Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@vera/README.md` around lines 115 - 117, The markdown rows include inline code
spans with leading/trailing spaces (e.g. "`  core.py`", "`  cache.py`", "` 
session.py`") which trigger MD038; remove the inner padding so the spans are
backticked without spaces (e.g. change "`  core.py`" to "`core.py`") and apply
the same change to the other affected rows (including the additional occurrences
mentioned around lines 119-123) to satisfy markdownlint.

Learnt from: aallan
Repo: aallan/vera PR: 720
File: vera/README.md:115-117
Timestamp: 2026-06-11T00:03:34.481Z
Learning: In `vera/README.md`, the Module Map table uses leading-space code spans (e.g. `` `  core.py` ``) as the established visual convention for sub-module rows (indented entries under a package row). This is intentional and consistent with the pre-existing `checker/` section. Do not flag these as MD038 violations; there is no markdownlint gate in the repo's CI or pre-commit configuration, and changing only new rows would break visual consistency with existing ones.

Learnt from: aallan
Repo: aallan/vera PR: 571
File: HISTORY.md:301-301
Timestamp: 2026-05-05T17:09:21.991Z
Learning: In the aallan/vera repo documentation, compute the canonical test count using pytest collection, not by line-counting source code. Use `pytest tests/ --collect-only -q` (or `python scripts/check_doc_counts.py`) as the source of truth. Do not verify “number of tests” by counting `def test_` / `class Test` lines with ripgrep (`rg`) or by shell-based line counts, because parametrized/fixture-driven tests can produce multiple collected test items from a single `def` line and this will undercount.

Learnt from: aallan
Repo: aallan/vera PR: 577
File: HISTORY.md:301-311
Timestamp: 2026-05-06T08:33:56.675Z
Learning: Do not trust pytest collection counts from the CodeRabbit sandbox when reporting test counts in Markdown docs (e.g., README.md, HISTORY.md, TESTING.md, ROADMAP.md). The sandbox lacks the full Python dependencies and yields inaccurate counts. Verify counts in the real development environment by running 'pytest tests/ --collect-only -q' or by using 'scripts/check_doc_counts.py', and base the documented numbers on that output. Apply this guidance broadly to Markdown docs in the repo, following the canonical approach referenced by learning 3b066ed3-0d66-41bd-a058-ccdf04052082.

Learnt from: aallan
Repo: aallan/vera PR: 613
File: ROADMAP.md:7-7
Timestamp: 2026-05-07T20:05:33.932Z
Learning: When reviewing Markdown documentation in aallan/vera that involves numeric count/summary verification (including tests, conformance programs, examples, and spec chapters), treat `python scripts/check_doc_counts.py` as the single authoritative source of truth. Do not use cross-references against `README.md` or other Markdown files to validate counts, since they may be temporarily stale during PR series. If you see a count discrepancy detected via `rg`/`grep` against `README.md`, treat it as non-actionable; only fail/flag the issue when `check_doc_counts.py` reports an error.

Learnt from: aallan
Repo: aallan/vera PR: 662
File: HISTORY.md:292-293
Timestamp: 2026-05-12T08:14:28.532Z
Learning: In aallan/vera, ensure the “tagged release” count stays perfectly in sync across exactly two locations per release: (1) the roll-up footer total in HISTORY.md (the “N tagged releases” line) and (2) the “N releases” figure in README.md (around line ~184 in the project-status sentence). ROADMAP.md intentionally does not include this counter. During PR review, flag changes that update only one of these two designated locations without updating the other to the same tagged-release count.

Learnt from: aallan
Repo: aallan/vera PR: 689
File: TESTING.md:0-0
Timestamp: 2026-05-19T11:05:07.302Z
Learning: When reviewing Vera documentation Markdown (e.g., TESTING.md), don’t update only the instance shown in the diff hunk for stale numeric count changes. Instead, treat the old count as a value that must be replaced everywhere in the same file: search the entire file for all literal occurrences of the old number (including table cells, overview/feature-coverage sections, round-trip sections, pre-commit/CI tables, and any other descriptive text) and ensure every occurrence is updated consistently as part of the same change. Use a whole-file search (e.g., grep-style patterns such as `34 example|all 34|All 34` or an equivalent literal-number match) to confirm no old-count references remain before commenting/approving.

Learnt from: aallan
Repo: aallan/vera PR: 689
File: CHANGELOG.md:0-0
Timestamp: 2026-05-19T12:34:26.776Z
Learning: When documenting `IO.read_char` behavior in the aallan/vera docs for Unix TTY inputs, explicitly state that it uses termios *cbreak* mode via `tty.setcbreak()` (so `ISIG` remains enabled and Ctrl-C works), and that it should *not* use raw mode. Also document that in the Unix TTY branch, Ctrl-D maps to EOF, while the non-TTY path reads a single byte using `sys.stdin.read(1)`.

Learnt from: aallan
Repo: aallan/vera PR: 697
File: ROADMAP.md:0-0
Timestamp: 2026-05-26T16:23:53.102Z
Learning: In Markdown documentation, do not accept citation/attribution URLs as valid based solely on plausible formatting. For external attribution links (e.g., tweets/X posts, blog posts, arXiv papers, GitHub permalinks, etc.), verify that the linked content is publicly accessible and matches the claimed author/date/title. If the source cannot be confirmed as accessible/accurate, flag the link as “unverifiable” and recommend replacing it with a confirmed URL or removing the URL while keeping the attribution text (e.g., “Szymon Teżewski (jasisz1)’s 24 May 2026 tweet”). Rule of thumb: verify or omit—never fabricate or assume.