docs(skills): record the iOS 27 hardware testing findings

[atdr]

Summary

Records what hardware testing of #88, #89, #95 and #96 on 2026-09-27 established, in the four repo skills it touches. Docs only, independent of the stack.

Why

The two-way audio campaign skill still said "still flaky sometimes" from July, and pointed at the iPhone → intercom direction. Today's per-hop measurement showed that on iOS 26 audio had been reliable apart from startup, and that the residue is the Pi → iPhone leg at session start (#102). It also cost most of an afternoon to discover that a whole class of "silent both ways" failures was the test rig, not the server. Both are worth having written down before the next person, or agent, reaches for the campaign.

Changes

• intercom-two-way-audio-campaign: status narrowed to Caller audio drops out in the HomeKit live view (Pi → iPhone leg) #102; new Phase 0a to rule out the test rig (MacBook answering via Continuity through the Home-app iPhone); the per-hop audio probe as the Phase 0 instrument, with a warning never to leave it enabled; the ~3 s ffIn startup silence and burst as known Gate 2 behaviour; the controller's 16 kHz Opus RTP clock recorded as harmless on the return path.
• intercom-debugging-playbook: a triage row for "silent both ways from the moment the live view opens" and a failure-archaeology entry for the Continuity false alarm. Prettier re-padded the tables, which is most of the line count.
• intercom-testing-and-validation: the far end of a test call must not be the Home-app iPhone, and a single-phone Twilio self-test call recipe (Twilio calls the intercom number from itself with inline TwiML, recorded dual-channel).
• telephony-audio-reference: answer supervision (why ringing can outlast an answer on the PSTN) and a correction: the call is answered as soon as /twiml returns <Connect><Stream>, so the Pi's ringback is a deliberate disguise rather than ringing on an unanswered call. Notes that the end of ringback is the visitor's cue to speak, which is why Caller audio drops out in the HomeKit live view (Pi → iPhone leg) #102's startup window loses their first words.

Test Plan

• prettier --check .claude/skills passes.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 69.47%. Comparing base (11c1820) to head (3a4a8e8).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #103   +/-   ##
=======================================
  Coverage   69.47%   69.47%           
=======================================
  Files          14       14           
  Lines        3240     3240           
=======================================
  Hits         2251     2251           
  Misses        989      989           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.