Chapter 2 templates

This page is generated from repo/templates/ch02/; the repo holds the latest version.

Attribute Map Worksheet (Chapter 2)

Note: the six attributes compete, so the deliverable is not "all of them"; it is a ranking. Rate each attribute's relevance per task type, then write the global ranking and the reason.

Six attributes × your task types

Relevance: high / medium / low

Attribute	Task type 1: ____	Task type 2: ____	Task type 3: ____
Correctness (the endpoint)
Process soundness (no dangerous moves on the path, no doubling down on an error)
Safety (no overstepping authority, no leaking, no promising)
Cost
Latency
Reproducibility (same input, stable distribution of behavior)

The ranking (the deliverable)

Example: Shore & Summit ranks Mini's attributes safety > correctness > cost > latency.

• My ranking: ____ > ____ > ____ > ____ > ____ > ____
• Reason (why first place is first):

"Who has this ranking lost to" log

Every time a metric conflict is ruled on by the ranking, add a row:

Date	Conflict (which two attributes fought)	Ruled in favor of	What the losing side paid

Source file: templates/ch02/attribute-map-worksheet.md

---

Intended Use & Action Boundary Sheet (extended, Chapter 2)

Note: Chapter 1's one-pager + attribute priorities + the severity table = this agent's spec v1. From Chapter 8 on, this document gets a new name: the file you must change before unlocking anything. Any capability unlock changes this file first, then flips the switch.

1. Intended use (three lines)

• For whom:
• Does what:
• Does not do:

2. Action boundary (three columns)

From Chapter 8 on, this table is refined into the Action Permission Matrix (tool × condition, templates/ch08/) and folded back here.

Autonomous	Needs confirmation	Forbidden

3. Attribute priorities

• Ranking (from the Attribute Map worksheet): ____ > ____ > ____ > ____
• Reason:

4. Severity table

sev	List (from the Severity Tiers worksheet)
sev-1
sev-2
sev-3

5. Verifiability inventory (three columns)

Only the third column will ever be handed to the judge (Chapter 5); not one row from the first two is allowed to leak in.

Endpoint verifiable now	Rewritable into verifiable	Rewrite won't go

Revision log

Date	Trigger (which unlock / which change)	What changed	Signature

Source file: templates/ch02/intended-use-action-boundary-sheet.md

---

Severity Tiers Worksheet (Chapter 2)

Note: upgrade the worst failures into a sev-1/2/3 list. Disputes are settled by two questions: "can this error be undone?" and "if not, who absorbs it?" If no one can name who absorbs it, file it sev-1.

The sev-1/2/3 list

• sev-1 (irreversible harm or unauthorized action): - -
• sev-2 (wrong information causing recoverable customer loss): - -
• sev-3 (experience and efficiency: tone, detours, slowness): - -

Four-verdict mapping

The four verdicts (pass / concern / unsafe / unclear) judge individual cases; sev grades failures.

Verdict	Maps to	Criterion
pass	none	no failure at the endpoint or on the path
concern	≈ sev-3, or an attempted sev-1/2
unsafe	≈ sev-1/sev-2 red line hit
unclear	to arbitration	undecidable cases go to human triage, never force-filed

Stratified report format

sev-1 gets its own count and never enters an average.

Tier	Cases	pass	concern	unsafe	unclear
sev-1 relevant
sev-2 relevant
sev-3 relevant

Source file: templates/ch02/severity-worksheet.md