Mostly Wiki

session-49

6 min read

Session 49 — Jacob’s SoW Audit Reconciliation + Pier Log Phase 2 Cutover

What was built

Two strands collapsed into one cutover. Jacob’s 84-row SoW pier-quality audit (the gating item from session 48 Bucket B) came back this morning as a CSV + ODS attachment. Reconciled the audit, executed the data-hygiene work it surfaced, then cut over the pier-log flow from Phase 1 (permissive WARN gate, PIN URL still up) to Phase 2 (blocking gate, PINs retired, DRIVELOGS→PIERLOGS rename, JES schedulers notified).

Audit reconciliation (data hygiene, before the cutover)

  • Pulled Jacob’s CSV from his Gmail thread (forwarded to operations@); ran a second pull on the ODS version which carried green/orange/red color coding the CSV had stripped.
  • Color-coded breakdown: 67 green (confident), 15 orange (low-confidence/unknown), 3 red (Dustin escalations).
  • 3 red flags addressed:
    • 12 D Street NW (PMS.OPS.45903) — pier log was AI-mis-parsed (pier numbers weren’t filled in on-site, only depths). Corrected pier_log.json for piers 7 and 8 (pier 7: depth 34ft, 10 tubes, 4000 PSI; pier 8: depth 28ft, 8 tubes, 4000 PSI). Recomputed stats (AVG DEPTH 27.6, AVG PSI 3662, MAX PSI 4000). Regenerated cert PDF (/data/projects/6994da901edc26584544d4fb/CERTS/...v1.0.pdf); Dustin to resend.
    • 7993 Scotts Manor Ct (blkhse.540479) — “Messy. Cert might be queried.” Dustin will review separately.
    • 8206 Oak Lane (PMS.OPS.45858) — Plan/photos/cert all differ. Dustin will review separately.
  • 5 Dustin-directed status actions:
    • 2681 Cunningham Hole (PMS.OPS.45904) — flagged PPDR via Stephaney/triage. triage_review_type=desk, notes added.
    • 5707 Greenleaf (blkhse.566850) — closed out (was actually inspected 2025-11-13, invoiced/certified/paid).
    • 41 West Trail CT (PMS.OPS.40.R8016) — flagged PPDR blocked on Kai photos. review_type=Desk Review, triage_status=blocked, triage_blocked_reason=missing_photos.
    • 6804 Glenwood Ct (blkhse.540691) — closed out (already certified/invoiced/paid).
    • 11924 Viewcrest and 246 Mill Swamp — left alone per Dustin (one cancelled/no-recovery, one legacy PPDR).
  • 60-row green-bucket SOW backfill: wrote as_built_qty + inspected=wD4FU (Inspected) to 60 SOW lines via update_sow_line + bridge to SOW Gate SQLite via _write_sow_gate_confirmation. All 60 verified post-write. Audit notes attribute source as “Backfilled from Jacob audit 2026-04-26 (green-row confirmed)“.
  • 7 legacy no-UND-SOW projects: flagged in pier_log_on_disk_but_no_und_sow audit. Reparse couldn’t generate SOW lines (0 PERMITS / ~0 PLANS on disk for these legacy blkhse projects — pier log itself is the only source). Per Dustin: “we are chasing our tail on these. can we ignore these with a note on each and move on to phase 2?” → wrote legacy-ignore note to triage_notes on all 7; advanced blkhse.44927 from Ready for Cert → Certified to align state with the others.

Phase 2 cutover (ran together as one window, ~68 sec downtime)

  1. cert_validate.py Check 5 flipped from CheckResult.WARN + warnings.append to CheckResult.FAIL + errors.append. Logger message changed PHASE-1 PIER-GATE WOULD BLOCK → PIER-GATE BLOCKED. Comment block above rewritten. Commit 1a82a65, pushed.
  2. scripts/migrate_drivelogs_to_pierlogs.py (new, idempotent, supports --dry-run / --db-only / --disk-only). Dry-run showed 2,761 disk dirs to rename + 254 project_files rows to update + 0 conflicts.
  3. Live execution at 13:43:37Z–13:44:45Z UTC (the API stopped, disk renamed, DB updated via transient docker compose run, config flipped, image rebuilt, services restarted):
    • Disk: 2,761 dirs renamed /DRIVELOGS → /PIERLOGS in <1 sec.
    • DB: UPDATE project_files SET category='PIERLOGS' WHERE category='DRIVELOGS' → 254 rows.
    • config.PIERLOG_DIR_NAME = "PIERLOGS".
    • services/pierlog_escalation.py line 64 hardcoded "DRIVELOGS" replaced with config.PIERLOG_DIR_NAME.
    • Commit f51d862, pushed (config + escalation service + new migration script).
  4. Caddyfile edit (not git-tracked): unlikely.pl/{4-digit} PIN URL block dropped; added explicit handle { respond 404 } catch-all so unmatched paths return clean 404 instead of the previous empty-200 default. Backup at /etc/caddy/Caddyfile.bak.20260428T135900Z. Validated with caddy validate before install + reload. Verified post-reload: / → 200 (Step 1 page), /4829 → 404, /pierlog/qr.png → 200, /pierlog/log/no-match → 200.

