⚠️ HISTORICAL / SUPERSEDED (S491, 22/07/2026). Point-in-time artefact
(16/04/2026 DB queries against the pre-cutover prod project). Its
new-project transfer model is dead — the ratified cutover is in place
onto existing prod (see reingest-cutover-runbook.md §7), reframed by DR-025
as a one-time onboarding ingest. The schema it classifies predates ID-131
(content_items retired S449; extraction tables re-parented to
source_document_id) and the cocoindex pipeline rebuild. Its restore
discipline (FK-ordered, category classification) remains a useful
reference; its table names, item counts, and entry-point queries must NOT be
executed as written.
Status: DRAFT (S176 WP3 — Liam decisions resolved + squash results) Last
updated: 16/04/2026
Purpose: classifies every ingress path and every persistent-state surface in
the Knowledge Hub by how it can be re-populated when bringing up a fresh
Supabase project (the target re-ingestion environment per the S165 decision
locked at
.planning/.archive/.specs/content-format-standardisation-spec.md:1323-1324).
The matrix is the input to the two-stage re-ingestion runbook
(docs/operations/two-stage-re-ingestion-runbook.md) — Stage 1 ingests the Phew
.docx corpus + replayable sources via the canonical pipeline; Stage 2
(conditional) re-imports the client’s new markdown set if Stage 1 quality
justifies it.
Read alongside:
docs/reference/data-entry-points.md (entry-point inventory; refreshed S174)docs/operations/re-ingestion-quality-protocol.md (6-step protocol + DOCX
pre-processing checklist; shipped S168)docs/operations/database-rebuild-runbook.md (pre-existing rebuild runbook;
see §Reconciliation below)
| Category | Definition | Confidence at restore |
|---|
| Replayable-from-source | Source artefact (URL / .docx / .md) preserved; re-run through the canonical pipeline | High |
| Re-seedable | Static config / taxonomy / template — restore via SQL dump or seed migration | High |
| Manual-recreate | No source artefact; must be manually re-authored (e.g. MCP-created guides via Claude Desktop) | Medium (depends on author availability) |
| Export-then-reimport | No source artefact but DB content can be exported as markdown and re-ingested via the canonical pipeline | Medium (loses some metadata) |
| Lost-acceptable | Historical / audit / observability state; acceptable to lose on fresh instance | n/a (no restore attempt) |
Item counts are from live DB queries run 16/04/2026 against project
rovrymhhffssilaftdwd. Entry point attribution uses these markers:
| Signal | Entry point |
|---|
metadata->>'source' = 'intelligence_pipeline' | EP11 (RSS) |
metadata->>'ingestion_source' = 'markdown_file' | EP2 (Python MD) |
metadata->>'ingestion_source' = 'upload' | EP3 (File Upload) |
metadata->>'ingestion_source' = 'manual' | EP5 (Manual API) |
source_url != '' AND source_file IS NOT NULL | EP8 (Q&A Import) |
source_url IS NOT NULL AND source_file IS NULL AND file_path IS NULL AND metadata->>'source' IS NULL | EP1/EP4 (URL) |
| No standard markers (batch_tag only or nothing) | EP9 (MCP) |
Mapped to entry points 1–11 from data-entry-points.md. Plus side-table state
surfaces that are not entry points but require restore decisions.
