FIN Intranet Human-Agent Projection Execution Program

Date: 2026-05-13 Status: Draft v0.1 Related audit: docs/audits/FIN_INTRA_GTM_PLANNING_CODEBASE_GAP_AUDIT_2026-05-13.md 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

1. Program Objective

Build FIN’s human-agent projection layer without losing the strategic planning scope. The program must support:

• FIN Canon intranet
• GitHub-backed planning/docs vault
• artifact and document graph
• doc MCP tools
• FIN Central multi-tenant operations
• Nathel anchor proof loop
• customer/employee portals
• Tier 0 MCP path
• Tier 2/3/4 offering rollout
• federal readiness path

2. Program Rules

1. No new document table.
2. Preserve four-table architecture.
3. Proof first, especially for Nathel-facing claims.
4. Keep implementation waves small enough for a solo operator.
5. Do not build marketplace before portal/intranet proof.
6. Do not enable automatic reverse projection in v1.
7. Treat Markdown as presumed canonical until DOCX diff proves otherwise.
8. Every wave has tests and a visible demo artifact.

3. Phase 0: Canonicalization And Audit Closure

Goal: Convert the sprawling planning session into controlled source material.

Deliverables:

• This audit/design/spec/program document set
• DOCX-vs-Markdown canonicality report
• source index for FIN Intra_GTM
• final decision on canonical doc source
• frontmatter/classification plan for pilot docs

Acceptance gates:

• all planning files inventoried
• DOCX companions compared to Markdown
• no plan relies on documents table
• open decisions captured

4. Phase 1: Content Model And Parser Foundation

Goal: Create the core doc/content model without UI sprawl.

Work items:

• add shared doc schemas
• add markdown parser
• add wikilink/backlink resolver
• add chunking helper
• add doc event payload schemas
• add fixture tests using FIN Intra_GTM

Acceptance gates:

• parser handles frontmatter, headings, tags, wikilinks
• doc schema validates pilot docs
• no DB migrations beyond allowed schema additions to existing JSON content paths
• tests pass locally

5. Phase 2: GitHub Vault Ingest And Doc MCP Tools

Goal: Make docs queryable by agents.

Work items:

• GitHub webhook route
• tenant vault config
• doc ingest pure functions
• knowledge.domain = doc upsert
• operational_memory chunk write path
• search_docs
• get_doc
• get_backlinks
• get_doc_graph
• list_by_tag
• get_decisions
• list_recent_changes

Acceptance gates:

• sample repo push indexes docs
• doc search returns citations
• unauthorized actor cannot access restricted doc
• doc MCP tools have contract tests
• events emitted for index/query

6. Phase 3: Human Reader And File Viewer Integration

Goal: Make humans able to browse docs, files, decisions, and proof packages in Workspace.

Work items:

• fix /files/:artifactId route
• add doc reader
• add doc explorer mode
• add backlinks/related artifacts panel
• add mobile reader path
• extend artifact upload support only after parser/security review

Acceptance gates:

• desktop doc reader renders pilot doc
• mobile doc reader renders pilot doc
• file share link works
• restricted doc hidden from unauthorized actor
• source attribution visible

7. Phase 4: FIN Canon Pilot

Goal: Ship a usable FIN Canon intranet for collaboration and strategic planning.

Work items:

• create private FIN Canon GitHub repo or designated private vault
• move/copy canonical Markdown docs into vault
• configure Quartz pilot behind Cloudflare Access
• index vault into FIN
• expose doc MCP endpoint to internal/partner agents
• publish onboarding doc

Acceptance gates:

• partner can read intranet URL
• agent can query docs via MCP
• updates are indexed after commit
• access is logged
• no private tenant data appears in public/partner scope

8. Phase 5: FIN Central Control Plane

Goal: Prepare for multi-tenant deployments.

