The floor mechanism

Each datapoint has a configured floor — the minimum score the datapoint can produce given absence of evidence. Floors prevent zero-collapse: a single missing structural signal does not drag the parent vector to zero.

The floor is not “zero plus a little.” It reflects the structural minimum the datapoint can reasonably produce. For some datapoints (e.g., schema-presence), absence of evidence is genuinely close to zero — a brand with no Schema.org markup at all has produced no positive evidence. For other datapoints (e.g., content-freshness), absence of evidence does not imply absence of value — a stable evergreen reference page is not stale even if its publication date is old. The floor for each datapoint reflects this distinction.

For the practitioner, floor-handling has practical implications:

• A datapoint at floor is not necessarily failing. It may be at floor because no evidence was collected (the page wasn’t crawled, the source was unavailable, the language wasn’t calibrated). The reliability flag on the AS measurement indicates this.
• A datapoint at floor with high confidence (evidence was collected, no signal was found) is genuinely deficient. Action work is appropriate.
• A datapoint at floor with low confidence (evidence was not collected) is uninformative. The remedy is to lift confidence (re-crawl, re-fetch, expand calibration), not to perform OMG action work.

Section 4 of each datapoint page distinguishes “what at floor looks like” — what the absence-of-evidence state structurally indicates for that specific datapoint.

---