Sharpen Module 1 Seam 1 so the example forces a real multi-file change #36

New Issue

2026-06-22T14:23:55-04:00

claude commented

2026-06-22 14:23:55 -04:00

Problem

Part B Seam 1's purpose is to make the learner feel that one-file copy-paste breaks across files, but a clear command is trivially implementable entirely inside cli.py (e.g. save(TaskList())), so the AI seeing only cli.py produces working code and python cli.py clear succeeds — the seam doesn't bite. (The AI might instead guess tlist.clear() and crash, making the demo non-deterministic.) The text even hedges ("if a clean implementation belonged there").

Evidence

modules/01-the-copy-paste-problem/README.md lines 170-175: "Add a clear command that removes all tasks." … "it couldn't see tasks.py, so if a clean implementation belonged there, it had to guess or cram it into the file it could see."

Why it matters

This is the load-bearing "feel the seam on purpose" lab in a LOCKED reference exemplar that later modules emulate; the seam can silently succeed instead of biting (unlike Seams 2 and 3).

Proposed change

Use a task whose clean implementation lives in tasks.py, e.g. "make list hide completed tasks." The clean fix edits render() in tasks.py (a pending() helper the AI can't see already exists), so an AI given only cli.py must reach into tlist.tasks and re-derive the [x]/[ ] box+index formatting — visible duplication, a sharper seam. Module 2 reuses the app with an independent task and never references the Module 1 clear command, so the chain is intact.

Acceptance criteria

Seam 1's task's clean implementation lives in the unseen tasks.py.
Following the lab produces visible duplication/guessing, not silent success.
No downstream module depends on the old clear Seam-1 task.

Affected files

modules/01-the-copy-paste-problem/README.md

References

Source finding F18 (realVotes 3/3). Touches locked exemplar Module 1 — confirm with course owner; preserve voice/structure.

Filed from an adversarial multi-agent course review (217 raw findings → 54 adversarially-verified survivors). Scoped for manual review; intentionally not auto-assigned to an agent.

## Problem Part B Seam 1's purpose is to make the learner feel that one-file copy-paste breaks across files, but a `clear` command is trivially implementable entirely inside `cli.py` (e.g. `save(TaskList())`), so the AI seeing only `cli.py` produces working code and `python cli.py clear` succeeds — the seam doesn't bite. (The AI might instead guess `tlist.clear()` and crash, making the demo non-deterministic.) The text even hedges ("if a clean implementation belonged there"). ## Evidence `modules/01-the-copy-paste-problem/README.md` lines 170-175: "Add a `clear` command that removes all tasks." … "it couldn't see `tasks.py`, so if a clean implementation belonged there, it had to guess or cram it into the file it could see." ## Why it matters This is the load-bearing "feel the seam on purpose" lab in a LOCKED reference exemplar that later modules emulate; the seam can silently succeed instead of biting (unlike Seams 2 and 3). ## Proposed change Use a task whose clean implementation lives in `tasks.py`, e.g. "make `list` hide completed tasks." The clean fix edits `render()` in `tasks.py` (a `pending()` helper the AI can't see already exists), so an AI given only `cli.py` must reach into `tlist.tasks` and re-derive the `[x]`/`[ ]` box+index formatting — visible duplication, a sharper seam. Module 2 reuses the app with an independent task and never references the Module 1 `clear` command, so the chain is intact. ## Acceptance criteria - [ ] Seam 1's task's clean implementation lives in the unseen `tasks.py`. - [ ] Following the lab produces visible duplication/guessing, not silent success. - [ ] No downstream module depends on the old `clear` Seam-1 task. ## Affected files - `modules/01-the-copy-paste-problem/README.md` ## References Source finding F18 (realVotes 3/3). Touches locked exemplar Module 1 — confirm with course owner; preserve voice/structure. --- *Filed from an adversarial multi-agent course review (217 raw findings → 54 adversarially-verified survivors). Scoped for manual review; intentionally not auto-assigned to an agent.*