Added#
Coverage drift in --against: what the comparison stopped being able to
see. The drivers name the dollars that moved; nothing named the findings
that stopped being measurable. A log where session was on 98% of calls and
is now on 4% has not fixed its conversation growth — it stopped recording the
field that would show it, and every session-shaped finding went quiet for a
reason that has nothing to do with the bill. A fixed finding and a blinded log
are opposite facts that the dollars render identically; only coverage tells
them apart, so the comparison now states any field whose share moved by 20
points or more, in either direction, loud on a collapse and quiet on a gain.
The findings a collapsed field took with it, named. "Some findings are silent" is not something a reader can act on; knowing that conversation growth, per-conversation cost, repeated turns, truncation retries and the cache-TTL fit are now silence rather than absence tells them exactly which sections of the report to distrust. One list per field, on all three surfaces.
The comparison gate refuses a comparison it cannot make. --max-growth-usd
now fails when the current log stopped recording a field the previous one
carried, before it judges the dollars at all. The bill can hold perfectly flat
while the log went blind, and a gate passing there would be certifying a
comparison nobody could check — the same refusal --max-day-usd makes on a
clockless log and --max-session-usd on a sessionless one. A field that
appeared never refuses: seeing more is not a reason to stop.
Coverage drift on every surface. The MCP's profile_usage and the web
bill render the same finding at the same 20-point threshold, so an agent
relaying "spend flat, all clear" cannot do so off a log that stopped
measuring.