todo2code

External corpus baseline

Runtime: todo2code 0.5.0 at 5f5ae5938ab77dcce474ba7abbd23686072776ec.

Each source was checked out as a detached, tracked-only worktree at the commit recorded below. Runs were offline and deterministic: tracked TASK.md, TODO.md and CHANGELOG.md were selected when present, documents were limited to README.md and docs/**/*.md, communication and task synthesis were disabled, and neither extraction nor summary used an LLM.

Repository	Commit	Time	Records	Relations	Topics aligned/all	Impl.	Plan	Docs	Diagnostics (I/W/R/B)	Warnings
semcod/code2llm	b297d60	18 s	16,899	41,747	107/628	59.4%	43.7%	31.4%	912/2,377/1,411/0	9
semcod/domd	b6c5ad2	5 s	10,611	7,470	9/241	11.8%	5.4%	5.4%	616/1,388/105/0	0
semcod/pactfix	daf301a	5 s	5,161	3,917	2/153	5.0%	1.8%	1.8%	197/419/48/0	5
semcod/code2logic	ba93489	12 s	21,423	16,927	27/359	17.7%	14.1%	14.1%	1,474/3,081/121/4	3
semcod/code2docs	c738aff	9 s	6,717	35,447	57/265	47.1%	77.0%	47.3%	283/876/396/0	0
semcod/redup	a175fb0	6 s	7,204	19,173	62/277	49.2%	55.9%	10.8%	476/1,205/703/0	0
subactor/platform	3e96573	6 s	10,628	11,002	25/688	5.9%	9.3%	8.9%	185/993/93/0	1

I/W/R/B means info/warning/review_required/blocking. Full commit hashes, graph fingerprints and diagnostic distributions are in baseline.json.

Warnings and explicit exceptions

• code2llm, pactfix and code2logic contain deliberately invalid parser fixtures and/or unsupported PHP, Ruby or C# inputs.
• Java extraction could not run for repositories containing Java because the clean runtime had no JDK. This is an explicit local exception; Java remains a required CI job.
• subactor/platform has one configuration file above the shared 524,288-byte limit.
• No repository-specific semantic options or thresholds were introduced.

Repeated defect selected for the first iteration

CHANGELOG_WITHOUT_IMPLEMENTATION occurs in all seven repositories (2,877 findings in total). Sampling separates two classes:

• substantive claims such as adding Jenkinsfile support or structured HR intent; these must remain reviewable when no implementation evidence exists;
• release-note mechanics such as Update project/calls.mmd, placeholder sections and summaries like ... and 12 more files; these are not behavioral claims and currently inflate both CHANGELOG_WITHOUT_IMPLEMENTATION and UNLINKED_RECORD.

Broadly linking changelog prose to module topics would manufacture evidence for the first class. The controlled change will instead classify only proven non-actionable release-note mechanics and leave substantive claims unchanged.

This site is open source. Improve this page.