| Field | Value |
|---|
| Category | Replayable-from-source |
| Source artefact | Original URLs. No canonical URL inventory file exists — extract from DB before re-ingestion. |
| Estimated item count | ~13 items (articles + policies with source_url set, no source_file, no file_path, no intelligence_pipeline metadata). Cannot distinguish EP1 (Python CLI) from EP4 (TS API) in existing data — both write source_url without a distinguishing marker. |
| Restore mechanism | Export URL list via SELECT source_url FROM content_items WHERE source_url IS NOT NULL AND source_file IS NULL AND file_path IS NULL AND metadata->>'source' IS NULL. Then python3 scripts/ingest.py --file urls.txt. |
| Resolution (S175) | No canonical URL inventory exists. The metadata.ingestion_source field is NULL for all 13 items (pre-dates the field). URL extraction query above is the only reliable method. Export this list to data/url-inventory-pre-reingest.txt as a pre-requisite step before wiping the old project. |
| Field | Value |
|---|
| Category | Replayable-from-source |
| Source artefact | .md files in documentation directory (Phew client corpus). 13 items have ingestion_source='markdown_file'. Additionally, 222 Q&A items written by EP8 (import_bid_library.py) also have source_file set — those are attributed to EP8, not EP2. |
| Estimated item count | 13 items (metadata->>'ingestion_source' = 'markdown_file'). |
| Restore mechanism | python3 scripts/ingest_markdown.py <dir> with --skip-existing, --tag, --author. |
| Resolution (S175) | Stage 1 vs Stage 2 question resolved. Stage 1 = Phew .docx corpus via EP8 + EP2 (existing markdown). Stage 2 (conditional) = client’s new markdown. The 13 markdown items are part of Stage 1. |
| Known gap | No chunking — need post-ingest bun run scripts/backfill-chunks.ts. |
| Field | Value |
|---|
| Category | Replayable-from-source (if storage files preserved) OR Export-then-reimport (if not) |
| Source artefact | Original PDF/DOCX/MD files in Supabase Storage documents bucket (13 files), tender-documents bucket (1 file), templates bucket (6 files). source_documents table (37 rows) provides lineage. |
| Estimated item count | 3 items (2 with file_path IS NOT NULL + 1 E2E upload sentinel with ingestion_source='upload'). |
| Restore mechanism | (a) Download files from old project’s Storage buckets → upload to new project → replay via POST /api/upload per file. OR (b) Export current content as markdown and re-ingest via EP2. |
| LIAM DECISION (S176) | Storage bucket strategy — RESOLVED. Files downloaded to docs/temp/. Investigation (S176 Wave 0) found: (a) documents bucket: 13 files are all E2E test artifacts (e2e-upload-sentinel-8-0-4.pdf duplicated from Playwright test runs), not production data; (b) tender-documents bucket: 1 PDF from manual testing of bid feature (only 1 draft bid workspace exists); (c) templates bucket: files from bid template testing (duplicated content). All 3 buckets contain only development/test artifacts — no production files. Buckets were created via Supabase dashboard (not config.toml or E2E tests). Decision: SKIP preservation. Buckets will be recreated empty by migrations (INSERT INTO storage.buckets now included in squashed migration). No production content at risk. |
| Field | Value |
|---|
| Category | Replayable-from-source |
| Source artefact | Original URLs (same as EP1 — source_url column). |
| Estimated item count | 0–13 items (cannot distinguish from EP1 in existing data; combined count is 13). |
| Restore mechanism | Use Python CLI (EP1) for bulk replay — same pipeline, no rate limit, no auth required. |
| Resolution (S175) | Recommend EP1 (Python CLI) for all bulk URL replay. The API route (EP4) adds rate limiting (10/min) and auth overhead with no benefit for batch operations. Collect all URLs into data/url-inventory-pre-reingest.txt and run via EP1. |
| Field | Value |
|---|
| Category | Lost-acceptable (negligible population, all test items) |
| Source artefact | None (web-form authored). |
| Estimated item count | 2 items (ingestion_source='manual'): “S167 Markdown Editor Test” and “S167 Chunking Pipeline Test”. Both created 14/04/2026 as development test artefacts. |
| Restore mechanism | None needed — test items, not production content. |
| Resolution (S175) | Reclassified from Export-then-reimport to Lost-acceptable. Both items are S167 development tests with no production value. No export script needed. If future manual items accumulate before re-ingestion, revisit — but current population is zero production items. |
| Field | Value |
|---|
| Category | Not applicable (never used for production data) |
| Source artefact | n/a |
| Estimated item count | 0 items (metadata->>'autosplit_batch_id' IS NOT NULL returns 0). |
| Restore mechanism | n/a |
| Resolution (S175) | No autosplit items exist. The batch API route exists and is tested but has never been used for production ingestion. All Q&A pairs were ingested via EP8 (Python import_bid_library.py). This entry point requires no restore action. |
| Field | Value |
|---|
