DATA-001 – BASE-005

Data engineering

0 of 15 done · 0 due for review.

DATA-001Choose the required fields for this document contract: tickets with text, label, author, and optional attachments.construct · what-counts-as-training-datanew
DATA-002Two rows share identical text and label. Which duplicate survives under first-occurrence-wins, and why does that rule matter?predict · what-counts-as-training-datanew
DATA-003A row has text: 12345 (a number). Repair this malformed record WITHOUT hiding the error.debug · what-counts-as-training-datanew
DATA-004Separate these three values: a missing label (absent), an unknown label ('unknown'), and an invalid label ('xqzz'). Which belongs in training?compare · what-counts-as-training-datanew
DATA-005Write the rejection reason for a leaked label: a row already used in the held-out set. Make the reason distinct from all other rejections.implement · what-counts-as-training-datanew
DATA-006Predict the result of this temporal split: tickets sorted by creation date, first 80% train, last 20% test. What risks does it carry?predict · split-the-evidencenew
DATA-007Author 'chen' appears in both train and test. Find the group that leaks and state the harm.debug · split-the-evidencenew
DATA-008Design a dataset version identifier: the same source rows cleaned twice must produce the same id, and a changed row must produce a different one.construct · reproducibility-is-part-of-correctnessnew
DATA-009An ingestion job failed after row 400 of 1,000. Recover it from its checkpoint without reprocessing the first 400 rows.operate · reproducibility-is-part-of-correctnessnew
DATA-010A retry resends the same ingestion batch after a timeout. Is this retry idempotent? Decide and justify.predict · serve-the-model-safelynew
BASE-001Build a majority-class baseline for these labels: 90 question, 6 bug, 4 feature. What does it predict for every row, and what is its accuracy?implement · the-baseline-earns-the-right-to-improvenew
BASE-002Build a keyword baseline for bug detection: flag rows containing 'crash', 'error', or 'broken'. State its blind spot.construct · the-baseline-earns-the-right-to-improvenew
BASE-003For a ranking problem (best ticket answers first), choose a baseline and predict how it performs: popularity, recency, or random.predict · ranking-is-a-separate-responsibilitynew
BASE-004Compare this model and baseline: model 0.91, baseline 0.90, same split. Is this improvement meaningful?compare · build-a-better-baseline-honestlynew
BASE-005A random baseline gets 0.5 AUC. Explain to a teammate why this random baseline is still useful, not useless.communicate · the-baseline-earns-the-right-to-improvenew

Opening…