[patty-chow]

Summary

Follow-up to #1 and the actual attendee acceptance walkthrough.

• Fix the confirmed first-connection failure bug: await public MongoClient.connect() on every async database acquisition, including each readiness-loop retry. Preserve explicit targets, validated cloud TLS, timeouts and retryWrites:false; no automatic replay of writes.
• Add first-failure, concurrent acquisition, shutdown and readiness unit tests plus a real isolated Compose regression that proves recovery with the same app PID and original waiting process. The harness owns and cleans only its ephemeral project.
• Replace intentional TODO error screens with accessible exercise checkpoints; remove misleading indefinite loading; retain real errors. Add persistent guide and calendar-import links and clarify editable implementation labeling.
• Add concrete Codespaces file/Ports navigation and progressive hints for both driver exercises.
• Add actual projected document/index inspection for migration before/after and an optional read-only Monday-by-track aggregation exercise with snapshot-derived expectations.
• Add Google Calendar web and Apple Calendar macOS import instructions and timezone/snapshot/reimport guidance.
• Preserve catalog, data volumes, credentials, starter/reference recovery, dependency pins, and Azure exclusion from the attendee scenario.

Verified results

• macOS final code: 46 tests passed, 1 existing Linux-only permission test skipped, 0 failures. In the restarted Codespace: 47/47 passed, no skips.
• 46 local documentation links checked; actual two-server database integration 2/2 passed.
• Actual local Docker and Codespaces reconnect regressions passed: original wait process recovered; first-use Site B failure recovered in the same app PID; concurrent acquisition and healthy-client outage recovery passed; notebook digests unchanged; own ephemeral project cleaned.
• Main attendee app browser proof: first-ever Site B connection failed while B was stopped; after restarting only B, the browser retry succeeded with the same actual server PID, start time, container and request-token hash. No app restart or workaround.
• SIGINT/SIGTERM interruption cleanup 2/2 passed, including repeated signals during cleanup and independent empty container/volume/network inventories. The review finding about interruption leaks is fixed.
• Full restarted-Codespace attendee flow passed: blue expected checkpoints, sequential TODOs, 216 entries, save/overlap/persistence, independent B, copy/retry/return, live inspection v1 -> v2, migration 216 then 0, optional aggregation 17 groups/74 entries, and real three-event browser ICS download. Starter/reference switching preserved both solved exercises and notebooks.
• Independent review approved application changes and the harness cleanup follow-up. Runtime/browser proof was at 8993fac; final harness retests were at a3f9c5a, which changed no app/UI code. Final commit records acceptance in documentation only.
• The same Codespace is verified Shutdown, with prior and final notebooks/solutions preserved. No new Azure resources or duplicate calendar imports in this retest.

Prior-environment evidence and limits

The previous main was exercised in a fresh two-core Codespace, real Google Calendar import (3/3 exact events), and a local-workstation-to-managed-Azure round trip. The temporary Azure cluster/group were deleted and verified absent. Those prior results are not presented as fresh testing of this revision. No new Azure resources are created for this change. Actual event laptop/network and a timed novice/classroom rehearsal remain outside the claim.