| Category | Lost-acceptable (UPDATE-only path — no items created) |
| Source artefact | n/a |
| Restore mechanism | Re-run bun run scripts/batch-reclassify.ts --execute on items restored via EP1–EP8 if classification quality needs improvement post-re-ingestion. |
| Resolution (S175) | No open questions — this is a maintenance path, not an ingress. Post-re-ingestion reclassification may be needed if the initial keyword-only classification from EP8 is insufficient. |
| Field | Value |
|---|
| Category | Replayable-from-source |
| Source artefact | Original .docx files (Phew bid library). |
| Estimated item count | 222 items (content_type='q_a_pair' AND source_url = '', indicating Python import wrote empty-string source_url). Note: 2 additional Q&A items exist with source_url IS NULL (from TEST-pigeon-post-demo batch) — those are attributed to EP9, not EP8. |
| Restore mechanism | Re-run python3 scripts/import_bid_library.py <dir> --batch-tag <tag> against the same .docx corpus. |
| Resolution (S175) | Idempotency check is not an issue on clean DB. The check uses title-match against existing content_items. Against a blank DB, no existing items → no duplicates found → all items inserted. --force is not implicitly invoked — the check simply passes benignly. Autosplit grouping non-determinism (S174 question) is also moot: 0 autosplit items exist — all Q&A was imported via this script, not the batch API. |
| Field | Value |
|---|
| Category | Manual-recreate (no source artefact preserved) |
| Source artefact | Original Claude Desktop conversation prompts. No metadata markers distinguish these items — they lack ingestion_source, source_url, source_file, and file_path. |
| Estimated item count | ~13 content_items without standard provenance markers: 7 × consolidation-phew-march-2026 batch (company docs — Corporate Identity, Company Profile, Social Value, Business Continuity, Certifications, GDPR, InfoSec Controls), 3 × TEST-pigeon-post-demo batch (LMS Product Description, Bid Q&A, Security Q&A), 1 × Telehouse fact sheet, 1 × SI Feed brief, 1 × Shaw Education Trust case study. Plus 9 guides (117 guide_sections) in the guides table — created via MCP create_guide tool, separate from content_items. |
| Restore mechanism | Re-run original Claude Desktop prompts against the new project’s MCP server. A/B compare current-DB items vs new. |
| Resolution (S175) | Prompt archive is WP3 scope (docs/operations/guide-regeneration-prompts.md). The 7 consolidation-phew items are company reference docs that could alternatively be re-authored from the original .docx sources if the prompts are unrecoverable. The 3 pigeon-post items are demo/test content — Lost-acceptable. The 9 guides + 117 sections are the highest-priority manual-recreate items (sector/product/company guides actively used in the MCP workflow). |
| Field | Value |
|---|
| Category | Not applicable (no bid data exists) |
| Source artefact | n/a |
| Estimated item count | 0 items. bid_questions: 0 rows. bid_responses: 0 rows. bid_response_history: 0 rows. Only 1 bid workspace exists (name: “BROKEN_NAME_PHASE4”, status: draft) — a development artefact. |
| Restore mechanism | n/a. After bid workspace tables are restored (if any real bids exist by then), re-run POST /api/bids/:id/outcome/integrate for each won bid. |
| Resolution (S175) | No batch endpoint exists — per-bid only. Moot for current data (0 bids). If real bids accumulate before re-ingestion, a batch script would need writing. The bid workspace schema is restored by migrations; bid data is populated by users through the UI. |
| Field | Value |
|---|
| Category | Re-seedable (feed config) + Replayable-from-source (promoted articles via their URLs) + Lost-acceptable (historical unpromoted articles) |
| Source artefacts | (a) feed_sources (25 rows) + feed_prompts (3 rows) = re-seedable config. (b) 75 promoted articles have content_item_id IS NOT NULL — their source_url (from feed_articles.external_url) enables replay via EP1. (c) 339 unpromoted feed_articles = historical LLM filter scores only. |
| Estimated item count | 75 promoted items in content_items (identified by metadata->>'source' = 'intelligence_pipeline'). 414 total feed_articles. 25 feed_sources. 3 feed_prompts. |
| Restore mechanism | (1) Re-seed feed_sources + feed_prompts via SQL dump. (2) Re-ingest promoted-article URLs via EP1 (Python CLI) using exported URL list. (3) After re-ingestion, re-link feed_articles.content_item_id to new content_items.id UUIDs via URL match. (4) Cron resumes polling — new articles flow normally. |
| LIAM DECISION (S176) | Historical feed_articles preservation — RESOLVED. Preserve feed_articles snapshot for promoted rows only (75). Accept loss of unpromoted rows (339) — intermediate LLM filter state with no user-facing value. FK re-link of content_item_id to new UUIDs via URL match is needed post-re-ingestion (Step 14 in restore order). |
| Known gap | No chunking — need post-ingest bun run scripts/backfill-chunks.ts. |
41 tables exist in public schema. Organised by restore category.
