A handout into a revision outline that shows which sections it covers
Measured on 2026-11-14 — numbers from one real run
The run log here comes from an actual run, not an estimate. Where token usage was not captured it says so rather than guessing.
When to use
Exams are coming and the handout runs to dozens of pages. You want it compressed into an outline, but the real fear is a missing half-chapter that the outline hides by looking complete.
Preparation
- Host version
- 0.1.2-alpha.5
- Model
- To fill
- Mode / permissions
- --profile headless / workspace-write / Windows 11 / node 22.22.2
- Plugins required
- None (add name & version after testing if used)
- Input materials
- A 2,144-character synthetic handout on pu-erh storage and ageing with seven top-level sections, including a planted inconsistency: the summary in 4.7 claims three household storage taboos while 4.4 states only two
Full instruction
Read handout.md (the full handout) and produce a revision outline at outline.md.
[Output structure]
1. Coverage table: one row per top-level chapter of the handout, marked 'covered / partially covered / not covered', with the outline item numbers that correspond. This table must list every chapter; none may be omitted.
2. Outline body: organised by chapter, one line per point. Each line: point | one-sentence explanation | importance (must-know / often-tested / awareness) | source (chapter-section)
3. Gap list: material in the handout that did not make the outline, and why (too long / redundant / peripheral)
[Hard constraints]
1. The coverage table must include every top-level chapter. 'Not covered' is allowed but needs a reason — never delete a chapter from the table, because then the table lies.
2. Importance must cite evidence (for example 'this section has three worked examples', 'bolded in the source'), never a feeling.
3. Every point carries a source (chapter-section) that resolves back to the original.
4. Add nothing the handout does not contain. Things you think are examinable but absent go in a separate 'not in the handout but worth revising' section.
5. Never delete an entire 'awareness' section — downgrading is fine, disappearing is not.
Report at the end: how many top-level chapters, covered/partial/uncovered counts, how many points, how many gaps. The coverage table row count must equal the chapter count.Run log
| Item | Value | Note |
|---|---|---|
| Time spent | 1 minute 11 seconds | 2026-11-14 |
| Result | Coverage table 7 rows against 7 sections; 24 points (9 must-know, 12 often-tested, 3 awareness); 2 gaps; artifact 10,350 bytes; no leftovers | |
| In / out tokens | Not captured | |
| Leftovers | None; the workspace holds only handout.md and outline.md. |
What the run showed
- The equation 'table rows equal chapter count' held: seven rows against seven sections, and it added 'verified by script: 24 KP- rows, 7 coverage rows, tier counts agree with the body'.
- It caught the planted inconsistency without patching it: flagging that 4.7 claims three household taboos while 4.4 states only two, and stating outright that it would not invent a third because that would be fabrication — listing the candidates (the no-cosmetics rule in 4.1, avoiding direct airflow) but marking them unverified.
- Importance was not assigned by feel: the handout has no worked examples, no bold, and no exam markers, and it said so before defining three structural signals from the handout itself (named in the summary, a standalone conclusion sentence, a quantified figure), citing which signal each row used. That is far more honest than forcing an 'examples mean must-know' rule.
- All three 'awareness' items stayed in the body — downgraded, never deleted, per the instruction that downgrading is fine and disappearing is not.
- It added its own caveat: 'must-know here only means the most exam-like tier within this handout's own structure, not a prediction about the real paper'. Self-limitation like that is rare and worth copying.
Two mechanisms, neither optional: the coverage equation prevents dropped chapters, and the gap list prevents 'looks complete but is missing things'. An outline without them gives you no way to see what was lost.
Output
After a successful run, list files and screenshots here, and note whether originals changed.
Acceptance checklist
Tap to tick each item off as you verify.
- Coverage table rows equal top-level chapters — use this equation to catch omissions
- No chapter was deleted from the table; 'not covered' rows carry reasons
- Importance cites evidence, not a feeling
- Every point resolves to a chapter-section in the source
- 'Not in handout' is strictly separate from the outline
Failures & boundaries
- The equation 'table rows equal chapter count' is the only omission guard here. Without it you get an outline that looks complete and silently lacks two chapters
- Importance tends to come back all 'must-know'. The tell is the evidence column filling with 'key content' and 'commonly examined' — that means going back to how the source was read
- An outline is inherently lower density than the handout, so loss is certain. The gap list tells you more about usability than the outline itself