Lab 06. Agent topology, context and memory as recorded decisions
Part 2 · Scheduled 13 Dec 2026 · 2-hour lab
Upcoming course outline. This class has not been recorded as studied.
Lab brief
Add observation masking and summary compaction next to a no-policy baseline; add file memory with ADD/UPDATE/DELETE/NOOP and a progress file; run one long task per arm. After class: 5 long tasks x 3 arms x k = 3 on the LAN server; the ADR with a C4 diagram. Tag part2-ref.
What to build and measure
- Add observation masking and summary compaction next to a no-policy baseline.
- Add a file memory with ADD, UPDATE, DELETE and NOOP, and a progress file; unit tests with the scripted LM.
- Run one long task per arm in the room; 5 tasks x 3 arms x k = 3 after class.
- Write the ADR with a C4 diagram for the context and memory policy, citing one measurement (Part 2 deliverable); tag part2-ref.
Why this class matters
- Industry: An ADR with a number is how a team decides a policy; students practise the document and the measurement.
- Academic: Three arms, k = 3, success and tokens: a small experiment in the paper format of Part 4.
- Elsewhere: Students see a planted stale fact get replaced, which is what memory is for.
In-room pass check
UPDATE and DELETE unit tests pass with the scripted LM; a planted stale fact is replaced; a second session resumes from the progress file; each arm finishes one long task.
Required materials
- 5 long tasks from the staff (ready by 22 Nov).
- LAN server for the after-class runs; laptops are too slow.
- ADR template.
These are materials the draft assigns to the course team. They are not bundled with these notes. See the course guide for the test-first verification log and the deliverables for part boundaries.
Results ledger
[Measurements, test output, run-record paths, and verification log]
Sources
- Mem0 paper (arXiv 2504.19413)