Moderation Record with Reentry both let a bad record recede, but by different logic and they should not be conflated. Decay is automatic and time-based — standing fades as a function of age, no action required. Reentry is conditional and earned — a sanction lifts on demonstrated change or met conditions, not merely on the clock. Using decay where reentry is warranted lets serious offenders wait out consequences; using reentry where decay suffices makes ordinary staleness a bureaucratic process.

References¶

[1] Exponential recency-weighting (an exponential moving average with a chosen half-life) is the standard way to make a running estimate track a changing quantity — old observations lose weight smoothly rather than being kept at full force or dropped abruptly. ↩