Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@ROADMAP.md`:
- Around line 34-45: Remove the shipped issue references from ROADMAP.md by
deleting mentions of `#463` and `#480` (and any retrospective phrasing around them)
so the roadmap only lists future work; then add one-line entries for each
shipped item to the appropriate Stage 9 row in HISTORY.md (e.g., "Closed `#463` —
IO.sleep/time/stderr" and "Closed `#480` — prerequisite structural refactor")
ensuring the HISTORY.md entry format matches existing Stage 9 table rows and
that surrounding ROADMAP.md prose remains grammatically correct after removal.

In `@spec/12-runtime.md`:
- Around line 218-254: Update the count of IO host function bindings in section
12.4.1 to reflect the three new ops (IO.sleep, IO.time, IO.stderr) so the text
that currently reads "seven host function bindings" now says "ten host function
bindings" (or equivalent) and ensure any nearby summary/listing references match
the definitions of 12.4.1.1 through 12.4.1.10.

In `@TESTING.md`:
- Line 58: Update the IO coverage line in TESTING.md that references
`test_codegen.py` to include the new IO operations by adding "sleep, time,
stderr" to the existing list of covered IO ops; find the table row mentioning
`test_codegen.py` and append these three symbols to the IO section so the
documentation matches the current codebase.

In `@tests/conformance/manifest.json`:
- Around line 565-577: Add a non-runtime conformance case for IO.sleep by adding
a new manifest entry similar to "ch07_io_time_stderr" but with a unique id
(e.g., "ch07_io_sleep"), file "ch07_io_sleep.vera", chapter 7, title mentioning
IO.sleep, level set to "check" or "verify", spec_ref "Section 7.2", and features
including "io_effect" and "io_sleep"; also create the corresponding
tests/conformance/ch07_io_sleep.vera file (simple deterministic check for
IO.sleep behavior) so the manifest references an actual test program.

In `@tests/test_codegen.py`:
- Around line 7186-7204: The test test_io_stderr_default_not_captured currently
only asserts exec_result.stderr == "" but needs to also assert that stdout
remains empty to ensure no cross-stream leakage; update the
test_io_stderr_default_not_captured function to include an explicit assertion on
exec_result.stdout (e.g., assert exec_result.stdout == "") immediately after the
stderr assertion so both streams are validated when capture_stderr is not
enabled.
- Around line 7127-7128: The test uses a hard-coded epoch threshold to validate
IO.time which can fail on skewed/frozen clocks; change the assertion to capture
host time immediately before and/or after running the command and assert
int(exec_result.stdout) falls between those captured host timestamps (e.g.,
record host_time_before and host_time_after around the execution and assert
host_time_before <= int(exec_result.stdout) <= host_time_after) so the check
compares against the actual host clock rather than a fixed epoch constant.

In `@vera/browser/harness.mjs`:
- Around line 89-92: Update the example/documentation comment that describes the
emitted JSON to reflect the new shape which now includes stderr; locate the
comment next to the output object construction (variable output) where
getStdout(), getStderr(), and getState() are used and modify the example JSON to
include a stderr field alongside stdout and state (and remove any stale fields),
so the documented schema matches the actual emitted object.

In `@vera/codegen/api.py`:
- Around line 104-108: The ExecuteResult dataclass field order was changed
causing the public positional constructor to shift; restore the original
positional shape by reordering the fields so they are defined as value, stdout,
state, exit_code (and keep stderr as a defaulted field after state or keep it as
a keyword-only with a default) — update the ExecuteResult declaration (the
dataclass and its field definitions: value, stdout, state, stderr, exit_code) so
state appears before stderr and stderr retains its default "" to avoid breaking
external positional call sites.

