FIN Projection Catalog

Date: 2026-05-15 Status: Draft v0.1 (skeleton — sections 1-5 seeded; sections 6-9 deferred to v0.2) Related design: docs/design/FIN_INTRANET_HUMAN_AGENT_PROJECTION_LAYER_DESIGN_2026-05-13.md Related spec: docs/engineering/FIN_INTRANET_HUMAN_AGENT_PROJECTION_ENGINEERING_SPEC_2026-05-13.md (v0.2) Related audit: docs/audits/FIN_INTRA_GTM_PLANNING_CODEBASE_GAP_AUDIT_2026-05-13.md Related strategy: docs/strategy/FIN_DATA_PLANE_AUDIT_2026-04-30.md

0. Purpose

This catalog enumerates every markdown projection FIN produces. Each projection is a deterministic transform from canonical sources (events + knowledge rows + storage bytes) into a human-readable, agent-readable, citable artifact. The catalog defines the entire surface area of FIN’s system of intelligence.

The catalog exists to:

• prevent ad-hoc projection sprawl
• give engineering, agents, customers, and regulators a single index of what FIN generates
• bind every projection to a wave, a renderer version, an audience filter, and a citation policy
• form the contribution boundary for the public Harvest Directory

Anything not in this catalog should not be projected. New projections require a catalog entry, a Wave assignment, and a renderer-version registration before implementation.

1. Principles

1. One entity, one current profile. Each Taproot entity row has exactly one current entity_profile doc plus optional immutable weekly/monthly snapshots.
2. Immutable snapshots. Snapshots are written once, never updated. They are the citation targets for federal and historical claims.
3. Renderer version recorded. Every projection row carries renderer_version. Bumps require an explicit backfill plan.
4. Citations on every claim. No projected fact ships without fact_citations[] pointing to events, knowledge rows, or external authoritative sources.
5. Anonymization is a policy artifact. Public-anonymized projections reference a versioned, reviewable policy. The policy is auditable and diff-able.
6. Determinism is required for legal-grade projections. recall_trace, proof_package, and federal-audience projections MUST produce identical output for identical inputs.
7. No projection without a wave. Every entry in §5 is tagged Wave 1 / 2 / 3 / 4 / Deferred. Implementation cannot precede catalog tagging.
8. No new doc_kinds via inline invention. New doc_kind values require a spec amendment and migrate through §4 before §5.
9. JSONB is the storage; markdown is the rendering. Each projection is a knowledge.domain = doc row with JSONB content. The markdown body lives inside that JSONB. Bytes (charts, attachments) live behind StorageProvider.
10. Public mirrors are derived, never authored. Harvest Directory entries are projected from tenant data through the anonymization gate. They are not separately authored, otherwise they drift.

2. Source Planes

Projections are categorized by which of the three data planes they read from and which they write to. Reading from a higher-trust plane and writing to a lower-trust plane requires an anonymization policy.

2.1 Plane 1 — Private Workspace

Source of: personal notes, role-scoped task lists, actor-memory digests, authored decision records scoped to one actor.

Reads from: actors.memory, authored markdown in GitHub, private events.

Writes to: knowledge.domain = doc with visibility = personal.

Audiences allowed: the actor, the actor’s privileged role-assigned partners only.

Examples: actor_daily_notes, agent_session_recap, personal_task_board.

2.2 Plane 2 — Relationship Data

Source of: tenant operational truth — orders, lots, arrivals, QC events, shipments, invoices, customer/vendor history, holds, temperature exceptions.

Reads from: events (most signal.* / lot.* / qc.* / shipment.* families), knowledge.domain = order|lot|shipment|customer|vendor, operational_memory extracts of emails and Teams threads.

Writes to: knowledge.domain = doc with visibility = company | customer_visible | restricted and data_plane = relationship_data.

Audiences allowed: tenant employees, FIN operators (for support), customer-visible subset goes to customer portal, vendor-visible subset goes to vendor portal.

