@vantageos/backend-doctor 0.1.0

package/fixtures/vp-1200-green/convex/schema.ts

@@ -0,0 +1,1243 @@
+import { defineSchema, defineTable } from "convex/server";
+import { v } from "convex/values";
+
+// ─────────────────────────────────────────────────────────────────────────────
+// Shared validators (reused across tables)
+// ─────────────────────────────────────────────────────────────────────────────
+
+export const memoryTypeValidator = v.union(
+	v.literal("user"),
+	v.literal("feedback"),
+	v.literal("project"),
+	v.literal("reference"),
+	v.literal("episode"),
+);
+
+// Open validator — any orchestrator name is accepted.
+// Known defaults: pi, tau, phi, sigma, omega, zeta, eta, kappa, alpha, lambda,
+// victor, proxima, theta, xi, epsilon, omicron, upsilon, system.
+// New orchestrators can be added without schema changes (see issue #132).
+export const creatorValidator = v.string();
+
+// Day 130 follow-up #2 (Eta REVISE, PR #1089) — inforgeable automation
+// signal. `createdBy` is a caller-supplied string on the PUBLIC
+// `tasks.create` mutation, so it can be forged (e.g. `createdBy: "system"`)
+// to escape the billing closure gate. `origin` is NOT accepted as an
+// argument on any public mutation — only the internal webhook path
+// (createOrUpdateReviewTask) writes it. The closure gate reads `origin`,
+// never `createdBy`, to decide whether a task is automation-created.
+export const taskOriginValidator = v.literal("automation");
+
+export const relationTypeValidator = v.union(
+	v.literal("updates"),
+	v.literal("extends"),
+	v.literal("derives"),
+);
+
+export const severityValidator = v.union(
+	v.literal("critical"),
+	v.literal("major"),
+	v.literal("minor"),
+);
+
+// ─────────────────────────────────────────────────────────────────────────────
+// Schema
+// ─────────────────────────────────────────────────────────────────────────────
+
+export default defineSchema({
+	// ── memories ────────────────────────────────────────────────────────────────
+	// Core memory store. Each row is one typed memory entry.
+	// Supermemory pattern: updates/extends/derives relations + isLatest flag.
+	// Mem0 pattern: typed memory (user/feedback/project/reference/episode).
+	// Episodic type includes structured episode metadata.
+	//
+	// Embedding + search is handled by @convex-dev/rag (component-owned schema).
+	// RAG entry key = memoryId string. Filters: namespace, type, isLatest.
+	memories: defineTable({
+		// Namespace: "global" | "orchestrator/pi" | "project/vantage-starter"
+		namespace: v.string(),
+
+		// Memory classification — drives retrieval strategy
+		type: memoryTypeValidator,
+
+		// Human-readable content (what the memory says)
+		content: v.string(),
+
+		// Which orchestrator (or system) created this memory
+		createdBy: creatorValidator,
+		instanceId: v.optional(v.string()), // which instance wrote this
+
+		// Graph relations to other memories (Supermemory pattern)
+		// "updates" supersedes target (sets target.isLatest = false)
+		// "extends" adds detail to target (both remain latest)
+		// "derives" is an inference drawn from target
+		relations: v.array(
+			v.object({
+				targetId: v.id("memories"),
+				type: relationTypeValidator,
+			}),
+		),
+
+		// True = this is the current authoritative version.
+		// False = superseded by a newer "updates" relation.
+		// Search always filters isLatest=true by default.
+		isLatest: v.boolean(),
+
+		// Optional TTL hint (ISO string). Cron job handles actual expiry.
+		// Example: "2026-06-01T00:00:00Z"
+		ttl: v.optional(v.string()),
+
+		// Episodic memory payload — only present when type="episode"
+		episode: v.optional(
+			v.object({
+				context: v.string(), // Situation that triggered this episode
+				goal: v.string(), // What was being attempted
+				action: v.string(), // What was actually done
+				outcome: v.string(), // What happened
+				insight: v.string(), // The lesson extracted (procedural memory)
+				severity: severityValidator,
+			}),
+		),
+
+		// Timestamp (ms since epoch)
+		createdAt: v.number(),
+		updatedAt: v.number(),
+	})
+		// Primary query patterns — all filtered to isLatest for active memory reads
+		.index("by_namespace", ["namespace", "isLatest"])
+		.index("by_type", ["type", "isLatest"])
+		.index("by_creator", ["createdBy", "isLatest"])
+		.index("by_namespace_type", ["namespace", "type", "isLatest"]),
+
+	// ── profiles ────────────────────────────────────────────────────────────────
+	// One row per INSTANCE. orchestratorId = role (pi/tau/phi).
+	// instanceId = unique running copy (pi-chromebook, pi-vps, tau-client-acme).
+	// Multiple instances can share the same role.
+	// static  = stable identity facts (role, workspace, capabilities)
+	// dynamic = mutable session state (current task, last seen, session count)
+	profiles: defineTable({
+		// Role: "pi" | "tau" | "phi"
+		orchestratorId: v.string(),
+
+		// Unique instance identifier: "pi-chromebook" | "pi-vps" | "tau-vps-1"
+		instanceId: v.optional(v.string()),
+
+		name: v.string(),
+
+		// Static profile — stable facts, infrequently updated
+		static: v.object({
+			role: v.string(),
+			workspace: v.string(),
+			capabilities: v.array(v.string()),
+		}),
+
+		// Dynamic profile — updated each session
+		dynamic: v.object({
+			currentTask: v.optional(v.string()),
+			// Durable end-of-day index (close-day step 9) — distinct from
+			// `currentTask` (live session summary) so a next-morning
+			// daily-start write does not clobber it. mission k574p02m DEFECT 1.
+			endOfDayIndex: v.optional(v.string()),
+			lastSeen: v.number(), // ms since epoch
+			sessionCount: v.number(),
+		}),
+	})
+		.index("by_orchestrator", ["orchestratorId"])
+		.index("by_instance", ["instanceId"]),
+
+	// ── messages ──────────────────────────────────────────────────────────────
+	// Inter-orchestrator messaging. VantagePeers native (supersedes pre-VP legacy).
+	// One message row per send. Recipients tracked via messageReceipts table.
+	// channel: "broadcast" | "pi" | "tau" | "phi" | "pi,tau"
+	messages: defineTable({
+		from: creatorValidator,
+		fromInstanceId: v.optional(v.string()), // "pi-chromebook", "tau-vps-1"
+		tenantId: v.optional(v.string()),
+		channel: v.string(),
+		content: v.string(),
+		sessionDay: v.optional(v.number()),
+		createdAt: v.number(),
+	})
+		.index("by_day", ["sessionDay"])
+		.index("by_from", ["from", "createdAt"])
+		.index("by_channel", ["channel", "createdAt"])
+		.index("by_tenant_channel", ["tenantId", "channel", "createdAt"])
+		// Day 103 — tenant-scoped listing for Clerk callers (task k176wgsrhha0fr0dxxahctvhw588q5a1).
+		// by_tenant_created lets listMessages push tenantId BEFORE .take(limit) for
+		// non-master callers, preventing under-fill when fleet (null-tenant) traffic
+		// dominates the recent window. by_tenant_channel cannot be reused here because
+		// channel is a required equality field in that compound (can't skip to createdAt).
+		.index("by_tenant_created", ["tenantId", "createdAt"])
+		// Day 102 v2.11.0 — CRUD baseline PR-C-bis option B (mission k575kc1r):
+		// Convex native BM25 search on message body, with filterFields for the
+		// common audit narrowing axes (from, channel, sessionDay).
