[napetrov]

Part of #4.
Depends on #164: both append tests to tests/test_novelty_routing.py, so this rebases cleanly only after #164 merges.

Summary

novelty routing --queries-from-headings stops reading code-block lines as requests, and a run with no usable query says why.

 headings_as_queries(body)
+  _outside_fences(body)                  # drop ``` and ~~~ blocks first
   headings: ^#{2,3}[ \t]+(.+)$           # a bare "##" line no longer takes the next line
 cmd_novelty_routing
-  if not queries: exit "no usable queries"
+  zero scored queries:
+    write routing.json  {"status": "no_queries", "reason": ...}
+    exit 1, naming the reason and --queries
   rate lines
+    n_queries N given, ...  triage: self-derived
+    WARNING if fewer than 5 scored

Evidence

Nine new tests in tests/test_novelty_routing.py. Against intel/main (commands/novelty.py and novelty/routing.py reverted, rest of the branch kept), nine fail:

before: intel/main routing.py + novelty.py
  FAILED test_a_heading_inside_a_code_fence_is_not_a_query[```]
  FAILED test_a_heading_inside_a_code_fence_is_not_a_query[~~~]
  FAILED test_a_fence_closes_only_on_its_own_character_and_length
  FAILED test_a_bare_hash_line_does_not_take_the_next_line_as_its_heading
  FAILED test_headings_inside_code_only_refuse_the_run_and_say_why
  FAILED test_a_query_file_with_nothing_in_it_refuses_the_run
  FAILED test_queries_without_a_content_word_refuse_the_run
  FAILED test_every_rate_line_carries_its_query_count_and_triage_tag
  FAILED test_fewer_than_five_scored_queries_are_warned_about_on_stdout
  9 failed, 48 passed

after: this branch, full suite below

Local checks on the rebased head:

uv run ruff check .                              All checks passed!
uv run mypy dokimi                               Success: no issues found in 222 source files
uv run python -m dokimi.schema_check             exit 0
uv run python -m dokimi.config_check             exit 0
uv run pytest -n auto -q                         6951 passed, 22 skipped

Merge Danger

Door: two-way. The exit-code change is the only one that a script can see, and it is reversible by reverting the commit.

Blast Radius: low, limited to novelty routing.

• Exit code 1 on no scored query. A run whose queries are non-empty but all filtered out (no content word) used to exit 0 with null rates. It now exits 1 and writes "status": "no_queries" with a reason. Scripts that read the null rates as a result must check the exit code or status. This is the one change to existing behaviour.
• Result file on the failure path. The no-query case now writes routing.json before exiting. A batch run finds a record instead of nothing.
• Fenced headings are no longer queries. --queries-from-headings offers fewer, and different, self-derived queries on artifacts with code blocks. They remain triage only and stay stamped queries_self_derived.
• Stdout format. Each rate line now reads n_queries N given, ... and carries triage: self-derived when the queries came from headings. Line-parsers need updating.
• Warning, not failure, under five queries. A run that scores fewer than five queries prints a warning on stdout.
• Merge order. Merge after feat(novelty): usage and cost block for routing --model #164 (see the first line).

Checklist

• Tests added/updated and pytest passes
• ruff, mypy, and schema checks pass
• Docs updated: docs/skill-novelty.md (routing section)
• No ad-hoc run artifacts committed
• New code and test files carry the SPDX license header (no new code files; the changelog fragment follows the existing fragments, which carry none)

🤖 Generated with Claude Code

[antonwolfy]

Review at 52d90cc. The fence skipping works for the plain cases, and the new rate lines and the no_queries record behave as described. Four points are worth fixing before merge: one bug in the fence matcher, two risks (a model-row count that does not match the lexical one, and tests that miss the fence-closing rules), and one nit.

What I ran, on a Linux review host at 52d90cc:

• make ci-local: 20 passed, 8 failed, 7 skipped. The 8 failures are ruff, zizmor, pre-commit, bandit, mypy, deptry, mkdocs build and pip-audit, all exit 127: the tools are missing from my venv because make dev stops at the tools install on this base, which #179 fixed on main. Static checks, mypy and the strict docs build pass on GitHub's CI at this head. Tests: 6952 passed, 21 skipped; changed-lines coverage gate passed; schema-compat, wheel build and smoke, and gitleaks passed.
• Reproductions for the code findings, with the LLM transport mocked.
• Mutation runs on _outside_fences: 8 mutants, 3 of which pass the full test_novelty_routing.py.

Checked, no issues

• Fence closing: closers need the opener's character and at least its length; trailing spaces, CRLF and unclosed fences behave as CommonMark.
• Heading regex: ##\n foo bar no longer takes the next line; ##\tTabbed heading is accepted; #### stays excluded.
• Consumers: nothing in the tree parses the old over N of M queries line, no schema covers routing.json, and the one test that reads it (test_novelty_journey.py) uses lexical.n_queries, which is unchanged.
• No-query path: it exits before any model call, and it overwrites a stale status: ok file at --out.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.80%. Comparing base (5064dc2) to head (6125691).
⚠️ Report is 11 commits behind head on main.
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #168      +/-   ##
==========================================
+ Coverage   91.64%   91.80%   +0.15%     
==========================================
  Files         222      223       +1     
  Lines       27409    27638     +229     
  Branches     4508     4553      +45     
==========================================
+ Hits        25118    25372     +254     
+ Misses       1620     1604      -16     
+ Partials      671      662       -9     

Files with missing lines	Coverage Δ
dokimi/dokimi/commands/novelty.py	95.84% <100.00%> (+2.30%)	⬆️
dokimi/dokimi/novelty/routing.py	98.85% <100.00%> (+0.09%)	⬆️

... and 15 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[antonwolfy]

I pushed a commit per nit, up to 46a0897, which fix the nit from my review:

• 46a0897: dokimi/dokimi/commands/novelty.py:373: the warning now says 'query' for a count of one and 'queries' otherwise.

make ci-local at 46a0897: 20 passed, 8 failed (all exit 127, tools missing on host), 7 skipped.

