The Groundwork Guide / Scorecard
WHERE DO YOU STAND

The AI Readiness Scorecard.

Twenty-four questions across the six layers — the same diagnostic Adam uses in audit sessions. Five minutes, honest answers, a named weakest layer.

How the scoring works

Twenty-four questions, four per layer. About five minutes. Answer honestly — the value is in knowing exactly where you stand, not in getting a high score.

Every question is weighted. Standard questions count one point. High impact questions count two and carry a tag. Yes earns the full weight. Partial earns half, rounded up. No earns zero. Thirty-six points total.

Open the note under any question to see why it matters and what the fix is. Results show your overall percentage, a layer-by-layer breakdown, your top priorities, and the recommended next move.

Layer 1: Identity

Would a stranger write like you — and decide like you — after reading your files? The fixes live in Layer 1.

1. Do you have a written Identity Charter, including what you will never become? High impact
Why it matters, and the fix

Why it matters. Without a written identity, the model fills the gap with the average of the internet. That is why AI output feels generic.

The fix. Write 01_Identity/Identity_Charter.md: one honest paragraph, what you stand for, and at least three things you will never become.

2. Are your voice rules written as enforceable rules, with on-voice and off-voice examples? High impact
Why it matters, and the fix

Why it matters. “Professional but friendly” is useless to a machine. Rules plus labeled examples are enforceable; adjectives are not.

The fix. Write 01_Identity/Brand_Voice.md with tone, style rules, vocabulary, and at least one real on-voice and off-voice example.

3. Are your decision principles written down — rules that would have changed a real decision this year?
Why it matters, and the fix

Why it matters. Principles are the tiebreakers the system reaches for when two good options conflict. Unwritten, they get improvised.

The fix. Write 01_Identity/Decision_Principles.md: five to nine rules drawn from hard calls you actually made.

4. Could a stranger read your identity files and then write and refuse the way you would?
Why it matters, and the fix

Why it matters. This is the pass condition for the whole layer. If a stranger could not do it, neither can an AI.

The fix. Load the three files into a fresh AI session and ask for a draft. Sharpen the files until the first pass sounds like you.

Layer 2: Knowledge

Can the system tell doctrine from a draft? The fixes live in Layer 2.

5. Does information you rely on carry an authority level (0–4), so doctrine outranks drafts? High impact
Why it matters, and the fix

Why it matters. Systems hallucinate policy because nothing told them what outranks what. A rumor and a contract must never be equal.

The fix. Write 03_Authority_Rules/Authority_Classification.md and classify documents as you touch them. Higher levels always override lower.

6. Do you have a Context Vault — a short, deliberate always-load list? High impact
Why it matters, and the fix

Why it matters. A vault that contains everything contains nothing. The briefing packet must be short enough to load on every task.

The fix. Write 02_Current_Context/Context_Vault.md: identity files, priorities, and constraints always load; domain files load on request.

7. Are your current priorities — and explicit non-priorities — written down and dated?
Why it matters, and the fix

Why it matters. Without this file, the system optimizes for whatever the last prompt implied. Non-priorities kill whole categories of useless suggestions.

The fix. Write 02_Current_Context/Strategic_Priorities.md, date it, and re-touch it whenever the cycle turns.

8. Are your key files digitized, consistently named, and findable in under a minute?
Why it matters, and the fix

Why it matters. Classification only works on files the system can read and find. Paper is invisible; scan_003.pdf is nearly so.

The fix. Name files DATE_DocType_Entity_Status, keep text as text, and give every document one home.

Layer 3: Governance

Does the system know when to act, when to draft, and when to stop? The fixes live in Layer 3.

9. Is every AI task assigned a zone in writing — Auto-Execute, Draft & Wait, or Never Automate? High impact
Why it matters, and the fix

Why it matters. An agent without written rules does not have no rules — it has improvised rules, invented fresh in every session.

The fix. Write 03_Authority_Rules/Automation_Framework.md. Zone your top ten tasks; unlisted work defaults to Draft & Wait.

10. Are there written escalation rules the system must follow, including “do not proceed on silence”? High impact
Why it matters, and the fix

Why it matters. Escalation converts ambiguous situations into paused ones instead of guessed ones — even mid-task.

The fix. Write 03_Authority_Rules/Escalation_Rules.md: triggers, a named human, and the three-part escalation statement.

11. Is there an explicit allowed-tools list, including what the system may never access?
Why it matters, and the fix

Why it matters. The “never” list is the one that protects you. Money and signatures stay human, permanently.

The fix. Write 04_Capabilities/Allowed_Tools.md. Everything not on a list is off-limits by default.

