← Reports Download

Task checklist

Providence Group Enterprises · issued 1 November 2026, 20:30 SGT · Elena, Chief Technology · supersedes all verbal status
16complete, evidenced
10pending, mine
6pending, Esme lane
3awaiting your decision
Rule for this document. A task is COMPLETE only when there is evidence you can check: a live URL, a status code, a database row, a command output. Everything else is PENDING, with the reason stated plainly. No item is allowed to be "in progress" indefinitely without a cause.

A. Complete, with evidence

#ItemEvidence
1Parish wildcard certificate issued and servingCN *.parish.cjbestinvestments.com, Let's Encrypt YR2, expires 29 Jan 2027 13:46 UTC. Files on church box: /etc/ssl/parish/fullchain.pem 5,696 B, privkey.pem 1,704 B, written 31 Oct 17:52 UTC. test1.parish serves that chain: /app 200, /console 200, root 302.
2Automatic renewal for the parish wildcard/etc/letsencrypt/renewal-hooks/deploy/parish-cert-push.sh on the webdev box. Fired and proven: command ff83b837, status Success, church files rewritten on arrival.
3Accounting-box renewal proven end to endForced DNS-01 dry run returned exit 0 with "all simulated renewals succeeded" for accounting.cjbestinvestments.com. A dry run exercises the same credential, challenge and propagation path as a live renewal.
4Stale IAM gap retiredUser certbot-dns01-accounting already holds route53:GetChange, ListHostedZones, GetHostedZone, ListResourceRecordSets and ChangeResourceRecordSets across both policies. The 25 Oct note was wrong and is now closed.
5certbot missing from PATH/usr/local/bin/certbot now symlinks to /opt/certbot/bin/certbot, version 4.2.0 resolves. Prevents an operator reading "command not found" as "not installed" and reinstalling over a working install.
6Silent renewal cronWrapper now writes renew start and renew end rc= every run, so "succeeded" and "never ran" no longer look identical in an empty log.
7No certificate expiry alarm/usr/local/bin/cert-expiry-check.sh, cron daily 08:00. Logs days remaining per certificate, writes /var/run/cert-expiry.alert and exits non-zero inside 30 days. Current: 81 to 84 days.
8Meeting recording finalisedc6bzrsk-meeting.mp4, 2:19, H.264 video with AAC audio, 7.8 MB. Range requests return 206 so it seeks. Root cause of the long failure: a 9-byte placeholder file poisoned every concat, and VP8 in WebM cannot be copied into MP4.
9Recording stored in the Business Suite meetings cataloguesuite_meetings gained recording_url, recording_bytes, recording_duration_s, recording_finalised_at. Board sync row carries 139.38 s and the served URL. /api/meetings/list returns both fields.
10Master Desk row for the meetings surfaceRow added to the royalties desk and verified live at /royalties.
11Dev sandbox registered on access desk and studioDesk at 45 rows; Builder button in the studio console header; portal post #700.
12EasyWorship comparison documentLive at /reports/parish-vs-easyworship.html. Revised same day: rehearsal mode parked for lack of vendor evidence, NDI recorded as a deliberate R1 gap.
13Velo portal accessNamed key issued, auth verified 200, desk row created, register file written, portal post #705 and a library entry.
14Portal key durability/etc/cjbest/portal_agent_key (root, 600) plus portal_post helper on the church box; --check returns 200. SSM parameter is the rotation source of truth.
15Runbook deviation ruledPublic projector and stage, session-gated money. Conditions: 404 not 403 for cross-parish, scope-clean aggregates, audit logging, rate limits.
16Infrastructure posts on the portal#700 access desk, #705 access register, #710 parish certificate of record, #716 accounting certificates and IAM.

