Ryan Schultz 37b5eb4609 docs: ADR 008 — platform weight governance

Defines how platform weights are set and changed: founding vote
(initiator defines eligible members, averaged proposals), annual
community vote (all platforms simultaneously, median of submitted
distributions), and structural change tier.

Updates ADR 002 and ADR 007 to reflect the new mechanism, and
ARCHITECTURE.md to mark weight governance as resolved rather than
a future direction.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

2026-05-14 20:31:32 -05:00

2.5 KiB

Raw Permalink Blame History

ADR 002 — Simple weighted scoring, configurable via config.yaml

Date: 2026-05-10 Status: Accepted

Context

We needed a way to turn raw participation counts (posts, commits, edits, messages, funding) into a single comparable signal per user. Several approaches were available, ranging from simple weighted sums to machine learning models.

Options considered

Option	Notes
Simple weighted sum	Transparent, adjustable, immediately understandable
Normalized weighted sum	More "fair" across platforms with different scales
Graph-based reputation	Models relationships, not just activity counts
Machine learning model	Could learn patterns, but opaque and hard to argue with
Equal weight per platform	Simpler but ignores meaningful differences in effort

Decision

Simple weighted sum with weights defined in config.yaml.

The formula is:

score = (forum_posts × w₁) + (github_commits × w₂) + (wiki_edits × w₃)
      + (matrix_messages × w₄) + (funding_activity × w₅)

Starting weights (determined by founding vote per ADR 008; adjusted annually thereafter):

github_commits: 0.4 — highest weight; commits represent sustained technical effort
forum_posts: 0.2 — meaningful contribution but easier to produce at volume
wiki_edits: 0.2 — documentation is undervalued; deliberately weighted equally to forum
matrix_messages: 0.1 — low weight; high volume, hard to assess quality
funding_activity: 0.1 — included but intentionally small; money ≠ participation

Why not normalize?

Normalization (converting scores to percentages or z-scores) was deliberately deferred. Raw sums are easier to explain and trace. "Your score is 78.8" is harder to argue with than "your score is in the 94th percentile" when the dataset is small and the model is new.

Normalization can be added once the signal is validated socially.

Consequences

Scores are not directly comparable across communities with different activity volumes
Platforms with high message counts (Matrix) could dominate if weight is raised carelessly
Anyone who disagrees with the weights can fork config.yaml and run their own ranking (see Correctability over precision)
Platform weights may not be changed via direct config.yaml edits; changes go through the community vote process defined in ADR 008

2.5 KiB Raw Permalink Blame History Unescape Escape