+		.searchIndex("search_content", {
+			searchField: "content",
+			filterFields: ["from", "channel", "sessionDay", "tenantId"],
+		}),
+
+	// ── messageReceipts ──────────────────────────────────────────────────────
+	// One row per recipient per message. Tracks read status.
+	// check_new_messages = query receipts where recipient=X AND readAt=undefined
+	messageReceipts: defineTable({
+		messageId: v.id("messages"),
+		recipient: creatorValidator, // role-level: "pi" | "tau" | "phi"
+		recipientInstanceId: v.optional(v.string()), // instance-level: "pi-vps"
+		tenantId: v.optional(v.string()),
+		readAt: v.optional(v.number()), // undefined = unread, ms epoch = read
+	})
+		.index("by_recipient_unread", ["recipient", "readAt"])
+		.index("by_instance_unread", ["recipientInstanceId", "readAt"])
+		.index("by_message", ["messageId"])
+		.index("by_tenant_recipient_unread", ["tenantId", "recipient", "readAt"]),
+
+	// ── missions ──────────────────────────────────────────────────────────────
+	missions: defineTable({
+		name: v.string(),
+		description: v.optional(v.string()),
+		project: v.string(),
+		status: v.union(
+			v.literal("brainstorm"),
+			v.literal("plan"),
+			v.literal("execute"),
+			v.literal("validate"),
+			v.literal("complete"),
+			v.literal("cancelled"),
+		),
+		priority: v.union(
+			v.literal("urgent"),
+			v.literal("high"),
+			v.literal("medium"),
+			v.literal("low"),
+		),
+		pilot: creatorValidator,
+		agents: v.array(v.string()),
+		brief: v.optional(v.string()),
+		startDate: v.optional(v.number()),
+		targetDate: v.optional(v.number()),
+		progress: v.optional(v.number()),
+		createdBy: creatorValidator,
+		createdAt: v.number(),
+		updatedAt: v.number(),
+		// Beta multi-tenant scope. null/undefined = master (internal Alpha).
+		// Set to Clerk org slug (e.g. "acme-hr") for client-scoped rows.
+		orgId: v.optional(v.string()),
+		// Terminal cancelled status (Day 157) — creator-only, mandatory reason.
+		// Never counted as done/complete; excluded from open/active aliases.
+		cancelledBy: v.optional(creatorValidator),
+		cancelReason: v.optional(v.string()),
+	})
+		.index("by_project", ["project", "status"])
+		.index("by_pilot", ["pilot", "status"])
+		.index("by_status", ["status", "createdAt"])
+		.index("by_priority", ["priority", "status"])
+		.index("by_orgId", ["orgId"]),
+
+	// ── tasks ──────────────────────────────────────────────────────────────────
+	tasks: defineTable({
+		title: v.string(),
+		description: v.optional(v.string()),
+		project: v.optional(v.string()), // "vantage-starter", "perfect-ai-agent", etc.
+		tags: v.optional(v.array(v.string())),
+		// Orchestrator or person assigned — open string (issue #132)
+		assignedTo: v.string(),
+		priority: v.union(
+			v.literal("urgent"),
+			v.literal("high"),
+			v.literal("medium"),
+			v.literal("low"),
+		),
+		status: v.union(
+			v.literal("todo"),
+			v.literal("in_progress"),
+			v.literal("review"),
+			v.literal("blocked"),
+			v.literal("done"),
+			v.literal("cancelled"),
+		),
+		completionNote: v.optional(v.string()), // what was actually done — written on complete/review
+		assignedToInstance: v.optional(v.string()), // instance-level assignment: "pi-vps", "tau-chromebook"
+		claimedByInstance: v.optional(v.string()), // which instance is working on this
+		dependsOn: v.optional(v.array(v.id("tasks"))), // tasks that must be done first
+		missionId: v.optional(v.id("missions")),
+		estimatedMinutes: v.optional(v.number()),
+		actualMinutes: v.optional(v.number()),
+		startedAt: v.optional(v.number()),
+		completedAt: v.optional(v.number()),
+		dueDate: v.optional(v.number()),
+		createdBy: creatorValidator,
+		createdAt: v.number(),
+		updatedAt: v.number(),
+		// Beta multi-tenant scope. null/undefined = master (internal Alpha).
+		// Set to Clerk org slug (e.g. "acme-hr") for client-scoped rows.
+		orgId: v.optional(v.string()),
+		// Day 130 follow-up #2 — inforgeable automation signal. ONLY the
+		// internal webhook mutation (createOrUpdateReviewTask) writes this;
+		// the public `tasks.create` mutation does not accept it as an arg.
+		// The billing closure gate reads this field, never `createdBy`
+		// (which is a caller-supplied, forgeable string on a public mutation).
+		origin: v.optional(taskOriginValidator),
+		// Terminal cancelled status (Day 157) — creator-only, mandatory reason.
+		// Never counted as done/complete; excluded from open/active aliases.
+		cancelledBy: v.optional(creatorValidator),
+		cancelReason: v.optional(v.string()),
+		// Day 159 — a block that names a commitment, not just a journal entry.
+		// blockedOnTaskId: the live task (assigned to someone else) this task
+		// waits on. blockTask validates existence/openness/other-assignee at
+		// call time; when the cited task reaches "done" every waiter here is
+		// swept back to "todo" (convex/tasks.ts unblockWaitersOn). Mutually
+		// exclusive with blockedOnNobodyReason in practice (blockTask sets
+		// exactly one), both kept optional so pre-existing "blocked" rows with
+		// neither field are valid (see migration list query
+		// convex/tasks.ts:listUnlinkedBlocked).
+		blockedOnTaskId: v.optional(v.id("tasks")),
+		// Explicit "# blocked-on-nobody: <reason>" marker text — required to
+		// block WITHOUT a blockedOnTaskId (nobody in the fleet owns the
+		// obstacle: operator decision, third-party outage). Forbids ANONYMOUS
+		// blocking without forbidding blocking itself.
+		blockedOnNobodyReason: v.optional(v.string()),
+	})
+		.index("by_assignee", ["assignedTo", "status"])
+		.index("by_project", ["project", "status"])
+		.index("by_priority", ["priority", "status"])
+		.index("by_status", ["status", "createdAt"])
+		.index("by_mission", ["missionId", "status"])
+		.index("by_instance", ["assignedToInstance", "status"])
+		.index("by_orgId", ["orgId"])
+		// Compound indexes added to close the silent-filter-drop defect in
+		// convex/tasks.ts `list`: when a caller supplies assignedTo/assignedToInstance
+		// TOGETHER with project, the query must apply BOTH filters via a matching
+		// index, never pick one and silently discard the other.
+		.index("by_assignee_project", ["assignedTo", "project", "status"])
+		.index("by_instance_project", ["assignedToInstance", "project", "status"])
+		// billingSummaryByProject (convex/tasks.ts) — the period MUST bound the
+		// query itself, not a post-hoc in-memory filter over a fixed-size scan
+		// (Day-131 live-defect fix: recent, billable work was silently invisible
+		// because the old handler capped the by_status(status, createdAt) scan
+		// BEFORE ever comparing completedAt against the requested window).
+		.index("by_status_completedAt", ["status", "completedAt"])
+		// Same fix, pushed one level further: when the caller also supplies a
+		// project filter, apply BOTH status+project+completedAt at the index
+		// level so a single-project billing query is never a post-hoc filter
+		// over a truncated cross-project scan (same disease, same fix).
