New Capability: Knowledge Infrastructure Layer (KIL) — UCA Layer 10

Date: 2026-04-21 UCA Layer: 10 (prose canonical), complementing Layers 1-9 (entity canonical) Origin: Agents and contributors had no structural path to team canon. Canon lived in MEMORY.md lines, scattered skill files, agent-definition headers, and an unindexed docs/ tree. Findability was "grep and hope," and onboarding required institutional memory of where things hid. KIL built the addressable prose layer that makes team canon reachable by routing, not by recall. Impact: Every agent and contributor now reaches any piece of team canon in one to three reads, starting from a single entry point (wiki/INDEX.md). New canon adds one Wiki file plus an index entry — sprawl is no longer the default.

What Was Built

The Knowledge Infrastructure Layer establishes /mnt/d/Projects/value-first-operations/wiki/ as the prose canonical layer of the Unified Canonical Architecture — the tenth layer atop the nine entity-canonical layers completed Apr 13. Where Layers 1-9 hold structured records (shows, episodes, articles, friends, organizations, methodology entities, customer health, company health), Layer 10 holds the prose that governs how the team operates: conventions, glossary, agent guide, architecture narrative, critical lessons, onboarding, per-BU pages, ADRs, and module orientation for cross-BU capabilities.

The layer works by addressing, not by archaeology. wiki/INDEX.md is the entry point. From there, conventions.md, glossary.md, and agent-guide.md cover 80% of routing queries. The glossary functions as a term index with canonical pointers — it never duplicates canonical content, it points to it (Sanity for framework concepts that render on the website, skills/methodology/ for operational and program-shaped content, per Canonical Architecture Rule Q1=C, codified as ADR 0002). Cross-layer references use the @import directive pattern (ADR 0005), so agents load canon structurally rather than reading duplicated prose stitched into every system prompt.

The migration compressed the team's canonical footprint while increasing its reachability. MEMORY.md dropped from 49KB to 5.5KB — a 79% reduction with zero broken pointers, because every removed line either moved to a Wiki file or was already covered by an enforcement skill. The monorepo CLAUDE.md thinned from 38KB to 21KB by replacing embedded prose with @import directives to the Wiki. Four canonical duplications were eliminated: methodology content that had been scattered across skills and agent files, QMS documentation fragmented across multiple roots, audit reports split between Leadership/reports/ and docs/quality/audits/, and the "runbook" vs "operating procedure" ambiguity (ADR 0004 retired "runbook" in favor of SOP canonical at docs/operating-procedures/). Folder structure across 206 files was rationalized: scripts/hooks/ became scripts/session-hooks/, website docs consolidated, client variants normalized, agent-internal reports moved to output/.

Infrastructure Changes

Change	Before	After
Prose canonical layer	None — canon scattered across MEMORY.md, skills/, agent defs, docs/	`wiki/` as UCA Layer 10 with INDEX, conventions, glossary, agent-guide, architecture, critical-lessons, onboarding, adr/, bu/, modules/
Entry point for team canon	None — "check everywhere"	`wiki/INDEX.md` — single addressable root
MEMORY.md size	49KB (dense index, many broken/implicit pointers)	5.5KB (79% reduction, zero broken pointers)
CLAUDE.md size	38KB (embedded prose for conventions, agent model, architecture narrative)	21KB (`@import` directives to Wiki pages; prose lives once, referenced everywhere)
Methodology canonical rule	Implicit — duplicated between Sanity, skills/methodology/, and agent prompts	Explicit — ADR 0002 (Q1=C): Sanity canonical for website-rendered framework concepts; `skills/methodology/` canonical for operational/program-shaped content
"Runbook" vs "operating procedure"	Two terms, one concept, inconsistent folder placement	ADR 0004: "runbook" retired; SOP canonical at `docs/operating-procedures/`
Audit/capability/CAR report location	Split between `Leadership/reports/` (primary) and scattered `docs/` roots	ADR 0003: `docs/quality/{audits,capabilities,cars}/` canonical; `Leadership/reports/` as mirror
Cross-layer reference pattern	Copy-paste duplication	`@import` directive (ADR 0005) — structural load, single source
Wiki write permission model	Undefined (no gateway existed)	ADR 0006: contributor-editable with frontmatter owner; no Ledger/Canon-style gateway (internal collaborative canon)
`rules.md` format	Undefined — clients had ad-hoc operational notes	ADR 0007: three-line format (Rule / Why / How-to-apply); team-visible per-client operational rules
Agent-internal reports	Polluted `docs/` tree and `Leadership/reports/`	Moved to per-agent `output/` directories
Folder drift (206 files)	Mixed conventions (`scripts/hooks`, capitalized transcript dirs, scattered website docs)	Normalized (`scripts/session-hooks`, lowercase transcripts, consolidated website docs)

