LAYER TWO · PHASE 1 · REQUIRED FIRST

Knowledge.

What the system is allowed to trust and use. Four documents that stop your AI from treating a rumor and a contract as equal.

Why this layer exists

Most AI systems treat every input as equal. A forwarded email, a signed lease, a half-finished draft, and a market rumor all land in the context window with the same weight. That is why systems hallucinate policy: nothing told them what outranks what.

Layer 2 installs the missing hierarchy. You classify what you know by authority, gather the small set of files the system must always load, and write down what you are optimizing for right now — and what you are explicitly not doing. The payoff is an AI that cites your files, at the right level of trust, instead of the internet's guesses.

The test: can the system tell doctrine from a draft? Ask it. If it cannot name the authority level of the document it just quoted, this layer is not done.

Authority Classification: Levels 0–4

This is the sharpest tool in Groundwork. Every piece of information gets a level, and higher levels always override lower levels. The Operator must cite the level when making material recommendations.

Level	Name	What it is	Examples
0	Core Doctrine	Non-negotiable. Overrides everything.	Identity Charter, Decision Principles, legal mandates, operating agreements
1	Institutional Standards	Approved ways of working.	Brand Voice, internal policies, approval thresholds, communication standards
2	Validated Knowledge	Reviewed and accepted.	Executed contracts, cleaned data, completed reports, third-party verified information
3	Working Material	Useful but not authoritative.	Drafts, preliminary analysis, internal notes, external proposals
4	Untrusted / Raw	Requires caution.	Forwarded emails, unverified quotes, market chatter, personal notes

Document 04: Authority Classification

03_Authority_Rules/Authority_Classification.md

# Authority Classification

## Level 0 — Core Doctrine
Non-negotiable. Overrides everything.
Examples: Identity Charter, Decision Principles,
legal mandates, operating agreements.

## Level 1 — Institutional Standards
Approved ways of working.
Examples: Brand Voice, internal policies,
approval thresholds, communication standards.

## Level 2 — Validated Knowledge
Reviewed and accepted.
Examples: executed contracts, cleaned data,
completed reports, verified third-party info.

## Level 3 — Working Material
Useful but not authoritative.
Examples: drafts, preliminary analysis,
internal notes, external proposals.

## Level 4 — Untrusted / Raw
Requires caution.
Examples: forwarded emails, unverified quotes,
market chatter, personal notes.

Rule: higher levels always override lower levels.
Cite the Authority Level when making material
recommendations.

How to apply it: you do not need to label every file you own on day one. Classify as you touch things. When a document enters the vault or gets used in a real decision, give it a level — a line of frontmatter or a single word in the filename is enough.

Document 05: Context Vault v1

02_Current_Context/Context_Vault.md

The short list of files the system must always load first. This is not your archive — it is the briefing packet.

# Context Vault v1

## Always load
- Identity Charter
- Brand Voice
- Decision Principles
- Current Strategic Priorities
- Active Constraints

## Domain files (add only what is necessary)
-
-
-

Keep this list short. A vault that contains
everything contains nothing.

How to apply it: the discipline is subtraction. Every file on the always-load list costs attention on every single task. If a file has not changed an answer in a month, demote it to a domain file that gets loaded on request.

Document 06: Strategic Priorities

02_Current_Context/Strategic_Priorities.md

What is actually being optimized this cycle — and what is not. Without this file, the system optimizes for whatever the last prompt implied.

# Strategic Priorities
Last updated: [date]

## Current top priorities (in order)
1.
2.
3.

## Active focus this cycle
-

## Explicit non-priorities
- [things you are deliberately not doing,
  so the system stops suggesting them]

How to apply it: the non-priorities section works harder than the priorities. “No new acquisitions this quarter” kills a whole category of well-meaning but useless AI suggestions. Date the file and re-touch it whenever the cycle turns.

Document 07: Active Constraints

02_Current_Context/Active_Constraints.md

Hard limits, soft limits, and current capacity. Constraints are context: an AI that does not know your limits will recommend plans you cannot run.

# Active Constraints

## Hard constraints
- No irreversible external actions without
  human approval
- No access to financial systems
- No legal document finalization
- No public communications without review

## Soft constraints
- Prefer internal tools over new third-party
  services
- Keep recommendations tightly scoped
- Prefer short drafts that need minimal revision

## Current capacity notes
-

How to apply it: hard constraints are enforced in Layer 3 governance; here they are stated as fact. Capacity notes are the human part — “one person, ten hours a week for this” changes every recommendation the system should make.

Vault hygiene: making files findable

Classification only works on files the system can actually read and find. Three habits cover most of it:

Name files so the name answers questions. 2026-08-01_Lease_AcmeCorp_Executed.pdf tells an AI the date, type, party, and status. scan_003.pdf tells it nothing.
Keep text as text. Markdown beats PDF, and OCR'd PDF beats an image. If a document only exists on paper, it is invisible to your system.
One home per document. Duplicates with different edits are an authority conflict you created yourself.

You are done with Layer 2 when

All four documents exist and the vault folder structure is real, not planned
The always-load list is short enough to read in under two minutes
Documents you actually use in decisions carry an authority level
Strategic Priorities is dated within the current cycle and lists explicit non-priorities
You can ask “what authority level is this claim based on?” and get a correct answer

Next: the system knows who you are and what to trust. Now it needs rules for what it may do. That is Governance — the three zones, escalation rules, and the scope of your first agent.

← Layer 1: Identity Layer 3: Governance →