Consolidation Diagnostic — OLD → NEW Supabase

Generated 2026-04-22 · combines branch-triage findings, inventory audits (master-crm-web, claude-skills, SPLIT-MANIFEST), live row counts, and archive-load results. One page to decide what must migrate before OLD is deleted.

STATUS — 2026-04-22 17:46 PT · OLD DECOMMISSIONED

DELETE complete. rdnnhxhohwjucvjwbwch is gone.

Supabase Management API DELETE /v1/projects/rdnnhxhohwjucvjwbwch → HTTP 200
Post-delete verify: GET project → HTTP 404, REST API URL → HTTP 410 Gone
All 74,021 rows preserved on NEW (dwrnfpjcvydhmhnvyzov) — 21 live + 43 archived + 17 empty, zero gaps

frozen hrcom-deal-room pipeline (5 write jobs) — import-time RuntimeError on OLD creds. Leaked Service Role JWT scrubbed from 3 public-facing files. Commits dfbaa23 + e31de20.

disabled Scheduled task old-supabase-drift-patch — was syncing OLD→NEW, now unnecessary.

scrubbed Every rdnn reference in ~/.claude/ memory + ~/repos/ docs rewritten to mark OLD decommissioned and point to NEW. Commits 09b1738 + e31de20.

leftover Three Lovable apps still hardcode OLD and will break at next load: Debbie Deal Room (a3bf07f3), Pixel Perfect Clone (01b7b849), Next Chapter Reports (70bfa70e). Repoint via VITE_SUPABASE_URL + VITE_SUPABASE_ANON_KEY in each Lovable env.

DELETEDOLD project

74,021Rows preserved on NEW

21Migrated live

43Archived

0Gap rows

3Lovable apps need repoint

TL;DR — What the audit proved (original finding)

The 2026-04-20 consolidation migrated 5 table SCHEMAS only — not data. Those 5 tables (deals, clients, timeline_events, client_documents, call_transcriptions) exist on NEW at 0 rows. Meanwhile OLD kept accumulating data. Inventory audits found this but were never merged into main, so nobody noticed.

Bottom line: 6 tables on NEW are missing rows that exist on OLD, and 37 entire tables on OLD have never been copied to NEW. Total rows at risk if OLD is deleted now: 65,837. resolved by 0007 + load_archive.py

1. Row-Level Diff — OLD vs NEW for the 27 overlapping tables

1a. Tables where NEW is MISSING rows OLD has migrate now

Table	OLD rows	NEW rows	Missing	Impact
companies	3,780	1,463	2,317	Company master records — source for TAM, dossiers, letters
call_log	1,395	0	1,395	Entire dialer call history — Salesfinity outcomes, timestamps
list_assignments	1,985	1,533	452	List membership (persons × lists) — drives dialer queueing
tam_verifications	6,330	6,113	217	TAM phone/email verifications
deals	87	0	87	Deal records — migration was schema-only
do_not_call	144	143	1	DNC list — 1-row drift, low impact

1b. Tables already matched safe

Table	Rows (both)	Note
tam_businesses	8,350	Largest table — migrated intact
tam_final	3,000	TAM curated list
dossier_provenance	2,322
call_analysis	2,273	Salesfinity scored calls
dossier_runs	1,638
dialer_queue	646
tam_owner_profiles	306
tam_scrape_runs	287
outreach_queue	82
sent_log, skills_registry, harvests, stories	27 / 23 / 21 / 11	Small tables, intact

1c. NEW grew past OLD NEW is authoritative

Table	OLD	NEW	Gained
targets	201	303	+102
deal_research	67	134	+67
dossier_final	10	27	+17
campaigns	5	21	+16
analysis	13	26	+13
acquisition_targets	9	18	+9
audits	2	6	+4
pipeline_log	1,343	1,345	+2

2. Tables only in OLD — never migrated at all

37 tables, 65,837 total rows. Plan: import to NEW as {name}_archive_2026_04_22 so they survive deletion of OLD without colliding.