Examples: nathel_daily_arrivals, customer_weekly_invoices, vendor_scorecard, recall_trace, proof_package.

2.3 Plane 3 — Universal Intelligence

Source of: commodity knowledge, variety details, regional patterns, seasonal windows, FDA/USDA/FSMA standards, post-harvest physiology, cross-tenant anonymized intelligence.

Reads from: knowledge.domain = commodity|variety|region|certification| season_window, external authoritative sources (USDA, FDA, university extension), anonymized aggregates across tenants.

Writes to: knowledge.domain = doc with visibility = tenant_private for the Taproot view and visibility = public_anonymized for Harvest Directory.

Audiences allowed: tenant agents and humans for tenant-private; public Tier 0 MCP clients, federal reviewers, contributors, search engines for public-anonymized.

Examples: commodity_profile.watermelon, variety_profile.seedless_mini, region_intelligence.sonora-MX, season_window.mexican_watermelon_spring, commodity_situation.watermelon.2026-W19.

3. Audiences

Each projection in §5 declares an audience filter. The filter determines visibility, anonymization, and which fields are projected.

3.1 Operator Agent

Privileged internal agents that perform multi-step ops work for a tenant. Receive full Plane 1 + 2 access for their assigned tenant. Receive Plane 3 full-fidelity (not anonymized) because they operate inside the tenant trust boundary.

3.2 Employee (Tenant)

Tenant operational staff using the PWA / Workspace. Receive Plane 2 docs filtered by their role (receiver vs. dispatcher vs. office manager) and full tenant Taproot.

3.3 Customer (Tenant’s customer)

Receive customer-visible subset of Plane 2 docs (their orders, their invoices, their delivery status, their permitted communications). Receive public Harvest Directory entries. Never see vendor identity, cost basis, margin, other customers, internal communications.

3.4 Vendor (Tenant’s vendor)

Receive vendor-visible subset of Plane 2 docs (their accepted lots, their QC outcomes, payment status). Never see other vendors, customer prices, margin, customer identity beyond the tenant.

3.5 FIN Operator

Anthropic-style internal operations role (you, future hires). Cross-tenant visibility for support and platform purposes, but anonymization required when projecting cross-tenant aggregates back to public.

3.6 Federal / Regulator

Receives recall traces, FSMA proof packages, certification artifacts. Strict schema; no commercial fields; deterministic renderer required.

3.7 Public / Tier 0 MCP

External developers, universities, USDA collaborators, customers without accounts. Receive only public_anonymized projections through the Tier 0 MCP contract. No tenant identity, no individual transactional data.

4. Cadence

Each projection declares one or more cadence triggers.

Cadence	Trigger	Examples
Event-driven (real-time)	Specific event types fire	recall_trace on compliance.recall_initiated; entity_profile rebuild on entity row change
Daily	Scheduled job, tenant-local 06:00	nathel_daily_arrivals, operator_morning_briefing, kb_health_summary
Weekly	Monday 06:00 tenant-local	customer_weekly_invoices, commodity_situation.<x>.<year-week> snapshot freeze
Monthly	1st of month 06:00 tenant-local	vendor_scorecard, monthly commodity_profile snapshot freeze
Quarterly	Quarter-end + 5 days	proof_package.<tenant>.<quarter>, federal readiness rollups
Annual	Calendar year-end + 14 days	season_window annual snapshots, year-in-review proof packages
On-demand (privileged)	MCP tool call by privileged actor	assemble_recall_trace, recall_drill
On-demand (agent-proposed)	Agent emits doc.edit_proposed	enrichment patches reviewed before merge

Scheduled jobs run as cron-triggered Workers. Each emits a projection_job.started and projection_job.completed event for FIN Central observability.