| Surface | Row count | Restore mechanism | Notes |
|---|
taxonomy_domains | 15 | Seed migration or bun run sync:taxonomy | Taxonomy is DB-driven; sync:taxonomy regenerates classification prompt + plugin files. No subtopic key rename issue — keys are stable strings. |
taxonomy_subtopics | 57 | Seed migration or bun run sync:taxonomy | Same as above. |
workspaces | 4 | SQL dump with UUID preservation | 1 bid workspace (draft), 2 intelligence workspaces, 1 GNews dedup workspace. UUIDs must be preserved — content_item_workspaces has FK dependency. |
content_item_workspaces | 75 | Re-created after workspace + content_item restore | FK to both workspaces.id and content_items.id. Must restore after both parent tables. UUIDs will change for content_items — requires re-linking. |
user_roles | 8 | bun run seed:e2e-users + manual admin seeding | get_user_role() function created by migrations. E2E users auto-seeded. Production users need manual role assignment. |
entity_aliases | 24 | SQL dump | Config-like lookup table. No FK dependencies. BLOCKING: no CREATE TABLE migration exists — supabase db push on a fresh project will NOT create this table. A migration must be authored before re-ingestion execution. |
layer_vocabulary | 4 | SQL dump or seed migration | Layer display labels. |
company_profiles | 1 | SQL dump | Single company profile (Phew Design Ltd). |
template_requirements | 96 | SQL dump | Template requirement definitions. No FK to content_items. |
templates | 0 | n/a (empty) | Schema exists; no data to restore. |
template_fields | 0 | n/a (empty) | Schema exists; no data to restore. |
template_completions | 0 | n/a (empty) | Schema exists; no data to restore. |
content_templates | 0 | n/a (empty) | Schema exists; no data to restore. |
coverage_targets | 0 | n/a (empty) | Schema exists; no data to restore. |
governance_config | 0 | n/a (empty) | Schema exists; no data to restore. |
review_assignments | 0 | n/a (empty) | Schema exists; no data to restore. |
verification_history | 1 | SQL dump (optional) | Single row; low value. |
feed_sources | 25 | SQL dump (see EP11) | RSS/Atom feed configuration. |
feed_prompts | 3 | SQL dump (see EP11) | LLM filter prompt templates. |
feed_flags | 1 | SQL dump (optional) — must restore AFTER feed_articles (Step 9) | Single flag row. FK: feed_article_id → feed_articles(id), prompt_version_id → feed_prompts(id). |
| Surface | Row count | Restore mechanism | Notes |
|---|
guides | 9 | Prompt-based recreation via Claude Desktop (see EP9). SQL dump as backup. | 4 sector, 3 product, 1 company, 1 research guide. Preferred path is prompt recreation for quality validation; SQL backup preserves content if prompts unrecoverable. |
guide_sections | 117 | Coupled to guides — recreated atomically. SQL dump as backup. | 9–19 sections per guide. |
| Surface | Row count | Notes |
|---|
entity_mentions | 714 | Re-populated by classifyContent() during re-ingestion. Clean re-classification is the point. |
entity_relationships | 3,636 | Same — regenerated during classification. |
content_chunks | 444 | Re-populated by regenerateChunks() for entry points that invoke it; backfill required for EP2/EP6/EP10/EP11 gaps. |
content_history | 36 | New v1 row per item created by canonical ingest pipelines. Pre-restore history is lost. Acceptable. |
content_citations | 0 | Empty; will be regenerated if citation features are used post-restore. |
| Surface | Row count | Notes |
|---|
pipeline_runs | 43 | Observability data. New runs created during re-ingestion. |
notifications | 133 | UI notification state. |
read_marks | 1 | Per-user UI state. |
digests | 1 | Single change report. See decision note below. |
ingestion_quality_log | 3 | Quality logging from Python pipeline. New entries created during re-ingestion. |
processing_queue | 3 | Ephemeral processing state. |
si_processing_queue | 2,206 | Sector Intelligence processing records. Historical — new runs create new entries. |
source_documents | 37 | Lost if Storage bucket not preserved (see EP3 decision). Restored if bucket preserved + files re-uploaded. |
source_document_diffs | 0 | Empty; coupled to source_documents. |
Digests decision (S176 — Liam confirmed): 1 digests row exists. Change
reports are generated fresh from content_history data. Post-re-ingestion, the
new content_history will be all v1 records — no meaningful change report
possible until content evolves. Accept loss — confirmed.