Implementation

File	Purpose
`wiki/INDEX.md`	Entry point — every Wiki navigation starts here
`wiki/conventions.md`	Team-wide operating rules (legal entity naming, language, taxonomy, voice, data tier hierarchy, git operations, routing rules, model tiering)
`wiki/glossary.md`	Term index with canonical pointers — never duplicates canon, always routes to it
`wiki/agent-guide.md`	Delegation model, gateway patterns (Ledger/Canon/Showcase), self-correction, verification-before-completion, data tier hierarchy, model tiering
`wiki/architecture.md`	System architecture narrative — Wiki ↔ UCA ↔ Obsidian ↔ Sanity ↔ WCP relationships
`wiki/critical-lessons.md`	Orphan lessons not covered by enforcement skills (Archivist owns)
`wiki/onboarding.md`	Contributor + agent onboarding entry point
`wiki/adr/0001-.md` through `wiki/adr/0007-.md`	Seven Architecture Decision Records codifying KIL decisions
`wiki/bu/{relay,trellis,marquee,provost,foundry,exchange}.md`	Per-BU canonical pages (one per business unit)
`wiki/modules/`	Cross-BU platform capability orientation (apps, commerce-hub per OQ-G8)
`CLAUDE.md` (monorepo)	Thinned — now uses `@import` directives to Wiki pages
`CLAUDE.md` (D drive cockpit)	`@imports` monorepo CLAUDE.md (no change to pattern, but the target is now Wiki-aware)
`.claude/codebase-index.md` § 480	Dewey section for Wiki (new classification)
`docs/plans/kil-audits/agents-tree-salvage-report.md`	Content salvage audit of 69-agent implementation tree (Phase 4 deliverable)

Architecture Decision Records

Seven ADRs codify the canonical rules KIL introduced:

ADR	Decision	Why it matters
0001	Directory name `wiki/` (vs `docs/wiki/`, `knowledge/`, `team/`)	Short, unambiguous, standard industry convention; no nested-in-docs confusion
0002	Methodology canonical rule Q1=C	Resolves "where does methodology X live?" permanently — Sanity when website-rendered framework; skills/ when operational/program
0003	`docs/quality/` canonical for audits/capabilities/CARs	Consolidates Q's domain; `Leadership/reports/` becomes the mirror
0004	"Runbook" retired → SOP canonical at `docs/operating-procedures/`	Eliminates terminology drift; one word for one concept
0005	`@import` directive pattern for cross-layer reference	Structural load instead of duplicated prose — canon lives once
0006	Wiki write permission model (contributor-editable + frontmatter owner)	No Ledger/Canon-style gateway — Wiki is internal collaborative canon, not external system writes
0007	`rules.md` three-line format (Rule / Why / How-to-apply)	Team-visible per-client operational rules, consistent structure across 18 active clients

Usage

For any agent or contributor asking "where does X live?"