+		.index("by_status_project_completedAt", ["status", "project", "completedAt"])
+		// Day-132 live-defect fix: `updatedSince` on `tasks.list` was filtered
+		// IN-MEMORY after a fixed-size widened scan, so narrowing the window
+		// never reduced the number of candidates fetched and the scan cap bit
+		// into normal usage on exactly two branches (measured in production):
+		// `assignedTo` alone, and `assignedTo` + `status`. These two compound
+		// indexes end in `updatedAt` so the bound lives in the query itself —
+		// no index added for missions/briefingNotes, whose branches were not
+		// measured to exceed the cap.
+		.index("by_assignee_updatedAt", ["assignedTo", "updatedAt"])
+		.index("by_assignee_status_updatedAt", ["assignedTo", "status", "updatedAt"])
+		// Day 159 — reciprocal unblock: find every task waiting on a given
+		// task so it can be swept back to "todo" the moment that task closes.
+		.index("by_blockedOnTaskId", ["blockedOnTaskId"])
+		// Day 102 v2.11.0 — CRUD baseline PR-C-bis option B (mission k575kc1r):
+		// Convex native BM25 search on task title, with filterFields for the
+		// common targeting axes (assignedTo, status, project, missionId).
+		.searchIndex("search_title", {
+			searchField: "title",
+			filterFields: ["assignedTo", "status", "project", "missionId", "orgId"],
+		}),
+
+	// ── diary ──────────────────────────────────────────────────────────────────
+	diary: defineTable({
+		date: v.string(), // "2026-03-25" ISO date
+		orchestrator: creatorValidator,
+		instanceId: v.optional(v.string()), // which instance wrote this
+		content: v.string(), // Full diary entry
+		highlights: v.optional(v.array(v.string())),
+		blockers: v.optional(v.array(v.string())),
+		// v2.4.8: auth-derived author captured at write time from oauthCtx.userId.
+		// Distinct from `orchestrator` (writer-intent label, client-supplied).
+		// Pre-v2.4.8 entries are backfilled with orchestrator as best-guess via
+		// migrations/diary_backfill_createdBy — NOT auth-verified.
+		createdBy: v.optional(creatorValidator),
+		createdAt: v.number(),
+	})
+		.index("by_orchestrator_date", ["orchestrator", "date"])
+		.index("by_date", ["date"]),
+	// by_createdBy_date intentionally omitted: createdBy filtering is handled
+	// as a universal post-take in-memory filter (mirrors tasks.ts:371-373 pattern).
+	// Adding an index pushdown optimization is deferred to a separate PR.
+
+	// ── briefingNotes ──────────────────────────────────────────────────────────
+	briefingNotes: defineTable({
+		title: v.string(),
+		topic: v.string(),
+		participants: v.array(v.string()), // ["pi", "laurent", "tau"]
+		content: v.string(), // Full briefing content
+		decisions: v.optional(v.array(v.string())),
+		linkedMemoryIds: v.optional(v.array(v.id("memories"))),
+		createdBy: creatorValidator,
+		createdAt: v.number(),
+		updatedAt: v.optional(v.number()), // set on first update
+		updatedBy: v.optional(creatorValidator), // orchestrator that last updated
+		// Beta multi-tenant scope. null/undefined = master (internal Alpha).
+		// Set to Clerk org slug (e.g. "acme-hr") for client-scoped rows.
+		orgId: v.optional(v.string()),
+	})
+		.index("by_topic", ["topic"])
+		.index("by_creator", ["createdBy", "createdAt"])
+		.index("by_orgId", ["orgId"])
+		// Day 102 v2.11.0 — CRUD baseline PR-C-bis option B (mission k575kc1r):
+		// Convex native BM25 search on briefing body, with filterFields for the
+		// common narrowing axes (topic, createdBy).
+		.searchIndex("search_content", {
+			searchField: "content",
+			filterFields: ["topic", "createdBy", "orgId"],
+		}),
+
+	// ── briefingNoteParticipants ──────────────────────────────────────────────
+	// Day 165 fix (task k175ga65p654z200ydj7s8qv5s8cnxfc): junction table so
+	// "is `identity` a participant on `noteId`" is an INDEX-RANGE predicate
+	// inside the get/list/search queries, never a post-query handler filter
+	// (R-11) and never an in-memory scan of `briefingNotes.participants`
+	// (Convex indexes can't range-query inside an array field's elements).
+	// One row per (noteId, participant) pair; kept in sync by create/update.
+	briefingNoteParticipants: defineTable({
+		noteId: v.id("briefingNotes"),
+		participant: v.string(),
+	})
+		.index("by_participant_note", ["participant", "noteId"])
+		.index("by_note", ["noteId"]),
+
+	// ── components ──────────────────────────────────────────────────────────
+	// Registry of agents, skills, hooks, plugins — backup + inventory.
+	// Content stores the full file so nothing is lost if filesystem is destroyed.
+	components: defineTable({
+		name: v.string(),
+		type: v.union(
+			v.literal("agent"),
+			v.literal("skill"),
+			v.literal("hook"),
+			v.literal("plugin"),
+		),
+		team: v.optional(v.string()), // e.g. "marketing", "development"
+		content: v.string(), // full file content
+		version: v.optional(v.string()),
+		project: v.optional(v.string()),
+		createdBy: creatorValidator,
+		createdAt: v.number(),
+		updatedAt: v.number(),
+	})
+		.index("by_type", ["type"])
+		.index("by_team", ["team", "type"])
+		.index("by_name_type", ["name", "type"]),
+
+	// ── mandates ──────────────────────────────────────────────────────────────
+	// Cross-orchestrator service requests. One orchestrator requests a service from another.
+	// Budget (token allocation) is agreed upfront; cost is recorded on settle.
+	mandates: defineTable({
+		requestedBy: creatorValidator, // who needs the service
+		fulfilledBy: creatorValidator, // who will provide the service
+		service: v.string(), // description of what is needed
+		budget: v.number(), // token budget allocated
+		status: v.union(
+			v.literal("requested"),
+			v.literal("accepted"),
+			v.literal("in_progress"),
+			v.literal("delivered"),
+			v.literal("settled"),
+		),
+		linkedTaskIds: v.optional(v.array(v.id("tasks"))), // tasks created to fulfill this mandate
+		tokensCost: v.optional(v.number()), // actual tokens consumed
+		createdAt: v.number(),
+		updatedAt: v.number(),
+		completedAt: v.optional(v.number()),
+		// AP2 authorization fields
+		spendingLimits: v.optional(
+			v.object({
+				maxPerTransaction: v.number(),
+				maxPerPeriod: v.number(),
+				periodDays: v.optional(v.number()), // default 30
+			}),
+		),
+		approvedCategories: v.optional(v.array(v.string())), // e.g. ["seo", "content", "development"]
+		mandateDocument: v.optional(v.string()), // signed authorization text or reference
+	})
+		.index("by_requestedBy", ["requestedBy", "status"])
+		.index("by_fulfilledBy", ["fulfilledBy", "status"])
+		.index("by_status", ["status", "createdAt"]),
+
+	// ── issues ────────────────────────────────────────────────────────────────
+	// GitHub issues tracked in VantagePeers. Synced via webhook.