Not approving yet. Still open: the bug and the two risks from my review.

[antonwolfy]

Review at fc0539e. The three points from my earlier review are fixed: the fence matcher follows CommonMark on the indent and the info string, the model row says n_queries N given, and the tests I asked for are there. One test gap is left before merge.

What I ran, on a Linux review host at fc0539e:

• make ci-local: 28 passed, 1 failed, 7 skipped. The failure is pip-audit on the lock files, which this PR does not touch. Tests: 7223 passed, 1 skipped; changed-lines coverage gate, schema-compat, docs build, wheel build and smoke, and gitleaks passed.
• Mutation runs on _FENCE, _outside_fences and the heading regex: 18 mutants; some pass the full test_novelty_routing.py. The ones that matter are on the inline comment.

Checked, no issues

• Fence rules: a tab indent, a 4-space indent, a backtick in a backtick opener's info string, a ~~~ line inside a backtick block, a ```js line inside a block, trailing spaces/tab and CRLF on the closer all behave as CommonMark; each of the first five now has a test that fails when the rule is removed.
• No-query path: all-fenced headings, a blank queries file, an empty JSON list and an all-stopword file each exit 1 with a reason and write status: no_queries; the model is not called.
• Consumers: the only reader of routing.json in the tree is test_novelty_journey.py, which uses lexical.n_queries; the added status key breaks nothing.
• Growth ratchet: novelty.py is 863 lines, matching the raised baseline, and cmd_novelty_routing complexity is lowered to 12.

[xaleryb]

• Ran: full suite, ruff, mypy strict and bandit at the head, a CommonMark differential on fenced documents, and a mutation pass over the new fence and output code.
• Open (F1, untested branch): "```\n## Hidden heading one\n ```\n## Real heading one\n" gives ["Real heading one"] at the head. If the closer check is changed to not line[len(fence.group(1)):].strip(), that becomes [] and every test still passes. Indented fences are how fences look inside list items. Also unpinned: "status": "ok", the triage: self-derived tag on the model row, the < 5 warning threshold, and a tab after ##.
• Pushed b6e3d7d (test-only): parametrized open/close indents 0-3 for backtick and tilde fences, the status == "ok" check, a model-row tag check under --queries-from-headings --model, a warning at 4 scored queries, and a ##\t heading. Each mutant above now turns a test red.
• Left for the author: resolve the three outdated inline threads, which are fixed at the head.

[xaleryb]

Approved at b6e3d7d. Every earlier point re-run at this head passes and no must-fix or mid item is open; full suite, ruff and mypy clean on Linux; CI green.

[antonwolfy]

Review at b6e3d7d. The commit since fc0539e adds test-only changes, and they close @xaleryb's gaps on indented closers, status: ok, the model-row tag and the warning threshold. One test gap from my earlier review is still open, so I am not approving yet.

What I ran, on a Linux review host at b6e3d7d:

• make ci-local: 28 passed, 1 failed, 7 skipped. The failure is pip-audit on the lock files, which this PR does not touch. Tests: 7257 passed, 1 skipped; changed-lines coverage gate, schema-compat, docs build, wheel build and smoke, and gitleaks passed.
• Two mutants run against the full tests/test_novelty_routing.py (98 tests): both pass.

Still open from my earlier review:

• dokimi/tests/test_novelty_routing.py:593: the minimum fence length and the heading separator are still unpinned. With _FENCE at {2,} for both characters, headings_as_queries("~~strike~~ text\n## Real heading one\n") returns [] instead of ['Real heading one'], and all 98 tests pass. With the heading regex at ^#{2,3}[ \t]*(.+)$, headings_as_queries("##hashtag word\n") returns ['hashtag word'] instead of [], and all 98 pass. The new ##\tTab heading words test does not help: it passes under [ \t]* too. The new indent tests use three-character fences only. Suggestion: add headings_as_queries("~~strike~~ text\n## Real heading one\n") == ["Real heading one"] and headings_as_queries("##hashtag word\n") == [].

Checked, no issues

• Zero-query path: an all-stopword file, a blank file and an all-fenced heading file each exit 1 with a reason and write status: no_queries.
• Rate line and warning: triage: self-derived shows on the rate line only for headings-derived queries; the warning fires below five scored queries.
• Docs and changelog: the docs tests, the growth ratchet and ruff pass on the changed files.

[antonwolfy]

LGTM. The two tests added in 6125691 pin both points from my earlier thread, and nothing else changed since b6e3d7d.

What I ran, on a Linux review host at 6125691:

• make ci-local: 28 passed, 1 failed, 7 skipped. The failure is pip-audit on the lock files, which this PR does not touch. Tests: 7261 passed, 1 skipped; changed-lines coverage gate, schema-compat, docs build, wheel build and smoke, and gitleaks passed.
• Mutation runs against the full tests/test_novelty_routing.py (102 tests): _FENCE at {2,} for backtick, for tilde and for both, and the heading regex at [ \t]*, each turn a test red. Seven further mutants on the indent bounds, the closer length, the closer character, trailing text on a closer and the backtick info-string lookahead are each killed by one test.

Checked, no issues

• Fence minimum: ~~strike~~ text, ``x`` text and ``x text keep the heading after them.
• Heading separator: ##hashtag word yields no query.
• Docs and changelog: exit code 1, status: no_queries with a reason, the n_queries N of M given / n_queries N given rate lines, the triage: self-derived tag and the under-five warning match the code.