5. Doc-Kind Catalog

This is the master enum of doc kinds. Every projection in §6 (inventory) must use one of these kinds. Adding a kind requires a spec amendment.

doc_kind	Source plane	Cadence range	Visibility states	Determinism required
authored	Any	Event (on commit)	All five states	No
system_projected	Any	Daily / weekly / event	Tenant-scoped	Yes
decision_record	Plane 1 / 2	On commit	company / restricted	No (authored)
runbook	Plane 2 / 3	On commit	company / restricted	No (authored)
sop	Plane 2 / 3	On commit	company / restricted	No (authored)
proof_package	Plane 2	Quarterly / on-demand	company / customer_visible / restricted	Yes
event_digest	Plane 2	Daily / weekly / on-demand	company / customer_visible	Yes
portal_page	Plane 2	Event / daily	customer_visible / vendor_visible	Yes
entity_profile	Plane 3	Event-driven + weekly + monthly snapshots	tenant_private / public_anonymized	Yes
aggregate_intelligence	Plane 3 (cross-tenant)	Weekly / monthly	public_anonymized	Yes
recall_trace	Plane 2	On-demand	restricted / federal-only by audience flag	Yes (mandatory)

6. Projection Inventory (Seeded)

Every entry below carries: name, doc_kind, source plane, audience, cadence, sources (events + knowledge), wave. Wave 1 entries are required for the Nathel proof loop and ship first. Wave 2 starts entity profiles. Wave 3 unlocks recall trace + Harvest Directory. Wave 4 is post-Nathel-funding expansion. Deferred entries are noted explicitly.

6.1 Plane 2 Operational Projections

nathel_daily_arrivals

• doc_kind: event_digest
• audience: operator agent, employee
• cadence: daily 06:00 tenant-local
• sources: signal.pp_arrival_logged, signal.temperature_exception, lot.received, qc.failed, qc.passed
• output shape: list of arrivals, anomalies highlighted, links to lot profiles, embedded sparkline of arrival count vs. trailing 14-day mean
• wave: Wave 1 (this is the spine-proving projection)

operator_morning_briefing

• doc_kind: event_digest
• audience: operator agent
• cadence: daily 05:30 tenant-local
• sources: yesterday’s events across all signal.* / lot.* / qc.* families; pending actions in actor_pulse_aggregate_projection; today’s planned arrivals; weather; holiday/transit flags
• output shape: prioritized brief, suggested first three actions, links to affected entities
• wave: Wave 1

qc_priorities_today

• doc_kind: event_digest
• audience: employee (QC role)
• cadence: daily 06:00, refreshed on new lot.received event
• sources: lot.received with qc_status = pending, signal.temperature_exception, qc.escalation_requested
• output shape: prioritized lot list with reason ranking, prior-week defect rate per commodity, expected shelf-life clock
• wave: Wave 2

lot_status_dashboard

• doc_kind: portal_page
• audience: employee
• cadence: event-driven on any lot.* event
• sources: lot_status_projection (existing migration 041), linked arrivals and shipments
• output shape: dashboard-style doc with live lot table, embedded heatmap of hold age, recent escalations
• wave: Wave 2

customer_weekly_invoices

• doc_kind: portal_page
• audience: customer (customer’s portal)
• cadence: weekly Monday 06:00 customer-local
• sources: invoice.generated, shipment.delivered, payment.received
• output shape: weekly invoice summary, payment status, recent deliveries with photos, link to billing contact
• wave: Wave 2 (one Nathel customer pilot)

customer_situation_brief

• doc_kind: portal_page
• audience: customer
• cadence: weekly Friday 16:00 customer-local
• sources: arrivals fulfilled vs. ordered, commodity availability for the customer’s typical order pattern, public Harvest Directory commodity situation
• output shape: forward-looking advisory; “this is what to expect next week”
• wave: Wave 3

