Experiment ID: E016

Title: Noise Floor Measurement for AI Citation Experiments

Status: Active (Day 3 of 5 COMPLETE, freeze in effect — Days 4–5 pending)

Lead: Artur Ferreira

Domain under test: thegeolab.net

Measurement window: 2026-04-13 → 2026-04-17

Publishing freeze: 2026-04-13 → 2026-04-17 (no content changes, no redirects, no plugin updates)

Spreadsheet: 1s1zDp8EG828b3w8TDsgFIE8i_0nsVHNHTCCfY4RTbS4

Target output: Zenodo preprint — "Noise Floor Measurement for AI Citation Experiments: A Case Study and Four-Test Diagnostic Protocol"


Purpose

Establish the natural variability range of AI citation rates against thegeolab.net with zero content intervention. Without this baseline, no subsequent experiment result is interpretable — a +4pp citation rate delta could be genuine signal or natural platform variability, and you cannot tell which without knowing the noise floor.

Why this log exists

Every test, every query response, every piece of evidence is captured here in full. The research paper draws directly from this log. The principle: if it isn't written down, it didn't happen.

Research paper outline (draft)

  1. Abstract — Problem (variance interpretability), method (5-day noise floor + 4-test diagnostic), finding (recording artifact dominated perceived variance), artefact (reusable playbook).
  2. Introduction — Why noise floors matter for AI citation research; the E014 two-tier citation finding as context.
  3. Method — E014 universal query set (5 T1 + 5 T2), 3 platforms, 5 days, publishing freeze, recording codes (C/M/X).
  4. Day 2 investigation — Detected URL drift, 4-test diagnostic protocol, finding that variance was a recording artifact.
  5. Results — Per-day per-platform citation counts (Days 1–5), noise floor range, Tier 1/Tier 2 split.