+	issues: defineTable({
+		repo: v.string(), // "myreeldream-ai/MyShortReel-beta"
+		issueNumber: v.number(),
+		title: v.string(),
+		body: v.string(), // truncated to 2000 chars
+		htmlUrl: v.string(),
+		labels: v.array(v.string()),
+		status: v.union(
+			v.literal("open"),
+			v.literal("in_progress"),
+			v.literal("fixed"),
+			v.literal("verified"),
+			v.literal("closed"),
+		),
+		priority: v.union(
+			v.literal("urgent"),
+			v.literal("high"),
+			v.literal("medium"),
+			v.literal("low"),
+		),
+		assignedOrchestrator: v.string(),
+		project: v.string(),
+		fixCommits: v.optional(v.array(v.string())),
+		fixedBy: v.optional(v.string()),
+		fixedAt: v.optional(v.number()),
+		verifiedBy: v.optional(v.string()),
+		verifiedAt: v.optional(v.number()),
+		linkedTaskIds: v.optional(v.array(v.string())),
+		githubCreatedAt: v.number(),
+		githubUpdatedAt: v.number(),
+		// External repo tracking (for Zeta contributions to third-party repos)
+		externalRepo: v.optional(v.string()), // "get-convex/better-auth"
+		externalIssueNumber: v.optional(v.number()),
+		externalIssueUrl: v.optional(v.string()),
+		prUrl: v.optional(v.string()),
+		prStatus: v.optional(
+			v.union(
+				v.literal("draft"),
+				v.literal("open"),
+				v.literal("merged"),
+				v.literal("closed"),
+			),
+		),
+		forkRepo: v.optional(v.string()), // "elpiarthera/better-auth"
+	})
+		.index("by_repo_number", ["repo", "issueNumber"])
+		.index("by_status", ["status"])
+		.index("by_project", ["project"])
+		.index("by_assigned", ["assignedOrchestrator", "status"])
+		.index("by_external_repo", ["externalRepo", "prStatus"]),
+
+	// ── githubRepoMapping ─────────────────────────────────────────────────────
+	// Maps GitHub repos to orchestrators. Used by webhook handler to route events.
+	//
+	// Day 98 (k173yr5n1) — auto-IRP cascade overhaul Mechanism (a):
+	//   lastDeployedSHA + lastDeployedAt track the most recent successful Convex
+	//   deploy of the repo. createDeployTaskWithDedup compares pr.mergedAt vs
+	//   lastDeployedAt — if a deploy completed AFTER the PR merged, the PR is
+	//   already shipped via a bundled chain and no per-PR Deploy task is spawned.
+	//   Recorded via githubRepoMapping.recordDeployment mutation after each
+	//   successful `npx convex deploy --yes`.
+	githubRepoMapping: defineTable({
+		repo: v.string(), // "myreeldream-ai/MyShortReel-beta"
+		orchestrator: v.string(), // "omega", "tau", "sigma", etc.
+		project: v.string(), // "myreeldream", "vantage-starter", etc.
+		active: v.boolean(),
+		lastDeployedSHA: v.optional(v.string()), // Day 98 — most recent prod-deployed commit OID
+		lastDeployedAt: v.optional(v.number()), // Day 98 — Unix ms timestamp of that deploy
+	}).index("by_repo", ["repo"]),
+
+	// ── businessUnits ─────────────────────────────────────────────────────────
+	// One row per ElPi Corp business unit. Tracks strategy, structure, and KPIs.
+	// managementFee: ElPi Corp takes this percentage of revenue (default 10%).
+	businessUnits: defineTable({
+		name: v.string(), // e.g. "VantagePeers", "VantageRegistry"
+		description: v.string(),
+		purpose: v.string(),
+		domain: v.optional(v.string()), // e.g. "vantagepeers.com"
+		orchestratorId: v.string(), // lead orchestrator — e.g. "sigma"
+		status: v.union(
+			v.literal("idea"),
+			v.literal("building"),
+			v.literal("live"),
+			v.literal("revenue"),
+		),
+		businessModel: v.string(), // how it makes money
+		targetCustomers: v.string(),
+		services: v.array(v.string()), // what it offers
+		pricing: v.string(),
+		revenueProjections: v.object({
+			y1: v.number(),
+			y2: v.number(),
+			y3: v.number(),
+		}),
+		coreTeam: v.object({
+			agents: v.array(v.string()),
+			skills: v.array(v.string()),
+			hooks: v.array(v.string()),
+			plugins: v.array(v.string()),
+		}),
+		coreProcesses: v.array(v.string()),
+		dependencies: v.array(v.string()), // which BUs it consumes
+		kpis: v.array(v.string()),
+		managementFee: v.number(), // ElPi Corp % cut (default 10)
+		createdAt: v.number(),
+		updatedAt: v.number(),
+	})
+		.index("by_orchestrator", ["orchestratorId"])
+		.index("by_status", ["status"])
+		.index("by_name", ["name"]),
+
+	// ── recurringTasks ────────────────────────────────────────────────────────
+	// Templates for tasks that auto-create on a schedule (daily standup, weekly scan, etc.)
+	// Convex cron checks every 15 min and creates tasks when nextRunAt <= now.
+	recurringTasks: defineTable({
+		title: v.string(),
+		description: v.optional(v.string()),
+		// Orchestrator or person assigned — open string (issue #132)
+		assignedTo: v.string(),
+		priority: v.union(
+			v.literal("urgent"),
+			v.literal("high"),
+			v.literal("medium"),
+			v.literal("low"),
+		),
+		project: v.optional(v.string()),
+		tags: v.optional(v.array(v.string())),
+		cronExpression: v.string(), // "0 9 * * *" = daily 9am, "0 9 * * 1" = Monday 9am
+		lastCreatedAt: v.optional(v.number()),
+		nextRunAt: v.number(),
+		active: v.boolean(),
+		createdBy: creatorValidator,
+		createdAt: v.number(),
+		updatedAt: v.number(),
+	})
+		.index("by_active", ["active", "nextRunAt"])
+		.index("by_assignee", ["assignedTo"]),
+
+	// ── fixPatterns ──────────────────────────────────────────────────────────────
+	// Knowledge base of fix patterns. Documents bugs, root causes, fix attempts
+	// (including failures), and validated fixes. Agents search this BEFORE fixing
+	// to avoid repeating mistakes. Semantic search via RAG on symptom + rootCause.
+	fixPatterns: defineTable({
+		symptom: v.string(), // What the bug looks like (searchable via RAG)
+		rootCause: v.string(), // Why it happens
+		validatedFix: v.optional(v.string()), // The fix that actually worked
+		files: v.optional(v.array(v.string())), // Files involved
+		tags: v.array(v.string()), // e.g. "react-hydration", "convex-subscription"
+		stack: v.array(v.string()), // e.g. "next.js", "convex", "clerk"
+		sourceProject: v.string(), // Origin project
+		linkedIssueIds: v.optional(v.array(v.string())), // VantagePeers issue IDs
+		createdBy: creatorValidator,
+		severity: severityValidator,
+		createdAt: v.number(),
+		updatedAt: v.number(),
+	})
+		.index("by_project", ["sourceProject"])
+		.index("by_severity", ["severity"])
+		.index("by_creator", ["createdBy"]),
+
+	// ── fixAttempts ──────────────────────────────────────────────────────────────
+	// Individual fix attempts for a pattern. Separate table to avoid unbounded
+	// array growth in fixPatterns documents (per Convex guidelines).
+	fixAttempts: defineTable({
+		patternId: v.id("fixPatterns"),
+		description: v.string(), // What was tried
+		commit: v.optional(v.string()), // Commit hash
+		worked: v.boolean(), // Did it fix the issue?