| Surface | Row count | Notes |
|---|
bid_questions | 0 | No real bids exist. |
bid_responses | 0 | No real bids exist. |
bid_response_history | 0 | No real bids exist. |
Workspace type='bid' | 1 | ”BROKEN_NAME_PHASE4” (draft) — development artefact. Do not restore. |
| Surface | Row count | Notes |
|---|
feed_articles (promoted, content_item_id IS NOT NULL) | 75 | Snapshot-preserve. Re-link content_item_id to new UUIDs after re-ingestion (Step 14). |
feed_articles (unpromoted) | 339 | Lost-acceptable. Intermediate LLM filter state with no user-facing value. See EP11 Liam decision. |
Note: The skeleton referenced bids and bid_workspaces tables — these do
not exist. Bid data lives in bid_questions/bid_responses linked to
workspaces with type='bid'. No production bid data exists; all tables are
empty or contain test artefacts.
| Bucket | Files | Notes |
|---|
documents | 13 | Private. Source files for EP3 uploads. Per-project — cannot share. |
tender-documents | 1 | Private. Single tender document. |
templates | 6 | Private. Template files. |
Total: 20 files. See EP3 decision for preservation strategy.
FK dependency chain determines ordering. Steps marked with ★ require operator
action beyond running a script.
Pre-requisite: Migration squash (S176 — COMPLETE). 43 migrations squashed to
1 (20260416102457_pre_squash_reconciliation.sql, 8,600 lines). Gaps found and
fixed: entity_aliases table (missing entirely), digests table (9 missing
columns), workspaces + processing_queue (missing created_by / updated_by
columns), 4 functions created via MCP (claim_next_job,
content_history_auto_version, get_source_documents,
get_verification_stats), 46 ALTER FUNCTION signature mismatches, 3 storage
buckets (missing INSERT statements), 6 storage RLS policies (missing for
documents + tender-documents buckets). Full reconciliation report:
docs/operations/schema-reconciliation-report-s176.md.
| Step | Action | Dependency | Est. time |
|---|
| 0 | ★ New Supabase project provisioned (eu-west-2, pgvector 0.8.0). Note project ID. Populate env vars. | Migration squash complete | 10 min |
| 1 | Apply migrations: /opt/homebrew/bin/supabase db push (single squashed migration) | Step 0 | 2 min |
| 2 | Regenerate types: supabase gen types typescript --project-id mgrmucazfiibsomdmndh --schema public > supabase/types/database.types.ts | Step 1 | 1 min |
| 3 | Seed auth users: bun run seed:e2e-users + verify pipeline service account | Step 1 | 2 min |
| 4 | Seed user_roles for production users (manual INSERT or dump-restore) | Step 3 | 5 min |
| 5 | Seed static config: taxonomy_domains, taxonomy_subtopics via bun run sync:taxonomy | Step 1 | 2 min |
| 6 | Seed remaining config: entity_aliases (24), layer_vocabulary (4), company_profiles (1), template_requirements (96) via SQL dump | Step 1 | 5 min |
| 7 | Restore workspaces (4 rows, with UUIDs preserved) | Step 1 | 2 min |
| 8 | Restore feed_sources (25) + feed_prompts (3) | Step 7 (workspace FK) | 5 min |
| 9 | ★ (If preserving) Restore feed_articles snapshot — promoted rows only (75), with content_item_id = NULL (re-linked in Step 14) | Step 8 | 10 min |
| 9b | Restore feed_flags (1 row) via SQL dump. FK: feed_article_id → feed_articles(id), prompt_version_id → feed_prompts(id) | Step 8, Step 9 | 1 min |
| 10 | Storage buckets — SKIPPED (S176 decision: all buckets contain only test artifacts). Buckets auto-created empty by migration. | Step 0 | 0 min |
| 11 | Stage 1 re-ingest — Phew .docx corpus: python3 scripts/import_bid_library.py <dir> --batch-tag phew-reingest-2026 (EP8, ~222 items) | Step 5 (taxonomy needed for classification) | ~30 min |
| 12 | Stage 1 re-ingest — markdown files: python3 scripts/ingest_markdown.py <dir> --tag phew-markdown-reingest (EP2, ~13 items) | Step 5 | ~5 min |
| 13 | Stage 1 re-ingest — URL inventory: python3 scripts/ingest.py --file data/url-inventory-pre-reingest.txt (EP1, ~13 items) | Step 5 | ~10 min |
| 14 | Re-link feed_articles.content_item_id for promoted rows via URL-match SQL (see runbook Step 14 for SQL) | Step 9, Step 13 | 10 min |
| 15 | Backfill chunks: bun run scripts/backfill-chunks.ts (for EP2/EP8/EP11 gap items) | Steps 11–13 | ~15 min |