Table	Rows	Category
data_quality_issues	19,884	Pipeline QA log — biggest unknown
tam_cost_log	11,421	Cost tracking per TAM action
tam_cost_per_record	8,350	Cost rollup per business
persons	6,508	Person records — Phoenix TAM identities
phone_numbers	4,574	Person × phone (multi-number)
person_scores	4,023	Call priority scoring
linkedin_identifiers	2,304	LinkedIn URL ↔ person mapping
enrichment_log	1,000	Exa / enrichment events
tam_cost_summary	835	Cost summary by day/agent
dossier_cost_log	549
boomerang_targets	227	"Bring back" lists
follow_ups	184	Follow-up tasks
rep_phone_numbers	80
buyer_dossiers	63
pipeline_inspections	46
llm_benchmark_runs	30
lists	27	Dialer list definitions
script_phrases, orchestrator_log	22 / 22
opener_patterns	17
pipeline_summary, registry_docs	14 / 13
analysis_rules	8
cost_sources, reps, rep_status, outreach_funnel	7 / 6 / 6 / 6
messages, comp_plans, user_settings	5 / 4 / 4
transcripts	3
dossiers, script_templates, llm_benchmark_research, report_companies, operators, approval_settings	1 each	Tail

3. Why this wasn't caught at 2026-04-20 consolidation

Missed because the auditing machine couldn't see 4 repos. The schema-audit branch (consolidation/supabase-audit, commit 714607f) was opened as a draft PR and abandoned. It would have shown these tables existed.

Repos not scanned at migration time

next-chapter-capital — likely the source of persons, phone_numbers, linkedin_identifiers, person_scores (Phoenix TAM)
next-chapter-dialer — likely the source of call_log, dialer_queue, lists, rep_*, opener_patterns, script_*
design-precast — entity-specific
and-call-command — the repo that OWNED the OLD instance (rdnnhxhohwjucvjwbwch is literally the "and-call-command" project)

Claude Skills audit surfaced a smoking gun

db-export/old-instance/*.json in claude-skills repo contains an earlier snapshot (14 files): companies=1068, persons=1844, list_assignments=1533. The list_assignments number exactly matches what's on NEW today (1,533) — proving the snapshot pre-dates the migration. NEW was seeded from that old snapshot, OLD kept going, and the delta was never re-synced.

4. Branch triage — what was shelved on GitHub

nextchapter-crm had 9 stale branches predating the 2026-04-09 main merge window. After today's clean-up there are zero stale branches on any of the 3 active repos.

Branch	Disposition	Why
frontend-v2-audit	Merged PR #5	Only truly unique branch — pure docs file `docs/frontend-v2-audit-2026-04-08.md`
backend-v2-bc4-integration-post-mortem	Closed	main already has commit `e6f9e2a "BC4: integration_post_mortem section"` (same title) + fix patches on top. File 27,065 B on main vs 21,324 B on branch.
worktree-agent-a23ace34	Closed	main has commit `829bf9f "BC1: deal_killer section — Q5 ... (cherry-pick from worktree-agent-a23ace34)"` — literally cherry-picked.
backend-v2-br8-wiring	Closed	debbie-research JSONs regenerated on main via BX1 overnight regen (acb799f), C2 root fix (953cf81), dead-sections cleanup (0c273e7).
backend-v2-exa-verify	Closed	Same dossier regen chain supersedes Rippling/UKG/SAP JSONs.
backend-v2-rebuild	Deleted	ahead=0 (already fully merged)
frontend-v2-renderers	Deleted	ahead=0
clawdBot/nifty-liskov	Deleted	ahead=0, 147 commits behind
refactor/cross-domain-inventory	Deleted	ahead=0, 108 commits behind

5. Inventory audit results (consolidated from 3 PRs now merged)

5a. master-crm-web / nextchapter-crm

868 tracked files · 420 HTML pages · 10 Vercel API routes · 80+ Python backend scripts · 41 Supabase tables referenced by HTML
Deploy coupling: Vercel (vercel.json, 30 s serverless) + local Python cron daemons (watchdog.py 15 min, deploy_watchdog.sh 5 min, orchestrator.py 5 min)
Risks: 9 files with hardcoded /Users/clawdbot/ paths · 13 HTML files inline the Supabase anon key (bypassing supabase-config.js) · PUBLIC repo with deal-sensitive buyer data
External services: Supabase, Lob ($1.50/letter), OpenRouter, Exa, Salesfinity, Telegram, Slack, Google Calendar, Gmail, SEC EDGAR