+		why: v.string(), // Why it worked or didn't
+		createdBy: creatorValidator,
+		createdAt: v.number(),
+	})
+		.index("by_pattern", ["patternId"])
+		.index("by_worked", ["patternId", "worked"]),
+
+	// ── missionTemplates ──────────────────────────────────────────────────────
+	// Reusable mission templates. Each template contains an ordered list of
+	// steps that get instantiated as tasks when a triggering event occurs.
+	// The "issue-resolution-v2" template drives the IRP (Issue Resolution Protocol).
+	missionTemplates: defineTable({
+		name: v.string(),
+		description: v.optional(v.string()),
+		brief: v.optional(v.string()), // Reused as the mission's brief on instantiation when the caller supplies none.
+		steps: v.array(
+			v.object({
+				title: v.string(),
+				description: v.string(),
+				tags: v.optional(v.array(v.string())),
+				assignedTo: v.optional(v.string()), // orchestrator role, e.g. "proxima"
+				assignedToInstance: v.optional(v.string()), // instance, e.g. "proxima-vps"
+				dependsOn: v.optional(v.array(v.number())), // indexes of required steps (0-based)
+			}),
+		),
+		isDefault: v.boolean(),
+		createdBy: creatorValidator,
+		createdAt: v.number(),
+		updatedAt: v.number(),
+		// Soft-delete marker — mirrors memories.softDeleteMemory's audit-preserving
+		// PATCH-a-flag motif. Chose a timestamp (not a boolean like memories'
+		// `isLatest`) because it also records *when* the template was retired,
+		// which the memories table gets for free via `updatedAt` alongside the
+		// flag — a lone boolean here would lose that provenance. undefined = active.
+		deletedAt: v.optional(v.number()),
+	})
+		.index("by_name", ["name"])
+		.index("by_default", ["isDefault"]),
+
+	// ── monitoredDeployments ─────────────────────────────────────────────────
+	// Registry of Convex deployments to poll for errors.
+	// deployKeyEnvVar: name of the env var holding the admin deploy key.
+	monitoredDeployments: defineTable({
+		name: v.string(),
+		deploymentUrl: v.string(),
+		deployKeyEnvVar: v.string(),
+		githubRepo: v.string(),
+		orchestrator: v.string(),
+		active: v.boolean(),
+		lastCursor: v.optional(v.number()),
+		createdAt: v.number(),
+	})
+		.index("by_active", ["active"])
+		.index("by_name", ["name"]),
+
+	// ── issueStats ───────────────────────────────────────────────────────────
+	// Daily issue resolution metrics per repo.
+	issueStats: defineTable({
+		repo: v.string(),
+		date: v.string(), // YYYY-MM-DD
+		totalIssues: v.number(),
+		resolvedIssues: v.number(),
+		medianTimeToFirstResponse: v.optional(v.number()), // minutes
+		medianTimeToFix: v.optional(v.number()), // minutes
+		fastestResolution: v.optional(v.number()), // minutes
+		slowestResolution: v.optional(v.number()), // minutes
+		avgTimeToFix: v.optional(v.number()), // minutes
+		// Before/after VantageOS Team (pivot: 2026-04-01)
+		beforeVantageOS: v.optional(
+			v.object({
+				totalIssues: v.number(),
+				resolvedIssues: v.number(),
+				medianTimeToFix: v.optional(v.number()),
+				avgTimeToFix: v.optional(v.number()),
+			}),
+		),
+		afterVantageOS: v.optional(
+			v.object({
+				totalIssues: v.number(),
+				resolvedIssues: v.number(),
+				medianTimeToFix: v.optional(v.number()),
+				avgTimeToFix: v.optional(v.number()),
+			}),
+		),
+		issueDetails: v.optional(
+			v.array(
+				v.object({
+					number: v.number(),
+					title: v.string(),
+					timeToFirstResponse: v.optional(v.number()),
+					timeToFix: v.optional(v.number()),
+					status: v.string(),
+				}),
+			),
+		),
+		calculatedAt: v.number(),
+	})
+		.index("by_repo_date", ["repo", "date"])
+		.index("by_date", ["date"]),
+
+	// ── kbUploads ─────────────────────────────────────────────────────────────
+	// TOFU (trust-on-first-use) storage ownership binding table.
+	// Created by B5 M1 storageId org-binding defense-in-depth (mission k5779q).
+	//
+	// On first call to storeDocumentChunked for a given storageId, an entry is
+	// inserted binding that storageId to the calling org. Subsequent calls with
+	// the SAME storageId but a DIFFERENT orgId are rejected with
+	// AUTH_STORAGE_NOT_OWNED to close the cross-tenant attack vector identified
+	// by Eta iter-2 post-merge follow-up (PR #992, commit 16bb32d).
+	//
+	// Index: by_storageId — O(1) ownership lookup before ctx.storage.get().
+	kbUploads: defineTable({
+		storageId: v.id("_storage"),
+		orgId: v.string(),
+		createdAt: v.number(),
+	}).index("by_storageId", ["storageId"]),
+
+	// ── oauth_clients ─────────────────────────────────────────────────────────
+	// OAuth 2.0 Dynamic Client Registration (RFC 7591).
+	// One row per registered OAuth client (Claude.ai custom connector, Nadia,
+	// future Perello VIP clients). Each client is bound to a scopeProfile that
+	// limits the `from=` allowlist and namespace read/write prefixes.
+	//
+	// clientSecretHash = SHA-256 hex of the raw client secret (raw never stored).
+	// Admin onboarding returns the raw secret exactly ONCE to the provisioner.
+	oauth_clients: defineTable({
+		clientId: v.string(), // public identifier (UUID)
+		clientSecretHash: v.string(), // SHA-256 hex of raw secret
+		redirectUris: v.array(v.string()),
+		name: v.string(), // human label, e.g. "nadia-acme-hr"
+		scopeProfile: v.string(), // FK to oauth_scope_profiles.profileId
+		createdAt: v.number(),
+		revokedAt: v.optional(v.number()),
+		// RFC 7591 §2 token_endpoint_auth_method (D6 — confidential client validation).
+		// Absent => treat as confidential ("client_secret_basic") for backward
+		// compatibility; existing rows MUST be backfilled in S2 with explicit value
+		// ("client_secret_basic" for confidential; "none" for known public clients).
+		tokenEndpointAuthMethod: v.optional(v.string()),
+	})
+		.index("by_clientId", ["clientId"])
+		.index("by_scopeProfile", ["scopeProfile"]),
+
+	// ── oauth_authorization_codes ────────────────────────────────────────────
+	// Short-lived authorization codes (RFC 6749 §4.1). TTL ~10 min.
+	// Cleaned periodically by cron (processExpiredOauthCodes).
+	oauth_authorization_codes: defineTable({
+		code: v.string(), // opaque random value
+		clientId: v.string(),
+		redirectUri: v.string(),
+		codeChallenge: v.string(), // PKCE S256 (RFC 7636)
+		scope: v.string(),
+		userId: v.string(), // who the code was issued for (== scopeProfile owner by default)
+		expiresAt: v.number(),
+	}).index("by_code", ["code"]),
+
+	// ── oauth_access_tokens ──────────────────────────────────────────────────
+	// Issued access tokens. tokenHash = SHA-256 hex of the raw token (raw never
+	// stored). The HTTP bearer middleware hashes the incoming Authorization
+	// header and looks the token up by hash, then enforces fromAllowList +
+	// namespaceRead/Write prefixes on every MCP tool call.