vendor_scorecard

• doc_kind: entity_profile (vendor entity)
• audience: FIN operator, employee
• cadence: monthly snapshot, live updated on vendor.* events
• sources: vendor entity row, lot.received from this vendor, qc.failed / qc.passed filtered by vendor, signal.temperature_exception, payment history
• output shape: vendor reliability score, defect rate trend (line chart), commodity mix (bar chart), notable incidents, payment terms summary
• wave: Wave 2

recall_trace.<lot_id>

• doc_kind: recall_trace
• audience: FIN operator (internal), customer (customer-safe), regulator (federal flag)
• cadence: on-demand via assemble_recall_trace MCP tool
• sources: graph traversal from lot:X → commodity → arrivals → shipments → customers → comms → corrective action events
• output shape: scope statement, affected lots, customer reach, root-cause candidates with citation chain, mitigation timeline, recordkeeping evidence
• wave: Wave 3

recall_drill.fsma-204-test

• doc_kind: runbook
• audience: federal, FIN operator
• cadence: on-demand (quarterly drill recommended)
• sources: synthetic recall scenario events, time-to-trace measurements
• output shape: drill plan, expected step durations, pass/fail criteria, evidence of completion
• wave: Wave 3

proof_package.<tenant>.<quarter>

• doc_kind: proof_package
• audience: FIN operator, customer (executive level, when shared), federal, partner (Anthropic/investor when shared)
• cadence: quarterly + on-demand
• sources: tenant KPIs, event volume by family, agent action counts, outcomes attributed to FIN intervention, customer testimonials linked from comms threads
• output shape: executive summary, KPI deltas with embedded charts, narrative of operational improvements, cited evidence for every claim, attachments ledger
• wave: Wave 3 (Nathel Q2 2026 is the first one)

daily_sales_situation

• doc_kind: event_digest
• audience: operator agent, employee (sales role)
• cadence: daily 06:00 + refreshed on each sales event
• sources: daily sales report extractor output (per project_pp_extractor_order memo), open orders, fill rates, customer rejections
• output shape: sales snapshot, attention list (orders at risk, customers with declining fill rate), suggested outreach
• wave: Wave 2

po_intelligence.<po_id>

• doc_kind: event_digest
• audience: operator agent, employee
• cadence: event-driven on any event carrying this po_id (per project_po_is_the_spine memo)
• sources: every event with the po_id; the 5 intelligence dimensions from the spine memo
• output shape: PO timeline, exceptions, linked lots, linked invoices, expected vs. actual position
• wave: Wave 2

sight_picture.weekly

• doc_kind: event_digest
• audience: operator agent, FIN operator, employee (executive role)
• cadence: weekly Friday 14:00
• sources: PO + sales + inventory projections vs. stock (per project_sight_picture_formula memo); three-layer formula (operational + organizational + public KB)
• output shape: sight-picture dashboard doc, multi-chart, narrative summary, forward look
• wave: Wave 2

6.2 Plane 3 Entity Profiles (Taproot)

commodity_profile.<commodity_id>

• doc_kind: entity_profile (entity_type: commodity)
• audience: operator agent, employee, customer (filtered), FIN operator, public via Harvest Directory mirror
• cadence: event-driven + weekly snapshot + monthly snapshot
• sources: knowledge.domain = commodity row, all signal.* events with this commodity_id, all lot.* events with this commodity_id, public authoritative sources (USDA cold chain, FDA classification)
• output shape: golden-profile frontmatter + narrative body covering cultivation, post-harvest physiology, common defects, handling SOPs, varieties, growers, regions, seasonal supply patterns, embedded charts (supply curve, price band, defect-rate trend), backlinks to all lots and arrivals
• wave: Wave 2 (first: watermelon, on Nathel pilot)

variety_profile.<variety_id>

• doc_kind: entity_profile (entity_type: variety)
• audience: same as commodity_profile
• cadence: same
• sources: knowledge.domain = variety row, lot events filtered by variety, customer preference signals
• output shape: variety detail sheet, parent commodity link, season window link, typical pack types, defect signatures specific to this variety, customer preference data (anonymized for public mirror)
• wave: Wave 2

