← TERMS

Track Record

Past committed decisions in a domain — and how they turned out

Track record is the accumulated history of named commitments in a domain and how those commitments turned out — relative to the confidence expressed at bind time. It is one of four components of authority weight (with qualification, certification, and authorisation). Introduction: But Commitment Is Not Enough.

Track record is domain-specific. An authority may have strong standing in fraud review and none in clinical diagnosis. Weight follows the relevant domain, not rank or charisma.

Why it matters

Trust is not a feeling. It is the record of commitments kept by someone with standing to make them — or eroded when commitments pile up unbacked by outcomes (The Promise, Where Everything Breaks).

A L0 event from someone with no relevant track record is structurally valid but epistemically weak. The Facets Composed treats authority weight as an input to risk judgment alongside tier, atomicity, and uncertainty — the facet that stops a high-ranking but low-standing voice from settling a question by position alone.

Commitment without standing is worthless. Standing without commitment is worthless opinion.

From reputation to data

When consequential decisions are captured as committed events — authority, evidence snapshot, moment — every authority accumulates a visible, queryable record of binds and outcomes. Track record stops being gossip and becomes measurable:

SignalWhat it measures
Brier scoreProbabilistic forecasts vs outcomes (lower = better calibrated)
Override rateHow often human review actually changes model or first-reviewer outcomes
Domain accuracyCorrect/incorrect or cost-weighted outcomes for non-probabilistic binds
Calibration curvesWhere stated confidence systematically diverges from reality

Without measured track records, meritocratic weighting becomes opinion weighting — politics, not epistemics. Strengthening the Commitment Boundary requires weights on collective panels to be pre-committed from prior records (Brier scores, historical accuracy, validated curves) — not set after votes are visible (compliance theatre). Seeded review cases and manual recency sampling supply ongoing human performance data for those records.

What track record is not

  • Certification — a credential attesting competence without full outcome history; a shortcut when strangers assign weight.
  • Qualification — training and expertise that may not yet appear in committed outcomes.
  • Authorisation — organisational permission to bind; rank without outcomes.
  • Presentation skillconfident deck rhetoric is not track record.

Positional authority can override epistemic authority when track records are invisible — The Buried Finding pattern in Article 10.

Building and using track records

  1. Record commits at L0 with named authority and domain scope.
  2. Link outcomes — forward evaluation of whether the bind was warranted (not silent rewrite of history).
  3. Score longitudinally — single events are insufficient; Brier and override rates need history.
  4. Apply weights before the next bindmeritocratic panels, escalation routing, model promotion decisions.

Enabling Intelligence lists track records alongside Brier scores and override rates under calibration and feedback.

Corpus stance

A2 — working context: Ordinary language for standing-from-outcomes; formalised here as a first-class component of authority weight once commitments are event-sourced and outcomes can be scored.

A3 — open question: Open formalisation of how components combine into a single weight figure: Outlook.