Teaching revision map · 2026-10-14.3
This map describes implemented instruction and learner activities, not measured learning gains.
P1 · Capstone difficulty gap
Module 10 now begins with three guided checkpoints: strict structural versus value failure; exact duplicate extras versus normalized-ID conflicts; and field validation plus source-index-aligned accumulation of multiple reasons. Four runnable toy examples use records different from the assessed workshop dataset. Each checkpoint gives expected intermediate masks/labels or records, explains why they differ, then asks for a changed-input prediction whose answer is separate in BRIDGE_SOLUTIONS.md. The policy is an acceptance table rather than one dense paragraph. The local and browser starters supply ingestion/orchestration and leave five explicitly contracted data-decision stages to the learner. A separate full-schema stage_checkpoint.csv is embedded in both starters with CHECKPOINT_STAGE 1–5 preview control, so the smaller conceptual tables are never mistakenly passed into the five-field stage interface. The lesson gives intermediate masks, reasons, evidence and totals for that fixture.
P1 · Inspection as a performed skill
Module 8 adds examples/08_inspection.py: shape, head, types, source missing markers, empty strings, post-conversion nonmissing count, individual rule failures, combined/inverse masks and gapped-label versus positional selection. It explains why empty source text differs from converted missingness. exercises/08_inspection.py asks for three observations with next actions on a new fixture before cleanup; the separate answers and solution code check accepted source labels and a valid zero duration.
P1 · Independent diagnosis
Module 6 no longer reveals the accumulator diagnosis before 6A. The starter gives the contract and symptom; exercises/06_diagnostic_worksheet.md requires expected/actual evidence, a reduced reproducer, a state trace, a hypothesis, one change and regression cases. A different runnable conversion failure and annotated traceback model how to locate the operation, bad value and upstream assignment. The diagnosis remains in separate answers.
P1 · Learner assessment and scoring
check_capstone.py loads exercises/10_capstone.py and passes that function to learner_capstone_checks.check_submission. The browser starter embeds the same checking function. Nine public scenarios exercise a fresh normal fixture, changed data, duplicates, conflicts, preserved multi-error evidence, empty input, structural failures, precision/date limits and zero price. Test code proves the checker calls its supplied function, rejects constant answers and detects modified evidence. Module 10 defines 0–4 anchors for each rubric criterion and a borderline example. Completion requires public checks, a satisfactory human changed-input explanation, threshold score and no critical failure.
P2 · Changed inputs and syntax bridges
The syllabus and Module 2 pair each changed input with a contract-derived expectation while preserving the original case. Module 4 adds an explicit local-parameter/caller-name trace and defines len/sum; negative errors are first run separately. Module 6 explains empty-list truthiness and relative/absolute tolerances, and its exit check now asks about an untested learner input. Module 7 contrasts list repetition with elementwise multiplication and adds non-square shape/value plus allclose transfer practice. Module 8 demonstrates mask inversion and index alignment. Module 9 explains zip, an expanded cents conversion loop, data-dependent chart headroom, empty/default behavior and two-decimal display formatting. Module 10 introduces the exact set, regex, .items(), Series-index and masked-assignment subset it requires.
P2 · Match the assessment verbs and voice
Module 3 adds container choice, retrieval/slicing/unpacking and a demonstrated nested-copy limitation. Module 9 adds independent chart-selection scenarios and a high/zero/empty output experiment. Readiness is explicitly a light placement screen supplemented by CSV export and diagnostic evidence. Source links now name useful topics. Full answers remain separate; the course's source-preservation and incomplete-subtotal safeguards remain intact.
Evidence and limits
Current local verification: 79 tests, 33 reference scripts, 62 syntax parses, unchanged dataset hashes. No claim of learner validation, live website integration or publication. Final browser/mobile/accessibility and beginner pilot checks remain necessary.