vendor_profile.<vendor_id>

• doc_kind: entity_profile (entity_type: vendor)
• audience: FIN operator, employee; vendor receives a filtered self-view
• cadence: event-driven + monthly snapshot
• sources: knowledge.domain = vendor row, all events tied to this vendor, comms history extracts
• output shape: vendor profile with commodity mix, reliability history, payment terms, contact graph, notable incidents
• wave: Wave 2

customer_profile.<customer_id>

• doc_kind: entity_profile (entity_type: customer)
• audience: FIN operator, employee (sales role); customer receives a filtered self-view via portal
• cadence: event-driven + weekly snapshot
• sources: knowledge.domain = customer row, orders, invoices, payment history, comms threads
• output shape: relationship profile, order pattern, commodity loyalty, payment timeliness, fill-rate history, suggested outreach moments
• wave: Wave 3

region_intelligence.<region_id>

• doc_kind: entity_profile (entity_type: region)
• audience: FIN operator, public via Harvest Directory
• cadence: weekly snapshot + monthly snapshot
• sources: regional commodity rollups across tenants (anonymized), weather/season correlation, freight pattern data
• output shape: region overview, dominant commodities, season windows, pricing patterns (banded for public mirror), transit corridors, recent notable supply events
• wave: Wave 3

season_window.<window_id>

• doc_kind: entity_profile (entity_type: season_window)
• audience: all (filtered)
• cadence: weekly during window, monthly outside window, annual snapshot
• sources: commodity × region × time aggregates
• output shape: season profile (e.g. “Mexican watermelon spring window”), expected start/peak/end weeks, typical pricing curve, supply ramp pattern, historical anomalies
• wave: Wave 3

grower_profile.<grower_id>

• doc_kind: entity_profile (entity_type: grower)
• audience: FIN operator, employee
• cadence: monthly snapshot
• sources: grower entity row, lot provenance chains, certification status
• output shape: grower detail, certifications, varieties grown, recent shipments, traceability completeness score
• wave: Wave 3

certification_profile.<cert_id>

• doc_kind: entity_profile (entity_type: certification)
• audience: all (filtered)
• cadence: on certification authority update
• sources: external authoritative source (USDA Organic, GAP, GlobalG.A.P.), internal verification events
• output shape: certification overview, what it covers, how to verify, list of growers/vendors holding it (tenant-private detail; public mirror lists only public-registry-published holders)
• wave: Wave 4 (post-funding expansion)

6.3 Plane 3 Aggregate Intelligence (Harvest Directory)

commodity_situation.<commodity_id>.<year-week>

• doc_kind: aggregate_intelligence
• audience: public via Harvest Directory + customer (filtered)
• cadence: weekly snapshot
• sources: cross-tenant commodity events, anonymized; public weather data; freight corridor data
• output shape: weekly situation brief — supply level, pricing band, notable disruptions, forward look 2 weeks
• wave: Wave 3 (first: watermelon, US Northeast market)

commodity_seasonal_outlook.<commodity_id>.<year>

• doc_kind: aggregate_intelligence
• audience: public + customer
• cadence: monthly through season, annual snapshot
• sources: multi-year cross-tenant aggregates, USDA seasonal forecasts
• output shape: outlook brief, season expectations, risk factors, comparable prior years
• wave: Wave 4

region_outlook.<region_id>.<year-week>

• doc_kind: aggregate_intelligence
• audience: public + FIN operator + federal
• cadence: weekly
• sources: cross-tenant regional aggregates, transportation disruption signals, weather
• output shape: region weekly outlook, dominant commodities, transit conditions, regulatory notes
• wave: Wave 4

