Data Classification Table

Module 9, Lesson 9.4 · three-bucket classification + routing rule · produces Section 2 of /capstone/security-posture.md

Save this as: the filled table stays in your notebook; the Section 2 draft copies into /capstone/security-posture.md. The re-route log in Part 3 is evidence that you did it, not just talked about it.

Use it again: every time you add a new artifact kind to your pipeline — a new data source, a new folder, a new inbound channel — classify it before it ships.

Sensitive data goes local or nowhere.

Classification drives routing — no case-by-case. Once an artifact is labeled, the model it may touch is already decided. You do not negotiate this rule against convenience.

Header

Student: Date:

Default cloud model (from Module 2):

Default local model stack:

Part 1 — Classification taxonomy reference

Class P

Public

Test: already lives on the open internet with no effort on my part.

Wikipedia articles. Public blog posts. Open datasets. Published research papers. Public company documentation.

Route: any model, local or cloud.

Class P

Personal

Test: about me or my immediate household, and I have not chosen to publish it.

Essay drafts. Calendar events. Email threads. Notes about friends. Tax prep scratch. Journal entries.

Route: default cloud model (Module 2), or local. Not third-party MCP-hosted models.

Class S

Sensitive

Test: would create real harm to me or someone else if disclosed.

Medical / mental-health detail. Financial account specifics. Immigration / legal status. Anyone else's password or key. Your own credentials.

Route: local model only, or no agent at all.

If in doubt, classify up (more strict). A public item mislabeled sensitive costs you a slightly slower local-model run. A sensitive item mislabeled public costs you something you may not be able to get back.

Part 2 — Artifact inventory

Walk every folder in your capstone. Classify. Write the action.

Sources to walk: /capstone/pipeline-v1/, /my-first-loop.md, the Module 5 inbox / calendar posture, the Module 7 plugin register, your Module 8 blueprint, your notes folders. Put P (public), P (personal), or S (sensitive) in the class column. If an artifact's current routing does not match its class, the action column is where you write “re-route to local” or “audit plugin X” or “delete from shared folder.”

#	Artifact path / folder	Contains what	Class (P/P/S)	Where currently sent	Correct routing	Action
1
2
3
4
5
6
7
8
9
10
11
12

Part 3 — Sensitive-data re-route log

Document the one re-route you complete this lesson

Pick one currently-cloud-routed sensitive flow from Part 2 and re-route it to your local model stack. This is a drill, not a thought experiment. Fill the log with real values, not aspirational ones.

Artifact or flow re-routed

Why this is classified sensitive (the test that puts it in class S)

New local model endpoint (Ollama / LM Studio URL and model name)

Verification test (what request you sent; what the response looked like; confirmation network egress was zero)

Notes / surprises (quality, latency, anything you want future-you to know)

Part 4 — Section 2 draft

Copy this block into /capstone/security-posture.md. Fill bucket lists from Part 2; fill the routing rule from Part 1.

## 2. Data classification

**Public** — information already on the open internet.
- _______________________________________________________________
- _______________________________________________________________
- _______________________________________________________________
Routing: any model, local or cloud.

**Personal** — about me / my household, not published.
- _______________________________________________________________
- _______________________________________________________________
- _______________________________________________________________
Routing: default cloud model (named in Section 0) or local. No
third-party MCP-hosted models.

**Sensitive** — real harm if disclosed.
- _______________________________________________________________
- _______________________________________________________________
- _______________________________________________________________
Routing: local model only (Ollama / LM Studio), or no agent at all.

**Routing rule:** the class is the routing. Classification is done
before the artifact enters a pipeline, not during. When in doubt,
classify up.

Part 5 — Future-self check

One question, three write-lines

The test for whether you are routing honestly: which of your current routing decisions would you be unwilling to put in a public log of your security posture? That answer is where you are negotiating against the rule. Name it now, or it will surface at the worst possible time.

Section 2 ready-to-freeze checklist

Every capstone folder is walked; every artifact I found has a row in Part 2 with a class and a correct routing.
Any row where current routing does not match the class has an action (re-route, delete, audit) with a target date.
One sensitive-data flow has actually been re-routed and verified (Part 3 log is filled with real values).
Section 2 draft pasted into /capstone/security-posture.md; bucket lists and routing rule are filled.
The future-self question has an honest answer — not an aspirational one.

This worksheet accompanies Lesson 9.4 of AI Architect Academy. The three-bucket taxonomy (public / personal / sensitive), the “classify up when in doubt” rule, and the sensitive-goes-local commitment are concept. Specific local-model commands (Ollama pull, LM Studio loader, endpoint URLs) live in /recipe-book/routing-sensitive-data-to-a-local-model.md and are recipe.