Clarify limits of meeting audio diagnostics

[r3dbars]

Why

A healthy Transcripted recording does not establish that another meeting app transmitted speech. The existing output_ducking_detected=false flag only means sampled hardware volume scalars did not drop; it cannot exclude software ducking.

Product Impact

• Affects: meetings; meeting reliability diagnostics and QA.
• Capture behavior, VPIO preferences, existing grades, and existing event meanings are unchanged. This is not a cross-app audio repair.

What changed

Add three fixed scope descriptors to existing capture diagnostic contexts: capture_health_scope=own_capture, cross_app_capture_status=unmeasured, and output_ducking_measurement=hardware_volume_scalars. Preserve them through the eight existing capture analytics schemas, support packets, and existing Sentry failure tags without forwarding new event types. Events without capture snapshots and older events can lack these descriptors.

Add regression coverage for unchanged legacy classifications and privacy/policy delivery. Require receiving-participant audio proof in the meeting QA matrix and add a same-build, built-in microphone crossover checklist. Remove the old suggestion that clean telemetry substitutes for manual audio proof. No automatic capture harness is needed for this first test.

How I checked it

• Agent preflight, mapped test selection, diff check, analytics emitter check, taxonomy normalization check.
• bash build-deps.sh --force.
• TRANSCRIPTED_SKIP_LAUNCH_SMOKE=1 bash build.sh --no-open; bundle performance budget passed. Launch smoke intentionally skipped to preserve the running app/audio environment.
• bash run-tests.sh: 12,719 passed, zero failed.
• bash run-integration-smoke.sh: core and wake smoke passed; seven MicRecordingFileMergerTests passed.
• Independent Astra Medium static review of base 792d82e159b2f5546e2b474d4a3100a4fe6fa442 through head 7d36c2490ed282b83c5793dea10e888c5efafa97: no remaining actionable findings.
• Live two-client Zoom/Meet audio, actual analytics delivery, and app-launch smoke: not run. No root-cause or hardware compatibility claim.

Test binaries used TRANSCRIPTED_DISABLE_FILE_LOGGER=1. The root Core package and public seam are unchanged; integration ran its mapped package-test filter.

Risk Review

Privacy and existing event consumers were reviewed. New values are constant categorical descriptors; there are no new hardware reads, process identities, audio content, storage migrations, or release/update changes. No customer screenshots or diagnostic payloads are included. Draft for human review; no merge or release.

Agent handoff

COORD_DONE: GREEN | this draft PR | diagnostic scope, regression coverage, receiving-audio QA | no cleanup | same-build live test remains | mapped checks and independent review passed | run checklist with a willing receiving participant