6.4 Decision Records, Runbooks, SOPs (Authored)

decision_record.<topic>.<date>

• doc_kind: decision_record
• audience: company / restricted
• cadence: authored on commit
• sources: GitHub vault, linked events as evidence
• output shape: ADR-style: context, options, decision, consequences, citations
• wave: Wave 1 (vault ingest must support these from day one)

runbook.<process>

• doc_kind: runbook
• audience: company
• cadence: authored, with last_validated_at event tracking
• sources: GitHub vault + linked SOP events
• output shape: step-by-step operational runbook with linked entities, expected durations, escalation paths
• wave: Wave 1

sop.<role>.<process>

• doc_kind: sop
• audience: employee, role-scoped
• cadence: authored
• sources: GitHub vault
• output shape: standard operating procedure with photos, video links, checklist, sign-off events
• wave: Wave 2

6.5 Agent Harness Projections

agent_harness_session.<session_id>

• doc_kind: event_digest
• audience: FIN operator, employee (the actor that ran the session), customer/vendor (filtered, when their portal session)
• cadence: event-driven (closed on session end) + post-session enrichment
• sources: agent.session.started, agent.tool_called, agent.tool_returned, agent.message_emitted, agent.action_proposed, agent.action_committed, provenance.walked, doc.section_generated, model + version + prompt_version metadata
• output shape: session transcript, structured tool-call log, citations emitted, drafts proposed, decisions made, latency and cost rollup, replay manifest
• wave: Wave 2 (the harness itself ships Wave 2; the doc projection ships in parallel so every session becomes citable)

This projection is what makes agent sessions queryable as docs. Regulatory review, federal audit, customer dispute resolution, and gradeboard eval all read this projection.

6.6 Open Corpus Projections

open_corpus_export.<version>

• doc_kind: aggregate_intelligence
• audience: public (CC-BY-SA license), federal, partner (Anthropic / Hugging Face)
• cadence: quarterly versioned snapshot
• sources: all entity_profile snapshots in Harvest Directory at version cutoff, all aggregate_intelligence snapshots, all anonymized event_digest and proof_package summaries cleared by the anonymization policy
• output shape: versioned dataset manifest (parquet / JSONL), model card training-data appendix, anonymization policy version, fact count, citation count, eval results when used to train the FIN open model
• wave: Wave 4 (post-Nathel funding; release gated on minimum corpus size — ≥100 commodity profiles, ≥50 region profiles, ≥200 aggregate intelligence snapshots, ≥6 months operational ground truth)

This projection is the corpus released to Hugging Face for training the FIN open produce-intelligence model. Every fact in the corpus is citable back to events through the provenance chain. Federal-grade reproducibility property.

6.7 Platform Observability Projections

kb_health_summary

• doc_kind: system_projected
• audience: FIN operator
• cadence: daily 06:00 UTC
• sources: existing kb_health_summary_projection (migration 043)
• output shape: domain counts, embedding coverage, stale entries, ingestion drift flags
• wave: Wave 1 (this can ship immediately because the projection table already exists; just needs the markdown renderer)

channel_health.<tenant>

• doc_kind: system_projected
• audience: FIN operator
• cadence: every 15 minutes
• sources: email queue status, Telnyx SMS/voice health, GitHub vault webhook freshness, MCP key validity, render status
• output shape: per-tenant channel health dashboard doc
• wave: Wave 2 (FIN Central control plane)

tenant_readiness.<tenant>

• doc_kind: system_projected
• audience: FIN operator
• cadence: on tenant lifecycle events + daily reconciliation
• sources: tenant provisioning checklist from spec §13, smoke-test events
• output shape: readiness rollup, blockers, current onboarding step (per project_agentic_onboarding_program memo)
• wave: Wave 2

agent_enrichment_metrics

• doc_kind: system_projected
• audience: FIN operator
• cadence: weekly
• sources: doc.edit_proposed events, acceptance/rejection events, gradeboard measurements
• output shape: proposals per agent, acceptance rate, doc enrichment rate, reasoning quality lift (gradeboard A/B)
• wave: Wave 3

federal_readiness_tracker

• doc_kind: system_projected
• audience: FIN operator
• cadence: weekly
• sources: federal-runtime feature flags, recall-drill events, audit-export status, SBOM/provenance status, residency notes (per project_strategic_research memo)
• output shape: federal readiness rollup, missing artifacts, last drill date, next steps
• wave: Wave 4

