Measurement

This page holds the latest measurement of dreamcatcher against the standard. The release checklist replaces it before each major or minor release. Earlier measurements are in Git history.

Measured on 2026-11-01 at 54499f6b, for the v9.8.0 release preparation. The review covers the three pull requests merged since v9.7.0: the home page's point on long-horizon work and the narrower rule for documentation entries in the changelog (#889), a formatting fix to the home page (#890), and the order of the kinds of agent work as assignment, campaign, conversation (#895). The consistency review and the pass over open issues ran beside this measurement.

Results

Criterion	Measured	Status	Issue
Cohesive modules	0 reviewed modules without one nameable responsibility	met
Focused functions	0 reviewed functions without one nameable job	met
No unused exports	31 public definitions without a direct named importer, as at v9.7.0	met
One way	The order of the kinds comes from AgentWorkKind; one copy deferred	met	#896
No suppressions	0 noqa, 0 type: ignore, 3 platform pragmas	met
One vocabulary	0 disagreements; 1 open issue rewritten in the concepts' words	met
True documents	#895 corrected the rotation order in the concepts and the architecture	met
Docs by purpose	Tutorials, task guides, references and explanations remain separate	met
Versioned contract	State format 8 unchanged since v9.0.0; contract remains 1	met
Machine-checked rules	The 11 import-linter contracts and the hooks of v9.7.0 remain	met
Fast readable tests	59.19 seconds, 1,566 tests passed, 100% statement and branch coverage	met
Current tracker	12 open issues read; 1 rewritten in the concepts' words	met
Reviewed changes	2 of 3 merged pull requests answer every definition-of-done question	short
Rules not cases	11 ledger entries, unchanged	met
Invariants by construction	71 nullable class fields, as at v9.7.0	met
Few concepts	No concept added or removed	met
Symmetry	Each kind keeps its parallel operations; only their order changed	met

Deferred work and previous findings

The v9.6.0 measurement filed #870, the calm field's other names in code and tests. It remains open.

The coherence review of #895 found that init's next step and the status guidance for an empty section each say how to start each kind of agent work, in wording that has drifted apart. #895 deferred it, and issue #896 now holds it.

Open issues in this interval:

• The open issues are #32, #492, #762, #844, #856, #857, #870, #882, #893, #894, #896 and #897.

• 893 called a sub-issue a "stage"; it is rewritten.

• 882 still awaits the product owner's triage, which this pass does not

  replace.

Notes

• Cohesive modules, Focused functions. The largest modules are those of v9.7.0, at the same sizes: github.py (931 lines), agent_rounds.py (785), status/agent_work.py (629) and status/report.py (543). The source changes since v9.7.0 reorder existing members and text, and add or remove none.
• No unused exports, One way. The public definitions and their importers are those of v9.7.0. #895 made the route order follow AgentWorkKind. _report_next_steps still chooses init's next step through a chain of branches in that order, and #896 records it beside the duplicated wording.
• No suppressions, Machine-checked rules. The three coverage pragmas remain at teardown.py:20, teardown.py:110 and commands.py:270, all platform arms. The import contracts and every hook pass.
• One vocabulary, True documents. The consistency review found no disagreement in the code, the output or the documents. No replaced word appears in what changed since v9.7.0. The home page's new point speaks of "long-horizon work" and "a roadmap" as plain description, not as concepts. A sweep of every list of all three kinds found none left in the old order outside the released changelog, the upgrade notes and the dated specifications.
• Fast readable tests. The default suite passed with 1,566 tests, one platform-only skip, and 100% statement and branch coverage, in 59.19 seconds; a first run beside other work took 61.01 seconds. Its slowest test, test_a_closed_parent_ends_the_campaign_once_its_round_finishes at 21.46 seconds, sets much of that time. The browser group passed 69 tests and the read-only GitHub integration group 11, each with one platform-only skip.
• Reviewed changes. #890, a formatting fix to the home page made in GitHub's editor, has no description. #889 and #895 answer every question.
• Rules not cases, Invariants by construction, Few concepts, Symmetry. The ledger keeps its 11 entries and the nullable-field inventory its 71.

Method

• tools/measure_source.py supplies physical module and function line counts; review assesses responsibility rather than imposing a limit.
• The diff of src from v9.7.0 establishes that no definition, import or field changed, so the export and nullable-field inventories of v9.7.0 stand.
• A search for each word that a "Replaces:" line names covered the lines changed since v9.7.0, the source, the documents, the repository's skills and every open issue.
• Test results establish local coverage, timing and rendered behaviour. PR descriptions establish the definition-of-done record, and current issue bodies establish the state of the tracker.