---

Outside diff comments:
In `@FAQ.md`:
- Around line 206-210: Update the IO operations bullet to reflect the current
API by adding the new IO functions IO.sleep, IO.time, and IO.stderr to the
existing list (which currently mentions print, read_line, read_file, write_file,
args, exit, get_env); locate the FAQ.md bullet that enumerates IO operations and
expand it to include IO.sleep, IO.time, and IO.stderr so the documentation
matches the codebase.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@spec/12-runtime.md`:
- Around line 572-574: Update the browser public API snippet in Section 12.9.2
so it matches the runtime and Section 12.9.3 by adding the stderr helpers to the
exported/imported names; specifically include getStderr alongside getStdout,
getState, and resetState (referencing the getStderr symbol) so the example shows
the same public API surface as documented elsewhere. Ensure the snippet
imports/exports getStderr in the same style as the other helpers and that any
example usage reflects reading the stderr capture via getStderr().
- Around line 42-44: Section 12.2.3 currently overstates that any IO operation
other than `print` and `exit` requires `$alloc`; update the paragraph to state
that only IO imports which allocate or return heap data require `$alloc`, and
explicitly note that `vera.sleep`, `vera.time`, and `vera.stderr` (which map to
`IO.sleep`, `IO.time`, `IO.stderr`) do not allocate or return heap data and
therefore do not require `$alloc`; reference the `$alloc` symbol and the IO
import names (`vera.sleep`, `vera.time`, `vera.stderr`, `IO.sleep`, `IO.time`,
`IO.stderr`) and adjust the example/contract text in section 12.2.3 accordingly
so the claim matches these cases.

In `@TESTING.md`:
- Line 9: Summary: The aggregate test count in TESTING.md is inconsistent with
the component counts; update either the total "3,356" or the component numbers
"3,344 passed, 11 skipped" so they sum correctly. Fix: open TESTING.md, locate
the table row containing "3,356 across 28 files (~34,000 lines of test code;
3,344 passed, 11 skipped)" and adjust the numeric values so the sum matches
(change the total to 3,355 or modify the passed/skipped counts to sum to 3,356),
and then commit the corrected markdown.
- Around line 109-110: The TESTING.md prose mismatches the table: after raising
the `verify` count to 4 the paragraph still states “Three programs” and lists
only three files; update the paragraph under the `verify`/`run` table to reflect
four programs by either adding the missing fourth program to the list or
changing the sentence to the correct count, and ensure the listed filenames
match the actual verify-level test files referenced by the `verify` table entry.
- Line 76: The table row for `test_conformance.py` lists 375 tests but the "Via
pytest" example still says `365 tests`; update the "Via pytest" example string
(the text containing "Via pytest" and "365 tests") to `375 tests` so both
references match, and scan TESTING.md for any other occurrences of `365 tests`
or the conformance total and make them consistent with `375` (reference:
`test_conformance.py` and the "Via pytest" example text).

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@CONTRIBUTING.md`:
- Line 88: The sentence claiming "23 run on every commit after `pre-commit
install`" is inaccurate because many commit-stage hooks are file/type-gated in
`.pre-commit-config.yaml`; update the wording to say "23 commit-stage hooks" (or
equivalent) and clarify that some of those run only when matching files are
staged, keep the note that the two push-stage hooks are
`check-changelog-updated` and `uv-lock-check`, and ensure the Markdown review
rule `**/*.md` remains consistent with the actual gating behavior described.

In `@TESTING.md`:
- Line 9: The summary line under the "Tests" table showing "12 skipped"
conflicts with the "Skipped-tests" breakdown that lists and enumerates 11 cases;
update the markdown so both places report the same skipped count: either add the
missing skipped case to the "Skipped-tests" section or change the table summary
from "12 skipped" to "11 skipped" (ensure the table row labeled "**Tests**" and
the "Skipped-tests" enumeration match exactly), and run a quick grep of the
enumerated case headings to confirm all skipped items are present and correctly
numbered.
- Line 377: The statement claiming "23 run on every commit" is inaccurate
because many hooks use trigger filters (`files`/`types`) and only run when those
filters match; update the sentence that starts "Every push is checked by 25
hooks..." to say that there are 25 configured hooks (23 that are configured to
run on commits but are subject to per-hook `files`/`types` filters, and
2—`check-changelog-updated`, `uv-lock-check`—that run only at push time with
`--hook-type pre-push`), and make the same clarification where the numeric claim
appears again (the other occurrence referencing 406) so the README reflects
trigger-based execution rather than "every commit".

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@scripts/check_doc_counts.py`:
- Line 199: The CONTRIBUTING.md hook-count regex is stricter than the TESTING.md
one; update the parsing so both accept the optional "configured " token. Replace
the existing literal pattern used for CONTRIBUTING.md (the one matching "checked
by (\d+) hooks") with the same flexible pattern r"checked by (\d+) (?:configured
)?hooks" (or factor both into a shared constant) so counts like "checked by 25
configured hooks" are recognized; ensure the change is applied where
CONTRIBUTING.md is parsed in scripts/check_doc_counts.py.