6.8 Onboarding & Adoption Projections

onboarding_progress.<tenant>.<actor>

• doc_kind: portal_page
• audience: employee (the actor), employee (admin), FIN operator
• cadence: event-driven on onboarding step completion
• sources: 10-step onboarding events (per project_pk_v2_implementation memo), project_agentic_onboarding_program 3-act cohort design
• output shape: progress doc, next step prompt, blockers, suggested helper resources
• wave: Wave 1 (Nathel cohort needs this immediately)

cohort_status.<tenant>

• doc_kind: system_projected
• audience: FIN operator
• cadence: daily
• sources: aggregate onboarding events across cohort actors
• output shape: cohort progress, lagging actors, common blockers, suggested interventions
• wave: Wave 2

6.9 Marketplace Projections (Deferred)

Marketplace projections are deferred to Wave 4 per the marketplace planning memo’s 3-phase A2A path. Catalog placeholders only:

• marketplace_offer.<offer_id> — portal_page, customer audience
• marketplace_match_log.<offer_id> — event_digest, FIN operator
• marketplace_blind_intelligence.<commodity>.<window> — aggregate_intelligence, public

Each will be defined when the marketplace phase opens.

6.10 Cross-Analysis & Data Fusion Projections (Reserved for v0.2)

Cross-analysis projections compose multiple entity profiles or event families into new insight artifacts. These are reserved for catalog v0.2 once the single-source entity profiles in §6.2 are proven.

Placeholders (full specs in v0.2):

• commodity_season_heatmap — embedded inside commodity_profile
• vendor_commodity_fulfillment_quality — embedded inside vendor_profile
• region_commodity_pricing_intelligence — embedded inside region_intelligence
• customer_commodity_loyalty — embedded inside customer_profile
• lot_lineage_waterfall — embedded inside recall_trace
• co_occurrence_graph.<commodity> — embedded inside commodity_profile “frequently shipped with”
• customer_substitution_patterns.<customer> — embedded inside customer_profile

7. (Reserved) Wave Assignment Rollup

Inline wave tags are recorded on each entry in §6. A consolidated rollup table is deferred to catalog v0.2 once Wave 1 ships and the Wave 2 entry list is firm.

8. (Reserved) Renderer Version Registry

Per-projection renderer_version history lives in docs/projections/renderer-versions.md (to be created in Wave 1). The catalog references the registry but does not duplicate it.

9. (Reserved) Harvest Directory Public Mirror Specification

The full public mirror specification — including anonymization policy references, contribution model (CC-BY-SA, PR review), and the public projection inventory — is deferred to catalog v0.2 alongside the Wave 3 work that ships the first public commodity profile.

10. Contribution Rules

• New projection entry: add to §6 in a single PR with doc_kind, source plane, audience, cadence, sources, expected output shape, and wave tag.
• New doc_kind: requires spec amendment first (§3 + §4 of the Engineering Spec), then catalog §5 update, then §6 entries.
• New audience: requires permission model update in the Engineering Spec before any projection can target it.
• Wave promotion (e.g. Deferred → Wave 4): requires explicit dated note in the catalog entry.
• Renderer version bumps: recorded in the renderer-version registry, not inline in §6.

11. Catalog Maintenance

This catalog is itself a doc and lives in the GitHub vault. It is indexed into knowledge.domain = doc like any other authored doc. The catalog is its own decision_record for the system of intelligence surface area.

Stale projection entries (no production traffic for 90 days) are flagged in the agent_enrichment_metrics projection so they can be reviewed and archived.