+	oauth_access_tokens: defineTable({
+		tokenHash: v.string(),
+		clientId: v.string(),
+		userId: v.string(),
+		scopes: v.array(v.string()), // OAuth scopes (vantage:read, vantage:write)
+		scopeProfile: v.string(), // snapshot at issue time
+		fromAllowList: v.array(v.string()),
+		namespaceReadPrefixes: v.array(v.string()),
+		namespaceWritePrefixes: v.array(v.string()),
+		expiresAt: v.number(),
+		refreshTokenHash: v.optional(v.string()),
+		createdAt: v.number(),
+		revokedAt: v.optional(v.number()),
+	})
+		.index("by_tokenHash", ["tokenHash"])
+		.index("by_clientId", ["clientId"]),
+
+	// ── oauth_refresh_tokens ─────────────────────────────────────────────────
+	// Refresh tokens for renewing expired access tokens. Same hashing rules.
+	oauth_refresh_tokens: defineTable({
+		tokenHash: v.string(),
+		clientId: v.string(),
+		userId: v.string(),
+		scopeProfile: v.string(),
+		expiresAt: v.number(),
+		createdAt: v.number(),
+		revokedAt: v.optional(v.number()),
+	})
+		.index("by_tokenHash", ["tokenHash"])
+		.index("by_clientId", ["clientId"]),
+
+	// ── oauth_scope_profiles ─────────────────────────────────────────────────
+	// Reusable scope templates. An OAuth client references a profile by
+	// profileId; the access_token materialises the profile into the token row
+	// so scope changes on the profile don't retroactively grant or revoke.
+	//
+	// Seeded on first deploy via oauth.seedDefaultProfiles: master, nadia-acme-hr,
+	// client-generic (deny-by-default template).
+	oauth_scope_profiles: defineTable({
+		profileId: v.string(), // stable slug e.g. "nadia-acme-hr"
+		description: v.string(),
+		fromAllowList: v.array(v.string()),
+		namespaceReadPrefixes: v.array(v.string()),
+		namespaceWritePrefixes: v.array(v.string()),
+		createdAt: v.number(),
+		updatedAt: v.number(),
+	}).index("by_profileId", ["profileId"]),
+
+	// ── mcpTenants ────────────────────────────────────────────────────────────
+	// Registry of VIP tenants for HTTP MCP transport.
+	// Each tenant has a hashed bearer token and a target Convex deployment URL.
+	// The HTTP MCP server (Railway) looks up tenants by tokenHash on every request
+	// and proxies to the correct Convex deployment.
+	// tokenHash = SHA-256 hex of the raw bearer token (raw token never stored).
+	mcpTenants: defineTable({
+		tokenHash: v.string(), // SHA-256 hex of bearer token
+		tenantName: v.string(), // e.g. "perello-consulting-vip-1"
+		convexUrl: v.string(), // e.g. "https://xxxx.convex.cloud"
+		createdAt: v.number(),
+		enabledAt: v.optional(v.number()), // undefined = disabled
+		lastUsedAt: v.optional(v.number()),
+		revokedAt: v.optional(v.number()),
+	}).index("by_tokenHash", ["tokenHash"]),
+
+	// ── errorMonitorFilterRules ──────────────────────────────────────────────
+	// Runtime-configurable filter rules for the auto-IRP bot.
+	// Each rule matches (functionName, errorMessageRegex) and assigns a severity:
+	//   "skip"          → drop silently, no upsert / issue
+	//   "log-only"      → console.log only
+	//   "create-issue"  → normal flow (used to override a default skip)
+	//
+	// Defaults are seeded by errorMonitorFilters.seedDefaultRules. Sigma/Pi
+	// can add or disable rules without redeploying via addFilterRule /
+	// disableFilterRule — internalMutation (admin-only, invoked with a deploy
+	// key via `npx convex run`; converted from public to close a DoS-adjacent
+	// suppression surface, see errorMonitorFilters.ts).
+	//
+	// Linked: memory j573cwcs3znp0xsvtg34x435jh84b0eg, pattern m978zeg4b2e9nx67z2hg5rwgfs85hf7f.
+	errorMonitorFilterRules: defineTable({
+		functionName: v.string(), // exact match, e.g. "tasks:complete"
+		errorMessageRegex: v.string(), // RegExp source
+		regexFlags: v.optional(v.string()), // e.g. "i"
+		reason: v.string(), // human-readable why this is filtered
+		severity: v.union(
+			v.literal("skip"),
+			v.literal("log-only"),
+			v.literal("create-issue"),
+		),
+		active: v.boolean(),
+		createdAt: v.number(),
+		// v1.0.1 observability — added in PR follow-up to PR #354 review.
+		// `lastMatchedAt` and `matchCount` are bumped from `pollDeploymentLogs`
+		// via a fire-and-forget mutation when a rule's severity is "skip" or
+		// "log-only" (the silenced classes), so operators can see which rules
+		// are actually firing in prod and which are dead weight.
+		lastMatchedAt: v.optional(v.number()), // Unix ms of last match
+		matchCount: v.optional(v.number()), // running counter, treat undefined as 0
+		// v1.0.1 precedence — higher `priority` = evaluated first. Stable sort
+		// by `_creationTime` for ties. Treat undefined as 0 (unprioritized).
+		priority: v.optional(v.number()),
+	})
+		.index("by_active", ["active"])
+		.index("by_function", ["functionName", "active"]),
+
+	// ── oauthClients ─────────────────────────────────────────────────────────
+	// OAuth 2.1 Dynamic Client Registration (RFC 7591) clients.
+	// Issued by the DCR /register endpoint without admin gating.
+	// clientSecret = 64-char hex (raw value — returned once on registration,
+	// stored here for PKCE / client_secret_post auth on the token endpoint).
+	oauthClients: defineTable({
+		clientId: v.string(), // crypto.randomUUID()
+		clientSecret: v.string(), // 64-char hex (raw — transmitted once)
+		clientName: v.string(),
+		redirectUris: v.array(v.string()),
+		createdAt: v.number(),
+		scope: v.optional(v.string()), // default "mcp:full"
+	}).index("by_clientId", ["clientId"]),
+
+	// ── oauthTokens ──────────────────────────────────────────────────────────
+	// Auth codes and access/refresh tokens for the DCR OAuth 2.1 flow.
+	// A single row covers both the auth-code phase (authCode set, accessToken
+	// absent) and the token phase (accessToken set, authCode consumed).
+	// Index on authCode supports code exchange; index on accessToken supports
+	// bearer validation; index on clientId supports revocation/listing.
+	oauthTokens: defineTable({
+		clientId: v.string(),
+		accessToken: v.string(), // crypto.randomUUID()
+		refreshToken: v.optional(v.string()),
+		scope: v.string(),
+		expiresAt: v.number(), // ms since epoch
+		authCode: v.optional(v.string()), // PKCE authorization code
+		codeChallenge: v.optional(v.string()),
+		codeChallengeMethod: v.optional(v.string()),
+		redirectUri: v.optional(v.string()),
+		used: v.optional(v.boolean()), // auth code single-use flag
+		createdAt: v.number(),
+	})
+		.index("by_accessToken", ["accessToken"])
+		.index("by_authCode", ["authCode"])
+		.index("by_clientId", ["clientId"])
+		.index("by_refreshToken", ["refreshToken"]),
+
+	// ── errorLogs ────────────────────────────────────────────────────────────
+	// Deduplicated log of detected function errors across monitored deployments.
+	// hash = simpleHash(functionName + ":" + errorMessage) for deduplication.
+	//
+	// Anti-flood additions (Day 76):
+	//   issueCreated   — true once the GH issue + IRP mission have been created.
+	//                    Prevents re-triggering if the errorLog is touched again.