Work items:

• tenant readiness projection
• channel health dashboard
• vault status dashboard
• render status dashboard
• MCP key dashboard
• KB freshness dashboard
• Nathel proof ledger
• federal readiness tracker
• Orgo exec-agent status placeholder

Acceptance gates:

• operator can see tenant readiness in one place
• failed channel/vault/render states are visible
• Nathel proof items can link events/artifacts/docs
• dashboard states have tests

9. Phase 6: Nathel Anchor Rollout

Goal: Convert live test evidence into funded anchor deployment.

Tracks:

• internal Nathel/FIN intranet section
• employee portal slice
• portal-only customer pilot
• portal-plus-agent customer pilot
• exec-agent candidate pilot
• proof package and case study

Acceptance gates:

• stakeholder-specific proof package exists
• at least one customer-safe portal slice is demonstrated
• agent-enabled portal pilot has deterministic boundaries
• value claims have evidence paths
• funding conversation package is complete

10. Phase 7: Tier 0 MCP Private Beta

Goal: Launch public-ish read-only MCP product without tenant data risk.

Work items:

• external MCP contract v1
• docs/examples
• key provisioning
• usage metering
• rate limits
• public produce intelligence tool subset
• private beta cohort

Acceptance gates:

• external tools cannot access tenant docs/events/artifacts
• usage events recorded
• docs site published
• beta users can call versioned tools

11. Phase 8: Tier 2 And Tier 3 Productization

Goal: Make the repeatable service offers real.

Tier 2 work:

• onboarding checklist
• multi-vendor connection template
• voice/WhatsApp readiness checklist
• live Obsidian/GitHub projection option
• support SLA tracking

Tier 3 work:

• Orgo integration contract
• visible operation audit log
• exec-agent provisioning
• real-time voice readiness
• monthly strategy report template

Acceptance gates:

• Tier 2 onboarding can be run from FIN Central
• Tier 3 Orgo pilot is specified before build
• service boundaries are explicit

12. Phase 9: Federal Readiness Package

Goal: Convert the architecture into credible federal positioning.

Work items:

• federal-safe MCP profile
• model/provider routing notes
• TAA/NDAA infrastructure inventory
• recall/traceability proof package
• audit export sample
• SBOM/provenance documentation

Acceptance gates:

• federal read-only intelligence MCP demo
• traceability proof bundle
• compliance posture doc
• no unsupported certification claims

13. Workstream Register

ID	Workstream	Priority	Owner	Dependencies
INTRA-001	Planning canonicalization	P0	FIN	none
INTRA-002	Doc schema/parser	P0	Engineering	INTRA-001
INTRA-003	GitHub vault ingest	P0	Engineering	INTRA-002
INTRA-004	Doc MCP tools	P0	Engineering	INTRA-002
INTRA-005	File/doc deep links	P1	Workspace	INTRA-004
INTRA-006	Doc reader/editor v1	P1	Workspace	INTRA-004
INTRA-007	Quartz FIN Canon pilot	P1	Ops	INTRA-003
INTRA-008	FIN Central readiness	P1	Platform	INTRA-003
INTRA-009	Nathel proof ledger	P1	Product/Ops	INTRA-008
INTRA-010	Tier 0 MCP beta	P1	Platform/GTM	INTRA-004
INTRA-011	Portal entitlement model	P1	Platform	INTRA-008
INTRA-012	Orgo exec-agent contract	P2	Platform/Ops	INTRA-008
INTRA-013	Federal readiness package	P2	Product/Ops	INTRA-010

14. Stop Conditions

Pause and reassess if:

• live Nathel proof work is delayed by speculative platform features
• doc architecture requires a fifth runtime table
• permission tests cannot prove safe customer/vendor filtering
• external MCP risks exposing tenant data
• operator workload exceeds the solo execution budget
• DOCX and Markdown conflicts reveal unresolved strategic divergence

15. First Implementation Slice Recommendation

The first code slice should be:

1. doc schema
2. markdown parser
3. fixture tests on two FIN Intra_GTM docs
4. knowledge.domain = doc upsert helper
5. get_doc and search_docs contract tests

This slice proves the architecture before UI, Quartz, portals, Orgo, or GTM work expands the surface area.