Is Strict Qualified Usage the right default for public metrics?

[roy-tong]

Our measurement policy treats usage as UNKNOWN by default: an observation only upgrades to Qualified Usage when evidence supports it (see product/ and QUALITY.md).

Why this bias: CI, health checks, benchmarks and manual curl calls all look identical at the call boundary. Without the strict default, every public metric silently absorbs noise - and then the metric is worthless for the thing we care about: proving real agent usage.

The trade-off: strict defaults undercount at first. A tool genuinely used by agents will show Qualified Usage of 0% until identity evidence is collected. That is a feature (no inflated claims), but it makes early numbers look bad.

Questions for the community:

1. Should public-facing metrics default to Strict Qualified Usage, or publish raw and qualified side by side?
2. Where should the line be between declared (self-claimed by the caller) and correlated (independent evidence)?
3. Is there a scenario where undercounting is more harmful than overcounting?

I lean toward: publish raw and qualified side by side, but gate claims (badges, comparison, billing) on Qualified. What breaks in that model?