12. Does every external-facing draft have one named human approver who reads the actual artifact?
Why it matters, and the fix

Why it matters. “Someone reviews it” means nobody reviews it. Draft & Wait only works if the wait is real.

The fix. Name the approver per category in Agent_Scope.md and keep a one-line approval log: date, item, approver.

Layer 4: Continuity

Would this survive the founder being gone for ninety days? The fixes live in Layer 4.

13. Would your AI system keep behaving as designed if you were gone for ninety days? High impact
Why it matters, and the fix

Why it matters. Most setups fail this on day one because everything lives in one head and one vendor's chat history.

The fix. Write 01_Identity/Continuity.md: what must survive, who inherits decision rights, how the system is stopped.

14. Is your operating context in portable files you control, not trapped in one vendor's history? High impact
Why it matters, and the fix

Why it matters. Platform independence is continuity. Markdown in a folder you control works with any AI; no platform owns it.

The fix. Move doctrine out of chat history into the vault. Back the folder up like it is money. Test one cold restore on a fresh platform.

15. Are succession and override rules written with real names — and does the successor have access?
Why it matters, and the fix

Why it matters. A succession rule naming a role you hope to hire is fiction. Names, access, and a known off-switch are the real thing.

The fix. Fill the succession and override sections of Continuity.md with names, then verify the passwords and vault access actually work.

16. Is working memory cleared regularly — promoted into real documents instead of hoarded?
Why it matters, and the fix

Why it matters. Session notes that never get cleared become a second, contradictory source of truth.

The fix. Keep 05_Working_Memory/Session_Notes.md. After major sessions: promote what matters to a real authority level, archive the rest.

Layer 5: Measurement

How would you know this is working next month? The fixes live in Layer 5.

17. Could you answer “is this working?” with evidence rather than a feeling? High impact
Why it matters, and the fix

Why it matters. Without evidence you either over-trust (drift goes unnoticed) or under-trust (you redo work the system did well).

The fix. Write 06_Output_Standards/Measurement_Notes.md with the four signals: on-voice, cites authority, appropriate escalations, trusted drafts.

18. Do you run a monthly review: what broke, what was wrong, what context was missing? High impact
Why it matters, and the fix

Why it matters. The review is where drift gets caught before a tenant or client catches it for you.

The fix. Put thirty minutes on the calendar monthly. Answer the four review questions and write one dated entry.

19. Do findings actually change the documents — miss, trace, edit, retest?
Why it matters, and the fix

Why it matters. Measurement that never edits a file is theater. Three misses with one root cause is a standing order to fix the document.

The fix. Trace every miss to the file that caused it: tone to Brand Voice, invented policy to doctrine, wrong action to zones or scope.

20. Do you know what share of drafts ship with light edits versus quiet rewrites?
Why it matters, and the fix

Why it matters. This is the bluntest signal there is. If you quietly rewrite everything, some document is lying to the system.

The fix. Track it roughly from your approval log: shipped as-is, light edits, or rewritten. The trend matters more than the number.

Layer 6: Operating Rhythm

Who owns each layer, and when is it reviewed? The fixes live in Layer 6.

21. Does every layer have a named owner? High impact
Why it matters, and the fix

Why it matters. A foundation nobody owns quietly becomes fiction while the Operator keeps citing it with full confidence.

The fix. Write 06_Output_Standards/Operating_Rhythm.md. In a small operation every line may say the same name — that is fine, as long as it is written.

22. Are weekly, monthly, and quarterly reviews on an actual calendar — and happening? High impact
Why it matters, and the fix

Why it matters. Rhythm is minutes per week, not hours. Without dates, decay wins by default.

The fix. Fifteen minutes weekly (clear session notes), thirty monthly (measurement review), an hour quarterly (re-score, audit zones, verify succession).

23. Is there a written update protocol — changes proposed, approved by the owner, versioned?
Why it matters, and the fix

Why it matters. Foundation edits are Draft & Wait work. The Operator may propose; it may never apply.

The fix. Add the update protocol to Operating_Rhythm.md: proposed in writing, approved by the layer's owner, versioned with a dated changelog line.

24. Have you re-scored or audited the whole system in the last quarter?
Why it matters, and the fix

Why it matters. The score trend is itself a measurement. A foundation that is not re-checked quarterly is drifting unobserved.

The fix. Re-run this scorecard at each quarterly review and compare layer by layer against last quarter.

0 of 24 answered.

Your answers stay on this page. Nothing is sent, stored, or tracked.

← Layer 6: Operating Rhythm Back to the main site →