Canonical resource access inventory

This inventory tracks code that reads or writes the canonical resource graph. It is intentionally organized by ownership and data flow rather than by table name alone. Keep it current when a resource access path is added, removed, or moved.

Canonical schema: web-tanstack/prisma/schema.prisma.

Invariants

resources owns canonical content metadata, lifecycle state, and nullable monitored-source provenance through source_id. Product semantics live in kind, special platform identity in resource_platform, and representation format/blob presence in file_type / storage_key.
The only valid (kind, resource_platform) pairs are blog/null, forum/hackernews, post/twitter, video/youtube, paper/null, image/null, and file/null.
resource_translations owns localized title, summary, and content. Resource-level tags and keywords belong to resources.
user_resources is the stable user-to-Resource fact; nullable profile_saved_at records an explicit profile save.
resource_placements optionally locates that fact in one collection or workspace and enforces common ownership. Workspace role distinguishes sources from results independently of file origin.
user_files owns private raw-blob attachment and filename; resources.byte_size owns quota size.
sources owns monitored feed configuration. Its kind (blog | news) is a separate feed/editorial classification; both source kinds produce the canonical blog Resource after successful processing. resources.source_id is the stable relation; non-monitored resources may keep source_id = NULL.
The app Worker owns user-facing save/file/container and blob lifecycle writes. Core owns acquisition, enrichment, translations, entities, and AI Search synchronization. Think uses app/core service bindings and does not query product-owned resource tables directly.

Schema and database-only objects

Path	Role	Audit status
`web-tanstack/prisma/schema.prisma`	Canonical Prisma schema	audited
`web-tanstack/prisma/manual-indexes.sql`	Constraints, indexes, and triggers	audited; canonical source/effective-date indexes back keyset reads, while obsolete publication-only and JSON source-name indexes are explicitly removed
`workers/core-worker/scripts/check-resource-identity.mjs`	Resource processing and identity contract	checks the Core kind/platform vocabulary against the database identity constraint; schema history is verified by migration replay and drift checks

Completed cutovers and one-shot data repairs live in git history, not beside the current schema where they could be mistaken for repeatable operator work.

App Worker: canonical reads

Path	Surface
`web-tanstack/src/server/resource-model.server.ts`	Shared Prisma projections, access predicates, localization, and persisted Resource identity mapping
`web-tanstack/src/server/resource-query.server.ts`	Resource detail, feed/search, ranked hydration, collection-preview, and validated media read queries
`web-tanstack/src/server/knowledge-service.ts`	Authorized and byte-bounded knowledge search/read plus flattened workspace context shared by Think Domain RPC and HTTP MCP
`web-tanstack/src/server/profile.functions.ts`	Current/public profile and paginated profile-resource Server Functions
`web-tanstack/src/server/collection.functions.ts`	Collection CRUD, detail, and paginated Resource-list Server Functions
`web-tanstack/src/server/resource-ingest.functions.ts`	Resource relationship, URL import, and Workflow-status Server Functions
`web-tanstack/src/server/resource-read.functions.ts`	Public feed, Resource detail rendering, and related-resource Server Functions
`web-tanstack/src/server/docs.ts`	Document/workspace persistence, Markdown conversion, versioning/sharing, and its TanStack Server Function surface
`web-tanstack/src/routes/api.mcp.ts`	Anonymous, stateless, read-only MCP boundary over the shared knowledge service
`web-tanstack/src/routes/api.media.asset.$.ts`, `web-tanstack/src/routes/api.media.image.$id.ts`	Authorized app-owned blob and Hosted Images delivery
`web-tanstack/src/lib/resource/{model,feed,save,toc}.ts`	Domain-owned browser/server Resource DTOs, feed vocabulary, save state, and reader navigation types

App Worker: authoritative writes

Path	Surface
`web-tanstack/src/server/resource-write.server.ts`	Resource-save, collection/workspace relationship, existing-URL admission, failure, and access transactions
`web-tanstack/src/server/collection.functions.ts`	Collection CRUD and atomic collection-plus-Resource creation
`web-tanstack/src/server/resource-ingest.functions.ts`	Resource relationship mutations, URL draft admission, and Workflow status reads
`web-tanstack/src/server/resource-workflow.server.ts`	Server-only signing and verification for user-bound Resource Workflow status tokens
`web-tanstack/src/server/billing.functions.ts`	Authenticated Autumn customer, checkout, and billing-portal endpoints
`web-tanstack/src/server/podcast.functions.ts`	Authorized workspace podcast reads
`web-tanstack/src/server/user-file.functions.ts`	User-file listing, reference guards, and deletion
`web-tanstack/src/server/user-file-persistence.server.ts`	Upload validation, quota, deduplication, and resource persistence
`web-tanstack/src/server/blob-lifecycle.server.ts`	Hosted Images/R2 provider operations and scheduled durable cleanup reconciliation
`web-tanstack/src/lib/media/upload.ts`	The one browser upload call: native `FormData`, shared admission constants, and the direct success payload
`web-tanstack/src/server/docs.ts`	Workspace/document creation, edits, optimistic versioning, sharing, and deletion
`web-tanstack/src/server/chat-sessions.ts`	Chat allocation, ownership, action receipts, metadata, deletion fencing, and reconciliation
`web-tanstack/src/server/domain-rpc.ts`	Request-scoped Domain RPC boundary used by Core and Think
`web-tanstack/src/server/podcast.server.ts`	Podcast creation, completion, failure, and durable object cleanup admission
`web-tanstack/src/server/chat-agent-proxy.server.ts`	Authenticated same-origin Think Agent HTTP/WebSocket proxy
`web-tanstack/src/server/maintenance.server.ts`	Scheduled blob and chat-session reconciliation orchestration
`web-tanstack/src/server.ts`	Thin Cloudflare Worker composition root
`web-tanstack/src/routes/api.upload.ts`	Authenticated native `FormData` admission into the shared media/resource boundaries

