Corpus completion: the ten missing constructs, a known-failures section, and the capture script for the two verbatim pages

Phase 2 of QuantEcon/project-theme-parity#2. Brief revised 2026-09-17 after review (mmcky): synthetic snippets as the rule; the two verbatim pages stay; no further lecture captures unless the coverage join asks for one. Revised again the same day: the coverage join itself moves to the project's pass (bin/coverage in project-theme-parity), because it needs the private lecture inventory as well as this public corpus, and coverage is a property of an inventory version × a corpus version — a pass, not a repository.

Where the corpus stands (measured 2026-09-17)

Running bin/feature-inventory (workspace-lectures) over this repository's lectures/ and joining it to the 2026-09-17 matrix: 130 of the 140 inventoried constructs are present. Absent:

Decisions that shape this item

• Synthetic by default (2026-09-17). Snippets are cheap, stable and provenance-tagged; verbatim lectures add execution time, data dependencies and drift. The two verbatim pages already here (lp_intro.md, troubleshooting.md) stay as the realism fixtures for the visual review and for interaction effects — numbering across many targets, a {numref} far from its figure, a bibliography with real citation density (references.md needs a lecture that cites to test BibTeX rendering). Nothing further is captured unless the project's coverage report shows a construct a snippet cannot express.
• Execution is measured too. The corpus must exercise how each stack renders execution outcomes, not only markup: errors, halts, long outputs, skipped cells. GPU-only cells are out of scope — that is the execution backend, not the theme.
• Real material for figures and files: images and static files referenced by snippets are copied from the lecture repositories' _static/lecture_specific/<slug>/ with a provenance line (repo, path, commit); datasets load from data-lectures URLs exactly as the lectures do.
• Corpus scope: the eight manifest repositories (QuantEcon/project-theme-parity#5).
• Coverage is computed by the project, not here: bin/coverage in project-theme-parity runs the inventory scanner over this repository at a recorded commit, joins it to the lecture inventory, writes passes/<date>/coverage.md, and fails the pass when a construct with lecture uses is absent. This README links to the latest coverage report rather than carrying a COVERAGE.md.

Work

1. Ten missing constructs as snippets, each with a provenance comment naming the lecture it is adapted from: the label: convention on a code cell and a figure (from lecture-wasm), {any}, no-execute, align, equation*, gathered, <font>, <style>.
2. A known-failures section at the end of code_cells.md: a raises-exception cell, a deliberate NameError without the tag (both stacks' halt behaviour), an output longer than the scroll cap, a skip-execution cell, a cell writing to stderr, a cell producing a wide DataFrame. Last on the page so a halt cannot hide other surfaces, as the :load: demonstration already is.
3. bin/capture for the two verbatim pages only: re-copy lp_intro.md and troubleshooting.md from lecture-python-intro at a named commit, record (repo, path, commit, date) in CAPTURES.yml, and replace the hand-written provenance comments with a generated header. The script fails loudly on a referenced file it cannot copy rather than editing the lecture.
4. Keep the Sphinx build clean under -W; update _toc.yml; re-sync test-lecture-theme-mystmd with bin/sync-from-sphinx; link the README to the project's latest coverage report.

Exit

All ten constructs and the known-failures section are on pages with provenance, both builds are green, the mystmd repository is re-derived, and the next pass in project-theme-parity reports 140 of 140 present.

Start by running bin/feature-inventory against lectures/ and reading code_cells.md, the existing verbatim pages, and their provenance. Add the ten missing constructs and known-failures section, then use bin/capture for lp_intro.md and troubleshooting.md with CAPTURES.yml and update _toc.yml and the README. Verify clean Sphinx builds, re-sync with bin/sync-from-sphinx, and confirm the next project-theme-parity coverage pass reports 140 of 140 constructs.