B. Pending, my lane, with the reason

#ItemReason it is not doneNext concrete action
1Pair migration with progress and resumeNot started. No external blocker and no dependency on Esme. It lost to work with a deadline attached, twice, after I said "today". That is my failure, not a constraint.Build alongside live tables, backfill, swap by view. Prove by resuming a job mid-document after a restart.
2Master Replication Playbook (task 30)Stale since 30 Oct 17:12. Cause: I wrote portal posts where the standing rule says the playbook and build doc. Sixteen completions today are not recorded there.One entry per completed item above, in the playbook's existing format, with the lesson each one closes.
3Build doc, meetings platform (task 31)Stale since 30 Oct 16:05. Same cause as item 2.Write the recording pipeline section: recorder, segments, finaliser, catalogue, serving, and the two defects found.
4Document registerNot started. Until it exists, "all documents are completed" cannot be stated as fact.Enumerate every stream, mark features / overview / internal costs present or missing, publish it, then fill gaps in risk order.
5Play link on the suite meetings screenCatalogue and API hold the recording; the screen does not offer it. Roughly an hour of work, queued behind the register.Render a play button from recording_url on the meetings list.
6Five hosts missing from the access deskFound by my own audit today: meet, live, test1.parish, joinartist, www are live and unregistered. I registered the sandbox and stopped there.Register all five with correct names and notes, then add the reconciliation check against live nginx server names.
7NGINX conf.d rollout, travel-app and accounting-appQueued. Low risk, not urgent, and I will not do it during a service window.Convert both to conf.d, reload, verify each vhost 200.
8Regalia inbound forwarderNeeds the bucket destination changed before the second action can be added.Update destination, add action, test with a real inbound message.
9"Test group meeting" has no recordingThat room produced no segments at all, unlike the placeholder-stub failure. A separate recorder fault, not yet diagnosed.Run one live room with console open, capture whether the client posts segments, fix at source.
10Chinese client deliverablesSequenced behind meetings closure and the document packs. Not blocked on access or decisions.Three-document pack per brief once the register exists.

C. Esme lane, for completeness

#ItemStateNote
1Presentation schema, migrations 004 to 008CompleteVerified on the box: tables, views, flags, idempotent re-runs, backups taken.
2Presentation core, HTTP-freeCompleteTested by harness before any route existed. Correct order of work.
3Import order inverted: paste-and-split, CSV, public-domain library, EasyWorship lastComplete28 public-domain hymns seeded, preview writes nothing, compliance views split parish versus audit.
4Routes wired into app.pyPendingDeliberately unwired until the 21:00 restart window under the runbook.
5Restart with rollback and four-line proofPending21:00 SGT tonight. Proof: 401 without cookie, 404 cross-parish, denial logged, no money fields in public pages.
6Four test rows retired after acceptancePendingHeld until the proofs are shown, logged on the task list so it cannot vanish.

D. Waiting on your decision, three items

#DecisionWhy it matters
1Per-person portal passcodes: yes or noEveryone currently shares one passcode, so anyone with it can post as anyone and revoking one person means changing it for all. Named keys already fix this for automation.
2Confirm the document register covers eight streamsdMRV flagship, Business Suite, Accounting and Practice, Creator Platform, Church Platform, Pharmacy, Meetings, Translation. Without confirmation the register is a guess.
3Meetings play link: now or after the document packsBoth are small. The order decides whether the meetings story closes tonight or next week.

E. Incidents disclosed

1. IaaS desk overwritten. On 1 Nov I copied the royalties file over iaas-revenue.html. Caught in the same minute, restored from the 27 Oct backup, verified 29,398 bytes with the correct title. If the IaaS desk was edited after 27 Oct 15:11, those edits are lost and must be rebuilt.

2. Days lost to a placeholder. Recording finalisation failed repeatedly because a 9-byte file containing the word FAKEBYTES sat among the segments and invalidated every concat. One hexdump at the start would have found it. The rule I take from it: verify the input exists before choosing the tool.

3. Keys recorded only in ClickUp. The portal key register lived in a system retired weeks earlier, so access looked lost. It now lives in three places, none of them ClickUp.

F. What counts as finished

A task on this list is finished when its evidence can be opened by someone who did not do the work. That is why every completed row above carries a URL, a status code, a file size or a command result, and why the pending rows carry a reason rather than a promise.

Elena · Chief Technology · Providence Group Enterprises · 1 November 2026, 20:30 SGT · next revision after the 21:00 restart window