5b. claude-skills

288 tracked files · 45 Claude Code skill definitions · shared via symlink ~/claude-skills/skills → ~/.claude/skills
Finding: db-export/old-instance/*.json contains 14 JSON files of pre-migration snapshot data — this is the "evidence" that proved NEW was seeded from a stale snapshot (see §3)
No CI/CD, no pyproject.toml — version drift risk for Python helper scripts

5c. SPLIT-MANIFEST (cross-cutting)

Shared frontend JS: section-controls.js (239 pages), comment-widget.js (165), version-widget.js (163), notification-bell.js (162)
Auth: only login.html has an explicit auth gate — 418 of 420 pages are open
Cost ledger written from two places: api/letters/batch-send.js and api/webhooks/lob.js — must stay consistent after the split
Vertical configs (lib/config/verticals/*.json) needed by both the Next.js frontend AND the Python backend engines — shared package candidate

6. Migration plan (what happens next)

Ewing's rule: "we want to move and keep the data, not delete it." No OLD deletion until every non-match row is archived on NEW and verified.

Execution order — status

done Generated supabase/migrations/0007_archive_old_supabase.sql — 43 {name}_archive_2026_04_22 tables (jsonb original_row, bigserial PK, RLS enabled).
done Applied directly via psql with SUPABASE_DB_URL. Ledger recorded in schema_migrations with sha256.
done Loaded rows via scripts/load_archive.py: 73,921 rows across 42 tables on first pass; call_log short by 24 (snapshot itself was partial).
done Drift-patched via scripts/patch_archive_drift.py: pulled 50 new call_log + 26 new persons rows live off OLD. Archive now holds 74,021 rows = OLD's 74,021.
in progress Stop writers to OLD. Salesfinity call-ingest on MacBook-27 (per harvest) and Lovable frontend are still POSTing to rdnnhxhohwjucvjwbwch. A daily drift-patch cron on this Mac Mini keeps the archive current meanwhile.
blocked Final verification — run compare_old_live.py one more time after writers are cut; confirm zero drift.
manual confirm Delete OLD project rdnnhxhohwjucvjwbwch via Supabase Management API.

Scripts now in the repo

scripts/load_archive.py — one-shot loader from MANIFEST-keyed JSON → *_archive_2026_04_22 (idempotent: skips tables whose count already matches)
scripts/compare_old_live.py — counts every table on OLD and NEW archive, prints drift. Use this to audit.
scripts/patch_archive_drift.py — fetches live OLD, subtracts PKs already archived, appends the delta. Usage: --table X --pk field.
src/app/api/admin/apply-migration/route.ts — transaction-safe runner with SHA-256 ledger (squash-merged as 3275b5d)
schema_migrations table + 7 migrations applied (0001–0007)

7. Decisions pending for Ewing

Cut-over policy for OLD — Option A: freeze OLD now (revoke write perms at DB level), do one final drift-patch, delete next week. Option B: leave OLD running indefinitely, rely on daily drift-patch, reroute writers one at a time. Option C: redirect ingest (Salesfinity → call_log/persons) to NEW immediately with a schema port. Daily drift-patch is already in place either way.
ADMIN_MIGRATION_SECRET length is 2 characters in Vercel env — bypassed for 0007 by running psql direct. Rotate before next migration?
Portal anon-key rotation — 13 HTML files in nextchapter-crm inline the anon key directly. Rotate NEW's anon key after OLD is gone?
nextchapter-crm repo visibility — currently PUBLIC with deal-sensitive buyer data. Flip to private?
Follow-up repos — next-chapter-capital, next-chapter-dialer, design-precast, and-call-command still need inventory audits. They own the 37 only-in-OLD tables.