| 16 | Re-ingest RSS-promoted URLs: extract URLs from preserved feed_articles → run via EP1 (75 items) | Steps 5, 8 | ~15 min |
| 17 | Re-link content_item_workspaces for intelligence workspace items | Steps 7, 16 | 5 min |
| 18 | ★ MCP guide regeneration (EP9): re-run Claude Desktop prompts for 9 guides + 117 sections. A/B compare. | Steps 11–13 (KB content needed for guide context) | 1–2 hours |
| 19 | ★ MCP content_item recreation (EP9): re-run prompts for ~13 orphan items (7 consolidation-phew, etc.) | Steps 11–13 | 30 min |
| 20 | ★ Stage 1 decision gate: Run quality protocol (snapshot, compare, human review). Proceed to Stage 2 only if all thresholds met. | Steps 11–19 | 1 hour |
| 21 | (Conditional) Stage 2 — client new markdown: python3 scripts/ingest_markdown.py <new-dir> + backfill chunks | Step 20 pass | Variable |
| 22 | Post-Stage-2 quality comparison: A/B against Stage 0 + Stage 1 snapshots | Step 21 | 30 min |
| 23 | Embedding smoke test: bun run scripts/embedding-smoke-test.ts | Steps 11–21 | 5 min |
| 24 | Final snapshot: bun run scripts/snapshot-content-state.ts | Step 23 | 5 min |
FK dependency justification:
content_item_workspaces → workspaces.id + content_items.id (Step 17
after Steps 7 + 11–16)feed_articles.content_item_id → content_items.id (Step 14 after Steps
9 + 13)feed_articles.feed_source_id → feed_sources.id (Step 9 after Step 8)feed_prompts.workspace_id → workspaces.id (Step 8 after Step 7)feed_sources.workspace_id → workspaces.id (Step 8 after Step 7)entity_mentions.content_item_id → content_items.id (auto-generated during
Steps 11–16)entity_relationships.source_item_id → content_items.id (auto-generated
during Steps 11–16)user_roles.user_id → auth.users.id (Step 4 after Step 3)guide_sections.guide_id → guides.id (Step 18 creates both atomically)
The existing docs/operations/database-rebuild-runbook.md covers four
scenarios: fresh demo DB, Phew re-ingest, disaster recovery, and local dev
reset. This restore matrix and the two-stage re-ingestion runbook (WP2)
supersede the “Phew re-ingest” scenario specifically.
| Runbook section | Overlap | Action needed |
|---|
| §3 Step 1 (db reset) | Overlaps — but the restore matrix uses a new project, not a reset of the existing one. | No change needed. The runbook’s reset path is for same-project rebuilds; the matrix’s new-project path is distinct. |
| §4 Step 2 (push migrations) | Identical — both use supabase db push. | No change needed. |
| §5 Step 3 (gen types) | Identical. | No change needed. |
| §6 Step 4 (seed users) | Identical — bun run seed:e2e-users. | No change needed. |
| §8 Step 6 (seed app data) | Partial overlap. The runbook’s “For Phew re-ingest” subsection points at ingest_markdown.py only. The restore matrix adds EP1 URL replay, EP8 Q&A import, EP11 RSS re-ingest, EP9 guide recreation, chunk backfill, and quality protocol. | Recommend: Add a sentence to §8 “For Phew re-ingest” pointing to the restore matrix and two-stage runbook as the authoritative procedure. Do not duplicate the full restore order in the runbook. |
| §12 Quality gate | Points at re-ingestion-quality-protocol.md. The restore matrix’s Step 20 (decision gate) invokes the same protocol. | No change needed — cross-reference is already correct. |
Gap identified: The runbook does not mention the new-project-ID placeholder
or storage bucket migration. These are covered exclusively by the restore matrix
and two-stage runbook. The runbook should add a cross-reference line:
For the Phew re-ingest into a new Supabase project, follow the
Blank-DB Restore Matrix and the
Two-Stage Re-Ingestion Runbook.
Before deleting or decommissioning the old project, extract these artefacts:
- S174 WP5: Skeleton (all entry points + side tables categorised; open
questions flagged).
- S175 WP1: DRAFT — inventory queries populated; open questions resolved or
escalated; restore order locked; reconciliation with rebuild runbook
documented; pre-export checklist added; missing side tables added (17 tables
not in skeleton).
- Post-provisioning (S176 — DONE): Project ID
mgrmucazfiibsomdmndh, URL
https://mgrmucazfiibsomdmndh.supabase.co. Placeholders updated in runbook
and matrix.
- Post-Stage-1 execution: Mark cells with actual restore behaviour; promote
to VALIDATED.