+	//   irpMissionId   — ID of the auto-IRP mission spawned for this error, so
+	//                    the auto-resolver can cascade-close it.
+	//   autoResolved   — true once the auto-resolver has closed the mission.
+	//                    Prevents the resolver from double-processing.
+	//   recurrenceThreshold — overrides the deployment-level default; the GH issue
+	//                    is NOT created until count >= this value.
+	errorLogs: defineTable({
+		hash: v.string(),
+		deployment: v.string(),
+		functionName: v.string(),
+		errorMessage: v.string(),
+		stackTrace: v.optional(v.string()),
+		firstSeen: v.number(),
+		lastSeen: v.number(),
+		count: v.number(),
+		issueNumber: v.optional(v.number()),
+		githubRepo: v.optional(v.string()),
+		// Day 76 anti-flood fields
+		issueCreated: v.optional(v.boolean()),
+		irpMissionId: v.optional(v.id("missions")),
+		autoResolved: v.optional(v.boolean()),
+		recurrenceThreshold: v.optional(v.number()),
+		// Day 128 fix (issue #1088 fabricated incident) — see
+		// convex/errorMonitorRecurrence.ts. Records the `count` value at the
+		// moment a 24h+ re-raise measurement window was armed. The RECURRING
+		// escalation label is only emitted once NEW occurrences counted after
+		// this baseline reach the effective threshold — never from group
+		// identity + a stale timestamp alone.
+		reRaiseBaselineCount: v.optional(v.number()),
+	})
+		.index("by_hash", ["hash"])
+		.index("by_deployment", ["deployment"])
+		.index("by_issue_created", ["issueCreated", "lastSeen"])
+		.index("by_issue_number", ["issueNumber"]),
+
+	// ── licenses ─────────────────────────────────────────────────────────────
+	// Open-core license registry. Raw license keys are NEVER stored — only the
+	// SHA-256 hex hash is persisted. The raw key is returned once on generate and
+	// is the customer's responsibility to store safely.
+	//
+	// Lifecycle: generated → activated (optional, sets activatedAt) → active until
+	// expiresAt, after which status moves to "expired" (handled by cron or on
+	// validate). Revocation sets status="revoked" immediately.
+	licenses: defineTable({
+		keyHash: v.string(), // sha256 hex of the raw license key
+		customerEmail: v.string(),
+		customerName: v.optional(v.string()),
+		productCode: v.string(), // e.g. "vantage-peers-self-host"
+		tier: v.string(), // e.g. "open-core-99-eur-yr"
+		purchasedAt: v.number(), // ms since epoch
+		activatedAt: v.optional(v.number()), // ms since epoch, set on first activation
+		expiresAt: v.number(), // purchasedAt + 365d (ms since epoch)
+		gumroadOrderId: v.optional(v.string()),
+		status: v.union(
+			v.literal("active"),
+			v.literal("trial"),
+			v.literal("revoked"),
+			v.literal("expired"),
+		),
+		githubRepos: v.optional(v.array(v.string())),
+		purchaseLocale: v.optional(v.union(v.literal("en"), v.literal("fr"))),
+		// Onboarding email delivery tracking (set by gumroadWebhook handler)
+		emailSent: v.optional(v.boolean()),
+		// Customer segmentation — set at purchase time or enriched later.
+		// Unset = unknown / backwards-compat for rows created before W3.
+		customerType: v.optional(
+			v.union(
+				v.literal("early-dev-voie-3"),
+				v.literal("smb-voie-2"),
+				v.literal("pro-voie-3"),
+				v.literal("partner-reseller"),
+				v.literal("standard"),
+			),
+		),
+		// True if the customer has expressed interest in reselling VantagePeers.
+		resellerCandidate: v.optional(v.boolean()),
+	})
+		.index("by_keyHash", ["keyHash"])
+		.index("by_customerEmail", ["customerEmail"])
+		.index("by_gumroadOrderId", ["gumroadOrderId"]),
+
+	// ── iframeEmbedSessions ──────────────────────────────────────────────────
+	// Session registry for VP Gen UI iframe embeds (SEP-1865 M3).
+	// Each session is bound to an origin and carries optional tenantId + userId
+	// for multi-tenant routing. Sessions expire via expiresAt; revoked flag
+	// provides immediate invalidation without waiting for TTL.
+	//
+	// Mission : sigma-vantage-peers-mcp-gui-iframe-embed-v1 (k5730xct6rvrwkvxhy5t5js12d87jwfw).
+	iframeEmbedSessions: defineTable({
+		sessionId: v.string(),
+		tenantId: v.optional(v.string()),
+		origin: v.string(),
+		userId: v.optional(v.string()),
+		createdAt: v.number(),
+		lastSeenAt: v.number(),
+		expiresAt: v.number(),
+		revoked: v.boolean(),
+	})
+		.index("by_session_id", ["sessionId"])
+		.index("by_origin_expires", ["origin", "expiresAt"]),
+
+	// ── errorMonitorConfig ───────────────────────────────────────────────────
+	// Singleton-ish dynamic configuration for the error-monitor subsystem.
+	// Key-value store where `key` is a stable slug (e.g. "pendingAliasReleases")
+	// and `value` is a string array.
+	//
+	// pendingAliasReleases — list of status aliases introduced in an in-flight
+	// release that are not yet deployed to prod. The auto-IRP bot synthesises
+	// ArgumentValidationError filter rules for each alias so that pre-deploy
+	// smoke-test noise doesn't spawn GitHub issues. Remove aliases post-deploy.
+	errorMonitorConfig: defineTable({
+		key: v.string(), // e.g. "pendingAliasReleases"
+		value: v.array(v.string()),
+		updatedAt: v.number(),
+	}).index("by_key", ["key"]),
+
+	// ── taskClosureConfig ────────────────────────────────────────────────────
+	// Day 130 (k17dhcmzqafve1ayzvh833kf558ae019) — server-side closure gate
+	// config. Key-value store, mirrors errorMonitorConfig pattern.
+	// Doctrine: no-hardcoded-business-knowledge — billable project list and
+	// stale-in-progress threshold live here, never as a code constant.
+	//
+	// Known keys:
+	//   "billableProjects"       — string[] of `task.project` values that
+	//     require a machine-timestamped startedAt before closure (billing
+	//     source = actualMinutes derived from startedAt→completedAt).
+	//     Seeded with ["vantage-immo"] — add client projects here, not in code.
+	//   "staleInProgressThresholdMs" — number encoded as a single-element
+	//     string[] (["86400000"]) — age (ms) after which an in_progress task
+	//     is surfaced as staleInProgress in check_messages. Default 24h.
+	taskClosureConfig: defineTable({
+		key: v.string(),
+		value: v.array(v.string()),
+		updatedAt: v.number(),
+	}).index("by_key", ["key"]),
+
+	// ── client_org_mapping ───────────────────────────────────────────────────
+	// Dashboard Beta multi-tenant scope registry. One row per Clerk organisation
+	// granted dashboard access. Provisioned manually by Pi / Laurent via Convex
+	// dashboard or npx convex run after merge.
+	//
+	// clerkOrgSlug: the Clerk organisation slug (e.g. "acme-hr", "a-client-org").
+	//   Maps to `identity.organizationId ?? identity.organizationSlug` in Convex
+	//   auth helpers.
+	// allowedOrchestrators: which orchestrator pilots/assignees this org can see.
+	//   ["*"] = master sentinel = all orchestrators (Laurent / internal use).
+	// scopes: permission tokens granted to this org.
+	//   Known values: "view-own-tasks", "view-own-missions",
+	//   "view-orchestrator-summary", "view-stats-aggregated", "cross-tenant-read".