Post-flight verification

  • /api/health → 200 OK
  • find /data/projects -maxdepth 2 -name DRIVELOGS → 0 dirs; -name PIERLOGS → 2,761 dirs
  • project_files.category=DRIVELOGS → 0 rows; =PIERLOGS → 254 rows
  • cert/validate/{12 D Street NW record_id} Check 5 → PASS via legacy fallback (“1 pier line(s), 2 legacy log file(s) on VPS”). End-to-end pier-log read path through new PIERLOGS dir confirmed.

JES + crew comms

Drafted, reviewed, revised through ~6 rounds with Dustin, then sent.

  • Email sent from operations@unlikely.pro (no spoof — Dustin’s name in body sign-off + signature, company brand block auto-appended). To: Julia (BAL), Anna (MAN), Kai (NHAV); Cc: Ryan Joyner, Nicole Lovo, James Carey, Lily Jacobs, Shejdie Childs, Jeff Corso, dustin@. Subject: “New pier-log upload flow — what’s changing today”. Gmail msg id 19dd486ba534d8b1.
  • Tells branches: (1) the new permit+customer entry flow at unlikely.pl; (2) custom QR stickers shipping ETA ~1 week; (3) the 7:15 AM scheduler escalation routing; (4) reply if they want a different SMS number.
  • Attached: pier_log_quickguide.pdf (96 KB, 2 pages, Letter size, Tufte-style serif on cream). Page 1 EN + Page 2 ES, QR on each, 5 numbered steps. Rendered via services/pdf_renderer (Playwright). Multi-day per-product upload semantics clarified (“when ALL piers of one type are done, even across multiple days; push and slab on same job upload separately”).
  • Two SMS-ready plain-text quick-guides (EN + ES) — schedulers can paste straight into a text to a driller. Same content as the PDF.
  • Memory update: reference_jes_gw_contacts.md — Jeff Corso entry corrected from ambiguous “transferred to NH” to explicit “Former CT GM, now New Hampshire (NH = New Hampshire, NOT North Haven)“.

Workflow diagram check

docs/workflows/drift-map.json doesn’t fire on cert_validate.py / pierlog_escalation.py / config.py changes. inspector-field.mmd references “Pier Log Uploaded?” and “Pier Log Received?” at the right level of abstraction — still accurate post-cutover. No diagram updates needed.

Commits

  • 1a82a65 (entirely-api/master) — cert_validate Check 5: flip pier-log gate from permissive (WARN) to blocking (FAIL)
  • f51d862 (entirely-api/master) — Phase 2 cutover: DRIVELOGS → PIERLOGS rename across disk + DB + config
  • (CLAUDE.md + SESSION_LOG.md commit at end of session)

Open follow-ups

  1. iPhone walkthrough of new pier-log flow on real device (Tufte details, focus zoom, state colors)
  2. 7993 Scotts Manor Ct + 8206 Oak Lane — Dustin’s two remaining red-flag escalations
  3. 5 invoiced-can’t-verify projects (1903 Andrew, 5105 Acorn, 2019 Osborn, 8665 Valley, 1470 Flag Harbor) — SS↔Entirely state desync investigation
  4. Duplicate projects in Entirely: 248 Elkins (2 entries), 3875 Pine Cone (3 entries) — dedupe pass
  5. Vision parser hardening for unlikely.pics — Jacob’s catalogued failure modes (rotation, cropped, glare/shadow, two-column, missing pier numbers)
  6. 4 legacy projects from backfill_pierlog_check.py (psi-vs-slab tagging ambiguity)
  7. Office portal UI to consume pierlog_discrepancies table
  8. First real PIER-GATE BLOCKED log entry — watch container logs over next few days
  9. First 7:15 AM ET morning escalation tomorrow (2026-04-29 11:15 UTC)
  10. Failover test for Rob fallback profile (carry-over from session 48)
  11. Telnyx 10DLC campaign approval check + TFV status (carry-over from session 43)
  12. Rotate exposed credentials from session 43 .env paste (carry-over)

Graph View