Transparency

Transparency

Scores should leave a paper trail.

Good Companies Directory publishes its opinionated rubric, source standards, score calculations, confidence basis, verification limits, and audit log. No quiet rewrites.

Methodology Commitments

Opinionated, not hiddenThe rubric expresses a moral view about institutional power. The point is not neutrality; the point is a visible standard applied consistently.
Sources before scoringClaims need source links and review metadata before they can support a score. Source quality affects confidence and review priority.
Verification has limitsVerification means the record was checked through a defined process. It is not endorsement, certification, warranty, or proof that no facts are missing.
No paid outcomesPaid verification can correct errors or omissions, but score changes require evidence and reviewer rationale under the published rubric.

Public Trail

ClaimsStatement, axis, source links, confidence, status, timestamp, and version history.
ScoresAxis values, caps, penalties, computed confidence, rubric version, and supporting claim IDs.
AppealsChallenges to claims, sources, or calculations with reviewer rationale.
ChangesTimestamped audit log entries for score and methodology updates.

Reviewer Model

The founder serves as initial lead reviewer and final authority. Contributors submit evidence but do not assign scores. Paid reviewers may verify and synthesize claims over time. Senior reviewers are promoted based on accuracy and consistency, while final scoring authority remains narrow and accountable.

Machine Assistance

A local LLM may assist with claim extraction, duplicate detection, normalization, and summarization. It does not assign scores or make final judgments. Human reviewers approve all LLM-assisted outputs before publication. AI-generated records remain tentative and capped below fully reviewed work until a reviewer checks source quality, specificity, counterevidence, and score mapping.

Consistency Checks

Computed confidencePublic confidence is computed from evidence coverage, source quality, disputes, recency, and review status. Seed confidence labels are kept in sync with that computation.
Profile thoroughnessDirectory profiles carry a non-scoring research-completeness label: Sketch, Partial, Developed, Extensive, or Exhaustive. AI-labeled values remain marked until a human reviewer updates them.
Direct evidenceProfile scoring tables distinguish direct axis-specific claims from contextual claims so a reader can see what actually supports the score.
Rare bonusesBonus columns stay conservative. Anti-Scapegoating Culture credits require sourced practice, not generic civility language or performative anti-bullying claims.
Regression testsPublication checks cover confidence drift, missing direct evidence markers, score wording, public exports, and profile rubric clutter.

Source Registry

Source classes are maintained in seed/source-registry.json. They affect confidence and review priority; they do not mechanically prove a claim true.

Class Weight Definition
Primary public record 100% Court, regulator, government, military, parliamentary, public-health, or other official public records.
Primary institutional record 90% Annual reports, audited financials, SEC filings, proxy statements, official policy documents, and organizational bylaws.
Independent reporting 82% Wire services, established newspapers, public-interest investigative journalism, and other independent newsrooms.
Official self-description 68% Official websites, about pages, impact pages, sustainability pages, press releases, and other self-published institutional claims.
General web source 58% Ordinary web sources that are usable but not independently high-authority for scoring.
Social or commentary source 40% Social platforms, video platforms, newsletters, blogs, and similar sources that usually need corroboration.

Correction Policy

Factual errorSubmit a source-backed challenge from the profile page. Reviewers check the source and publish a decision.
CounterevidenceContradictory evidence can lower confidence, mark a claim disputed, or change a score component.
Company responseCompany representatives may submit corrections, but payment, verification, or logo display never buys a higher score.
Public trailAccepted corrections and score-impacting changes are recorded in the audit log with rationale.

Public Data

Generated data exports live under /static/data/: entities, claims, score components, audit log, source registry, and a compact scored directory export.

Names and Logos

Good Companies Directory uses company names by default. Verified companies and groups may opt in to logo display, but logo display never affects scoring, claims, appeals, or reviewer authority. Logos, when shown, are fixed-size, secondary to the GCD badge, and may be removed for misuse, misleading submissions, or stale verification.

Audit Log

No score changes quietly.

Every claim and score update carries a timestamp and rationale. This log records the public trail for methodology and rating changes.

May 2026 Latest 100 entries

June 2026

Loading audit log…