17 lines
1.8 KiB
Markdown
17 lines
1.8 KiB
Markdown
{n} research agents have independently determined the blocks of the topic "{topic}". Exactly identical titles have already been merged; the number in parentheses says how many research passes name the block. Consolidate the list.
|
|
|
|
{entries}
|
|
|
|
Rules:
|
|
- Recognize the SAME concepts under different titles and merge them into one block. The mention counts of the merged entries add up (each research pass counts a concept only once).
|
|
- A block solves EXACTLY ONE PROBLEM. Entries that are variants of the same solution are combined into ONE block (right: one block `<input>` for all types, one block "Modalverben" for all modal verbs; wrong: one entry per input type or per verb, but also collective entries that mix several problems).
|
|
- A block is ATOMIC: exactly one idea, complete in itself. Test: you can remove nothing without making it incomplete — and nothing is missing to understand it.
|
|
- CONSOLIDATE the granularity: a block is a LEARNING UNIT, not a dictionary entry. If the research passes deliver dozens of micro-entries of the same kind (one CSS property, one verb, one gesture per entry), group them by problem (right: "Flexbox-Ausrichtung" instead of six entries for justify-content, align-items, …). More than ~150 blocks is almost always a granularity problem — then check specifically for such series.
|
|
- Then split into two lists: blocks that (after merging) are named by AT LEAST TWO research passes → `blocks`. Named only once or doubtful on the merits → `rest`. Discard only what is obviously fabricated.
|
|
- Drop the sources. Title and short description (max. ~12 words) in GERMAN (code identifiers stay original). Every title must be UNIQUE.
|
|
|
|
Write ONLY the JSON file to: {out_path}
|
|
|
|
Format (each entry a string "Title — Kurzbeschreibung"; no other text in the file):
|
|
{{"blocks": ["Title — Kurzbeschreibung"], "rest": ["Title — Kurzbeschreibung"]}}
|