Core Worker: engine reads and writes

Path	Role	Audit status
`workers/core-worker/src/ingest/stored-resource.ts`	Stored-blob and explicit-reacquisition reads	validates persisted identity and projects database rows into Core processing input
`workers/core-worker/src/ingest/analysis.ts`	Resource AI analysis and entity normalization	deterministic analysis output crosses the persistence boundary with the processed resource graph
`workers/core-worker/src/ingest/translation-workflow.ts`	Revision-fenced translation reads/writes	translation input loading and guarded persistence are colocated with their only Workflow consumer
`workers/core-worker/src/ingest/platforms/{rss,twitter}.ts`	Source discovery, dedup, provenance, and enqueue	each monitor owns its provider polling and draft admission
`workers/core-worker/src/ingest/workflow.ts`	Draft/stored-target acquisition, enrichment, and persist-last orchestration	URL/source/subject drafts finalize through one create-or-merge transaction; uploaded blobs and resyncs load stored rows explicitly
`workers/core-worker/src/ingest/acquisition.ts`, `workers/core-worker/src/ingest/platforms/web.ts`	Acquisition routing and generic web extraction	platform detection selects the dedicated acquisition path before canonical persistence
`workers/core-worker/src/search-index.ts`	Runtime AI Search projection, synchronization, deletion, and candidate ranking	`newsence-corpus-v6` stores only effective date, source ID, kind, and resource platform as custom metadata; grouped searches give direct kind matches and HN expansion independent native 50-candidate budgets
`workers/core-worker/src/operations/search-index-rebuild.ts`	Operator-triggered full-index rebuild and readiness	the durable Workflow uploads the live corpus plus delta, prunes stale items, and verifies native item counts
`workers/core-worker/src/index.ts`	RPC/workflow entrypoints	bounded batch RPCs wrap native Workflow `createBatch`, status fan-out, and grouped AI Search without moving product writes into Core

Think Worker consumers

Path	Role	Audit status
`workers/think-worker/src/agent.ts`	Native Think `knowledge` tool and app Domain RPC consumer	one typed action surface covers public search, typed reads, and current-workspace sources; RPC values are cloned and disposed
`workers/think-worker/src/tools/create-document.ts`	Document edit/create and generated-image actions	visible product mutations cross the app Domain boundary; Computer files remain scratch only
`web-tanstack/src/server/knowledge-service.ts`	App-owned read authority	validates Think and MCP requests and executes authorized, bounded knowledge reads

Frontend consumers

The frontend must consume server-owned DTOs and must not reconstruct database ownership rules. These consumers were audited during the read-layer refactor; resource/card identity is explicit, source filter state stores Source UUIDs rather than display names, duplicate profile/collection adapters and localization wrappers were removed, and effective dates now come from server policy. Direct consumers live in:

web-tanstack/src/lib/queries/* (including the single server-owned viewer save-state projection)
web-tanstack/src/routes/_workspaceApp._noResources.index.tsx
web-tanstack/src/components/feed/*
web-tanstack/src/components/resource/*
web-tanstack/src/lib/resource-workflow.ts (one normalized result list for URL/upload invalidation and polling)
web-tanstack/src/components/workspace/ResourcesRail.tsx
web-tanstack/src/components/collections/*
web-tanstack/src/components/chat/*
web-tanstack/src/lib/resource/{model,feed,save,toc}.ts

Completion searches

Run these from the repository root and classify every new match before marking the resource refactor complete:

rg -n --glob '!**/generated/**' \
  '\b(prisma|tx)\.(resource|resourceTranslation|userResource|resourcePlacement|userFile|source)\.' \
  web-tanstack/src

rg -n -i --glob '!**/generated/**' \
  '(from|join|update|into|delete from)[[:space:]]+"?resources"?|resource_translations|user_resources|resource_placements|user_files' \
  web-tanstack/src workers packages

Historical SQL and documentation matches are not runtime touchpoints, but must remain clearly labeled so stale models are not copied back into production code.

Completion audit

The completion searches above were rerun after round 9 and every runtime match maps to an inventory entry; remaining SQL matches are canonical schema, operator migrations, or documentation.
Live read-only invariant checks found no enriched corpus row missing its original translation, URL, title, or source; no duplicate normalized URL, invalid kind/platform pair, invalid library item, ownerless private resource, or user-file row without a storage key.
All resource, translation, profile-item/placement, and user-file indexes declared in manual-indexes.sql are present in the live database.
Final verification uses the web check, typecheck, and production build; core database contract, typecheck, and lint; Think typecheck and lint; plus real workflow, scheduled cron, AI Search, search RPC, related-resource, and PostgreSQL smoke checks.