+	// isActive: false = org is disabled (returns Forbidden) without deleting the row.
+	//
+	// Seed rows (post-merge, Sigma runs npx convex run):
+	//   acme-hr   → allowedOrchestrators=["victor"], scopes=["view-own-tasks","view-own-missions","view-orchestrator-summary"]
+	//   <redacted-client> → allowedOrchestrators=["phi"],    scopes=["view-own-tasks","view-own-missions"]
+	client_org_mapping: defineTable({
+		clerkOrgSlug: v.string(), // "acme-hr"
+		allowedOrchestrators: v.array(v.string()), // ["victor"] or ["*"] for master sentinel
+		scopes: v.array(v.string()), // ["view-own-tasks", "view-own-missions", ...]
+		displayName: v.string(), // "<redacted-client>"
+		isActive: v.boolean(),
+		createdAt: v.number(),
+	})
+		.index("by_clerk_slug", ["clerkOrgSlug"])
+		.index("by_isActive", ["isActive"]),
+
+	// ── userBearerTokens ─────────────────────────────────────────────────────
+	// Bearer tokens issued to VP webapp users via Clerk JWT exchange
+	// (POST /issueBearerFromClerk). Separate from the OAuth DCR flow in
+	// oauth_access_tokens — these are user-grade tokens bound to a Clerk userId
+	// rather than an OAuth client.
+	//
+	// tokenHash = SHA-256 hex of the raw token (raw value NEVER stored).
+	// The raw bearer is returned once in the HTTP response payload and never
+	// re-derivable from the database.
+	//
+	// TTL: 7 days (issuedAt + 604800000 ms).
+	userBearerTokens: defineTable({
+		tokenHash: v.string(), // SHA-256 hex of raw bearer
+		clerkUserId: v.string(), // Clerk `sub` claim
+		workspaceId: v.string(), // logical workspace identifier (clerkUserId-based slug)
+		extId: v.string(), // Chrome extension ID that requested the token
+		expiresAt: v.number(), // ms since epoch
+		revoked: v.boolean(),
+		createdAt: v.number(),
+		lastUsedAt: v.optional(v.number()),
+	})
+		.index("by_token_hash", ["tokenHash"])
+		.index("by_clerkUserId", ["clerkUserId"]),
+
+	// ── credentialsAuditLog ──────────────────────────────────────────────────
+	// Append-only audit trail for every Bearer issuance via issueBearerFromClerk.
+	// Never updated after insert — one row per successful (or failed) issuance.
+	credentialsAuditLog: defineTable({
+		clerkUserId: v.string(),
+		workspaceId: v.string(),
+		extId: v.string(),
+		extVersion: v.optional(v.string()),
+		issuedAt: v.number(),
+		ip: v.optional(v.string()),
+		userAgent: v.optional(v.string()),
+	})
+		.index("by_clerkUserId", ["clerkUserId"])
+		.index("by_workspaceId", ["workspaceId"]),
+
+	// ── credentialsRateLimits ────────────────────────────────────────────────
+	// Rolling rate-limit counter for issueBearerFromClerk.
+	// key = "<clerkUserId>-issueBearer"
+	// Resets automatically when windowStart + 60000 < now.
+	credentialsRateLimits: defineTable({
+		key: v.string(), // "<clerkUserId>-issueBearer"
+		count: v.number(), // requests in current window
+		windowStart: v.number(), // ms since epoch — start of 1-minute window
+	}).index("by_key", ["key"]),
+
+	// ── oauth_audit_log ──────────────────────────────────────────────────────
+	// Append-only audit trail for administrative scope profile mutations.
+	// One row per patchScopeProfileEmergency invocation.
+	// actorTokenHash = sha256Hex of the callerToken (raw token never stored).
+	// previousState + newState allow forensic reconstruction of leaked scopes.
+	// S1.2-mutation: captures Day 90 Nadia `global` leak remediation.
+	// S2.1-D9: clientsRetargeted added (additive, optional for backward compat).
+	oauth_audit_log: defineTable({
+		eventType: v.string(),
+		actorTokenHash: v.string(),
+		targetProfileId: v.string(),
+		previousState: v.object({
+			profileId: v.string(),
+			fromAllowList: v.array(v.string()),
+			namespaceReadPrefixes: v.array(v.string()),
+			namespaceWritePrefixes: v.array(v.string()),
+		}),
+		newState: v.object({
+			profileId: v.string(),
+			fromAllowList: v.array(v.string()),
+			namespaceReadPrefixes: v.array(v.string()),
+			namespaceWritePrefixes: v.array(v.string()),
+		}),
+		reason: v.string(),
+		cascadeRevokedCount: v.number(),
+		// S2.1-D9: number of oauth_clients rows retargeted during rename (0 if no rename).
+		// Optional for backward compat with pre-S2.1 rows.
+		clientsRetargeted: v.optional(v.number()),
+		createdAt: v.number(),
+	})
+		.index("by_targetProfileId", ["targetProfileId"])
+		.index("by_createdAt", ["createdAt"]),
+
+	// ── okfDurableExportProgress ─────────────────────────────────────────────
+	// I1 long-task survival — durable variant of `exportOkfBundle`
+	// (convex/okfBundleDurable.ts). One row per durable job (keyed by the
+	// `@vantageos/agent-engine` `jobId`), tracking pagination cursors across
+	// steps so a step function can resume exactly where the previous one left
+	// off. `orgId` here IS the requested namespace string (see
+	// okfBundleDurable.ts comment) — kept as its own field, never trusted as a
+	// client-supplied scope on any step: every step re-derives its DB reads
+	// from `namespace`/`orgId` on THIS row, written once at job creation by
+	// the authenticated `startOkfBundleExportDurable` mutation.
+	okfDurableExportProgress: defineTable({
+		jobId: v.string(), // opaque id returned by durableJob.start
+		orgId: v.string(), // == namespace; tenant scope for every step read
+		namespace: v.string(),
+		sinceMs: v.optional(v.number()),
+		memoriesCursor: v.union(v.string(), v.null()),
+		memoriesDone: v.boolean(),
+		briefingsCursor: v.union(v.string(), v.null()),
+		briefingsDone: v.boolean(),
+		tasksCursor: v.union(v.string(), v.null()),
+		tasksDone: v.boolean(),
+		memoryCount: v.number(),
+		briefingCount: v.number(),
+		taskCount: v.number(),
+		stepsCompleted: v.number(),
+		status: v.union(v.literal("running"), v.literal("assembled")),
+		createdAt: v.number(),
+		updatedAt: v.number(),
+	})
+		.index("by_jobId", ["jobId"])
+		.index("by_orgId_jobId", ["orgId", "jobId"]),
+
+	// ── okfDurableExportEntries ──────────────────────────────────────────────
+	// Serialized bundle files (path + markdown content) appended one page at a
+	// time by the durable step function. `orgId` is the FIRST index field on
+	// every index (multi-tenant convention) — a step for org A can never read
+	// org B's rows because every query below is scoped by `.eq("orgId", ...)`
+	// before anything else.
+	okfDurableExportEntries: defineTable({
+		orgId: v.string(),
+		jobId: v.string(),
+		family: v.union(
+			v.literal("memory"),
+			v.literal("briefing"),
+			v.literal("task"),
+		),
+		seq: v.number(),
+		path: v.string(),
+		content: v.string(),
+	})
+		.index("by_org_job", ["orgId", "jobId"])
+		.index("by_org_job_seq", ["orgId", "jobId", "seq"]),
+});