# Read the entry point
Read /mnt/d/Projects/value-first-operations/wiki/INDEX.md

# From INDEX, route to:
#   - Term definitions → glossary.md
#   - Operating rules → conventions.md
#   - Delegation/gateways/self-correction → agent-guide.md
#   - System architecture narrative → architecture.md
#   - BU-specific canon → bu/{relay,trellis,marquee,provost,foundry,exchange}.md
#   - Historical decisions → adr/{0001-0007}-*.md
#   - Cross-BU platform capabilities → modules/

For session-start context loading (already wired into monorepo CLAUDE.md):

## Canonical References (loaded at session start)

@wiki/INDEX.md
@wiki/conventions.md
@wiki/glossary.md
@wiki/agent-guide.md
@skills/enforcement/README.md
@skills/hubspot/data-model-reference.md
@skills/ai-leadership/INDEX.md

For adding new canon:

# 1. Create the Wiki file
touch wiki/{topic}.md

# 2. Add frontmatter with owner
# 3. Add a pointer in the relevant index (INDEX.md, glossary.md, or conventions.md)
# 4. If cross-layer referenced, add @import directive to the consuming file
# 5. Never duplicate — always point to the canonical source

For verifying a canonical claim:

Glossary entry → follow the canonical pointer → read the canonical source. If the Wiki entry and the canonical source disagree, the canonical source wins by definition; open a Wiki issue so the glossary is brought into alignment.

Leader Applications

V (Operations)

V is the primary beneficiary of KIL. Before the layer existed, V's delegation decisions required re-reading scattered files on every session to figure out whether a task belonged to Squire, Showcase, Aegis, or Mender. Now the delegation and gateway patterns live in wiki/agent-guide.md, loaded structurally via @import at session start. Specific applications:

Daily ops synthesis — V's morning briefing no longer needs to re-discover canonical addresses for each domain; the routing rules in wiki/conventions.md § Routing rules are loaded at session start.
Spawning specialists — Agent roster, model tiering, and gateway enforcement all reachable from wiki/agent-guide.md without additional reads.
Capability reports and CARs — ADR 0003 makes docs/quality/{capabilities,cars}/ the canonical write target, eliminating the prior ambiguity about where V's governance reports live.
Onboarding new agents — wiki/onboarding.md replaces the prior scatter of "here are the files to read" lists in agent definitions.

Sage (Customer)

Sage's relationship intelligence work benefits from two KIL artifacts: wiki/glossary.md (Collective taxonomy, practitioner terms, participation modes, interest pipeline vocabulary all canonical with pointers) and wiki/bu/trellis.md (Collective BU canonical, supporting Trellis's work under Sage's org). Sage's daily reports on relationship signals now pull terminology from the glossary rather than re-defining terms each time.

Pax (Finance)

Pax's work on investments, renewals, and commercial health benefits from wiki/modules/commerce-hub.md (cross-BU commerce orientation) and wiki/bu/exchange.md (Store BU canonical). The Commerce Hub module in particular gives Pax a canonical reference for how Orders, Invoices, Payments, Subscriptions, Carts, and Quotes relate — previously distributed across the HubSpot data model reference, agent definitions, and ad-hoc notes in finance reports.

Dependencies

Dependency	Status	Notes
UCA Layers 1-9 (entity canonical)	Confirmed (completed Apr 13, 2026)	KIL's Layer 10 sits atop the entity layers
Content Vault (`/mnt/d/data/content-vault.db`)	Confirmed operational	Referenced by `wiki/glossary.md` as the Vault canonical
`@import` directive support in Claude Code	Confirmed (Phase 4 Wave 1 verified via `wiki/test.md` sanity check marker)	Sanity check marker remains in place for verification; removal instruction at `docs/plans/kil-designs/wave1-import-sanity-check.md`
Enforcement skills (`skills/enforcement/`)	Confirmed operational	Wiki agent-guide.md points to enforcement skills as the runtime canonical; no duplication
Archivist agent	Confirmed operational	Owns `wiki/critical-lessons.md` curation
Aegis agent	Confirmed operational	Owns `wiki/INDEX.md`, `wiki/conventions.md`, `wiki/agent-guide.md`, `wiki/onboarding.md`, `wiki/architecture.md`
Scrivener agent	Confirmed operational	Owns `wiki/glossary.md`
BU leaders (Relay, Trellis, Marquee, Provost, Foundry, Exchange)	Confirmed operational	Each owns their `wiki/bu/{name}.md` page
Q agent	Confirmed operational	Owns `docs/quality/` per ADR 0003; Wiki references routed through Q's domain

