HOMEY KNOWLEDGE COLLECTION
Dataset explorer
Source pages, arranged into grounded training examples.
Looking for the scraped English lessons? View the processed English training examples and their source evidence.
Collected examples 0
Generated drafts include source evidence. Approval requires a human review.
| QUESTION & ANSWER | STRUCTURED FACT | SOURCE / SPLIT | REVIEW | Inspect |
|---|
Waiting for the first examples
As pages are processed, their questions, answers, and evidence will appear here.
Splits are assigned by Wikipedia article, keeping excerpts from the same article together. Related facts across different articles still need a leakage review before evaluation.
Source library 0
Article revisions and raw responses are retained for traceability.
| PAGE | COVERAGE | EXCERPTS | EXAMPLES | STATUS |
|---|
No pages yet
Add Wikipedia article URLs or titles to start collecting.
Full-page expansion · live
Loading the expanded collection…
Each section can produce multiple batches. Source checks, duplicate filtering and a second LLM review precede storage. These are drafts, not verified ground truth. “Scanned” means the model twice reported no new items; it does not prove exhaustive coverage.
| INSTRUCTION | INPUT → ANSWER | SOURCE | REVIEW |
|---|
Full-page queue and extraction history
| PAGE | SECTIONS SCANNED | NEW DRAFTS | NEEDS REVIEW |
|---|
Original pilot · retained for comparison
The 24 original examples below are separate from the new full-page collection.
Processed training examples 0
The model's instruction, input and expected answer. Examples remain drafts after the assistant's check.
| INSTRUCTION | INPUT → ANSWER | SKILL | SOURCE | Inspect |
|---|
Waiting for the first exercise
Each page is checked before the next one is processed.
Page-by-page review 0 / 23
Loading the teaching batch…
| PAGE | INPUTS PROCESSED | PAGE CHECK | REVIEW NOTE |
|---|
This batch processes selected teaching excerpts from 23 downloaded pages. It is not exhaustive extraction of every article. Assistant checks are recorded separately from human approval; no model weights are trained here.
Recent activity
Live collection events, validation failures, and queue changes.