Open Lore NetworkOLN
Network/Register/Experiment And Feature Flag Harness
Gap Register
G-064Public

Experiment & feature-flag harness

Tier 1 — Existential, blocks launch
Status
Open — not started
Owner
Creator
Why now
Entry 031 and G-059 treat A/B experimentation as a given — "ship every gamification component behind a flag" — but the code only emits Amplitude events: there is no flag init, no variant assignment, no holdout. This small, blocking prerequisite gates the entire experimentation thesis.
Blocks
G-059
Related
Entry 032, Entry 031, G-059, G-063, G-026, G-010

Stand up the experiment/feature-flag layer (Amplitude Experiment via the existing @amplitude/unified SDK): variant assignment, flag gating, holdouts, and the paired engagement + data-quality guardrail events that every gamification and ranking experiment reads.

Why this matters

The A/B harness is assumed everywhere (Entry 031, G-059) and exists nowhere in code — ViewTracker only calls amplitude.track. Building it is small and unblocks gamification (G-059) and search-ranking (G-063) experimentation alike. Open decisions:

  • Assignment & gating — initialize Amplitude Experiment on the existing @amplitude/unified SDK; deterministic per-user variant assignment; server- and client-side flag evaluation; holdout groups.
  • Guardrail event model — make data-quality events (revert rate, contradiction-flag, consensus disagreement, rubber-stamp rate) first-class alongside engagement events, so a variant can't win on engagement while silently regressing quality (G-026, G-059).
  • Readout discipline — encode the Entry 031 methodology: 30–60 day horizons, holdouts, and a pure-intrinsic arm to detect overjustification crowd-out.
  • Change governance — who can launch/promote an experiment, and which surfaces (e.g. ranking, G-063) are constitutional-tier rather than unilateral.

Related

  • Entry 032 — the bridge strategy that names this the first build
  • Entry 031 — the gamification A/B thesis this enables
  • G-059 — gamified contribution (the primary consumer; blocked by this)
  • G-063 — entity search ranking (the other experiment consumer)
  • G-026 — Fact-graph data quality (the guardrail events)
  • G-010 — discoverability & recommendation (experiment surface)