research: identify necessary and duplicated responsibilities in main and next

[mohanagy]

User story

As Madar's maintainer, I want each major responsibility in main and next connected to a current capability or necessary guarantee, so that I can identify justified simplification without deleting useful behavior.

Acceptance criteria

• Record both exact branch revisions and map public entry points to their major supporting responsibilities.
• Classify responsibilities as keep, consolidate, retire or uncertain, with source/dependency evidence. Separate production code, tests, scripts and generated or external artifacts.
• For any proposed retirement, identify consumers, compatibility obligations and protected behavior. Line count or few textual references alone do not justify removal.
• Deliver one inventory and a justified baseline choice for later product work. List unresolved questions without automatically extending the review.

Scope

One read-only inventory pass, with deeper tracing limited to a candidate TypeScript/Node workflow: cross-file behavior tracing for a bounded code change. That use case is a review lens, not a qualified product claim.

This story does not authorize tests, benchmark reruns, deletion, branch promotion, new infrastructure or production implementation. A completed inventory is a maintainer decision artifact, not proof that Madar saves work.

Starting revisions

• main: 3371ada8425efa7f8cabdac781fa227feaea7a6a
• next: 2b144504ddf924d64cce51db601bb599be0b6c44

Supersedes the repository-review portion of #740. Existing user-owned root changes, worktrees and historical evaluation evidence remain intact. The maintainer authorized this read-only review after reviewing the focused story drafts.

Result — 10 September 2026

Completed by the source-backed inventory in the closing comment. This closes the read-only discovery outcome only; product usefulness and implementation remain unproved/deferred under their own stories.

[mohanagy]

Completed: one read-only responsibility inventory

Reviewed exact main 3371ada8425efa7f8cabdac781fa227feaea7a6a and next 2b144504ddf924d64cce51db601bb599be0b6c44. Live branch identities remained unchanged. No tests, builds, diagnostic runners, source edits, deletions or model experiments were performed.

Result: targeted consolidation is justified for consideration; broad deletion is not. Next adds 13,875 net source lines and 53,137 test/script lines. These are physical lines including comments and fixtures, not unnecessary-code or runtime-cost measurements. Tests/scripts/docs are outside the configured clean production package; evaluation modules under src still ship. Sources: package and build scope.

Responsibility inventory

Responsibility	Disposition and evidence
Public CLI/MCP, retrieval, evidence packing and freshness	Keep: active entry points and shared runtime consumers. See CLI pack and MCP.
Task routing/pack assembly	Consolidation candidate: CLI orchestration repeats policy also owned by MCP. Preserve transport-specific defaults, rendering, sessions and authorization.
Final evidence status	Consolidation candidate: initial judgment, serialized reconciliation and strict MCP can revise status. No current contradiction was reproduced.
Source authentication, graph identity/artifacts, legacy readers	Keep: source rendering, publication and compatibility loading have current consumers.
SPI/legacy extraction	Uncertain simplification: default-auto composition actively uses both. No whole-family deletion justified.
Installers, languages, exports, federation and evaluation CLI	Active/optional; usage value not measured. tool profiles limit advertised tools; CLI imports still load optional evaluation modules.
Unexposed add parser	Small retirement candidate: declaration and its interface are the only scoped src/test references. Confirm supported external deep imports before removal.
Source-preservation/compatibility tests and evaluation-independence controls	Keep: they protect current guarantees. See preservation test and CI controls.
Worker-failure detection	Keep protection; consolidate guard scanning with repeated release scanning. Preserve logs and error/missing-log handling.
Historical qualification/receipt support	Current consumers remain. qualification validator has an uncertain release-policy role; contract preservation is not product usefulness.
Standalone Stage 3 artifact runner	Conditional archive candidate: runner had no package/CI/src/test caller found. Confirm external reproducibility needs; shared libraries remain used.
Worker diagnosis and local/generated/external artifacts	Diagnostic remains deferred under #756. Generated/untracked/external artifacts were excluded from tracked counts and preserved.

Baseline decision and stopping point

Use pinned main as the control and pinned next as the candidate. Main precedes the 22-commit change set; next retains the current artifact/source-preservation work. This is not a quality ranking, branch promotion or product-success verdict.

All four inventory criteria are satisfied: exact revisions, source/consumer map, qualified retirement boundaries, and a baseline decision with explicit uncertainty. Broad assertion redundancy, external feature usage, published-package contents and task-level benefit remain unproved; they do not automatically extend this review.

The actual documentation in draft #742 still needs to match the focused backlog. #754 remains deferred. Findings above are a shortlist, not authorization for a rewrite, another test platform, deletions or new runs.