Verification

Phase 5 validation evidence (completed 2026-04-21):

Findability probes — 3/3 agents (Sentinel, Pulse, Oracle) reached canonical ground truth in 1-3 reads starting from wiki/INDEX.md. No dead ends, no "I couldn't find it." Oracle's verdict: "Sub-second routing from agent entry point to canonical ground truth. The Wiki layer worked as designed."
Drift check — Archivist confirmed zero drift in MEMORY.md post-migration. Every removed line either landed in a Wiki file or was already covered by an enforcement skill.
Regression check — Zero Phase 4 regressions. Two critical pre-KIL latent gaps surfaced during validation (queued as post-KIL follow-ups, not KIL defects).
@import sanity check — wiki/test.md marker visible in session context confirms structural import resolution works. Marker removal instruction documented at docs/plans/kil-designs/wave1-import-sanity-check.md.

To verify the capability yourself:

# Confirm entry point exists and routes
Read /mnt/d/Projects/value-first-operations/wiki/INDEX.md

# Confirm @import loads at session start
# (Already wired into monorepo CLAUDE.md — visible in any new session)

# Confirm MEMORY.md is thinned but whole
wc -c /home/coach_chris/.claude/projects/-mnt-d/memory/MEMORY.md
# Expect: ~5,500 bytes (was ~49,000)

# Confirm CLAUDE.md is thinned
wc -c /mnt/d/Projects/value-first-operations/CLAUDE.md
# Expect: ~21,000 bytes (was ~38,000)

# Confirm ADRs exist
ls /mnt/d/Projects/value-first-operations/wiki/adr/
# Expect: 0001 through 0007

# Confirm BU pages exist
ls /mnt/d/Projects/value-first-operations/wiki/bu/
# Expect: relay, trellis, marquee, provost, foundry, exchange

Scope Boundaries (What This Is NOT)

NOT a methodology change. Value Path, 12 Traps, Core Beliefs, Four Conversations, Interest Pipeline, Three-Org Model, Unified Views, AI-Native Shift, Value Points — content unchanged. KIL clarified where each lives canonically (Q1=C rule, ADR 0002) but did not alter what any of them mean. The 8 Value Path stages are still the 8 Value Path stages.
NOT a tooling change. No new scripts. No new agents. No new MCP servers. No new slash commands. Existing agents have cleaner reference targets (they load wiki/*.md via @import instead of parsing denser CLAUDE.md prose), but their behavior and capability are unchanged.
NOT a customer-facing capability. This is internal knowledge infrastructure. Chris's clients, Collective members, and website visitors will see no direct change. The capability enables faster, more accurate delivery — a second-order improvement in every customer-facing artifact, but not a first-order customer-visible change.
NOT a gateway. Unlike Ledger (HubSpot writes) or Canon (Sanity writes), the Wiki has no write gateway. Per ADR 0006, Wiki is contributor-editable with frontmatter-declared owners. This is a deliberate design choice — Wiki is internal collaborative canon, not an external system mutation channel. If gateway-style governance is ever needed, that becomes a future ADR and a distinct capability.
NOT a replacement for enforcement skills. skills/enforcement/ remains the runtime canonical for agent execution behavior. When wiki/conventions.md and skills/enforcement/vf-platform-context.md agree, execution uses the enforcement skill. When they appear to disagree, that is a Wiki issue — they should never disagree.

Maintenance Contract

Each Wiki page carries a frontmatter-declared owner (per ADR 0006). When canonical content changes, the owner is responsible for keeping the Wiki page and the canonical source aligned — or for opening an issue when they drift.

Wiki surface	Owner	Maintenance trigger
`wiki/INDEX.md`	Aegis	New Wiki page added, major structural change
`wiki/conventions.md`	Aegis	Language rule change, routing rule change, new operational convention
`wiki/glossary.md`	Scrivener	New term enters team vocabulary, canonical source relocates
`wiki/agent-guide.md`	Aegis	New gateway introduced, model tier changes, delegation pattern updates
`wiki/architecture.md`	Aegis + Hone	UCA layer changes, cross-system relationship changes
`wiki/critical-lessons.md`	Archivist	New orphan lesson surfaces (not covered by enforcement skills)
`wiki/onboarding.md`	Aegis	Contributor or agent onboarding process changes
`wiki/adr/*.md`	Original ADR author	ADR amendments (rare — usually superseded rather than edited)
`wiki/bu/{name}.md`	Respective BU leader (Relay, Trellis, Marquee, Provost, Foundry, Exchange)	BU-specific vocabulary, pipeline, or canonical source changes
`wiki/modules/*.md`	Aegis + module-relevant leader	Cross-BU capability definition changes

When canon changes elsewhere:

If a canonical source (Sanity schema, skills/methodology/ file, HubSpot property, etc.) changes in a way that alters a definition or routing rule, the Wiki owner brings the Wiki entry into alignment in the same session.
If the Wiki entry is a pointer (not a definition), no change is usually needed — the pointer still routes correctly.
If a canonical source moves, update the pointer and note the move in the commit message referencing the relevant ADR.

When canon is added:

Add one Wiki file (if the concept deserves its own page) OR one entry in the relevant index file (glossary for terms, conventions for rules, agent-guide for agent-behavior patterns).
If the new concept requires a routing decision ("where does this live canonically?"), open an ADR.
Never create a new top-level Wiki directory without an ADR.

When the Wiki and a canonical source disagree:

The canonical source wins by definition. Open a Wiki issue so the entry is brought into alignment. Do not edit the Wiki entry to argue with canon; edit it to route to canon correctly.

Related ADRs

wiki/adr/0001-wiki-directory-name.md — wiki/ selected as the directory name
wiki/adr/0002-methodology-canonical-rule.md — Q1=C (Sanity vs skills/methodology/ stratification)
wiki/adr/0003-docs-quality-canonical.md — docs/quality/ canonical for audits, capabilities, CARs
wiki/adr/0004-runbook-retired.md — "Runbook" retired in favor of SOP canonical
wiki/adr/0005-at-import-directive.md — @import pattern for cross-layer reference
wiki/adr/0006-wiki-write-permission.md — Contributor-editable + frontmatter owner model
wiki/adr/0007-rules-md-format.md — Three-line rules.md format (Rule / Why / How-to-apply)

Related Skills and References

skills/enforcement/README.md — Runtime enforcement canonical (Wiki points to it; never duplicates it)
skills/methodology/ — Operational methodology canonical (Q1=C per ADR 0002)
skills/canonical/canonical-query-reference.md — UCA Layers 1-9 query tools
.claude/codebase-index.md § 480 — Dewey classification for wiki/
docs/plans/kil-designs/oq-decisions.md — KIL open question resolutions (Q1 through Q8)
docs/plans/kil-audits/agents-tree-salvage-report.md — Phase 4 content salvage audit
docs/plans/kil-designs/wave1-import-sanity-check.md — @import sanity check marker lifecycle

Capability: Knowledge Infrastructure Layer (KIL) — UCA Layer 10. Built 2026-04-21. Validated by Phase 5 findability probes across Sentinel, Pulse, and Oracle. Zero regressions, zero drift, three-for-three canonical-ground-truth routing.