Source revision: 1021c056 (main)
Scope: docs/findings/ only
Related: issue #152, Phase 3 (bounded corrections)
docs/README.md states that for archive and evidence collections
“category-level discoverability can be sufficient” and that one inbound link
per archival record must not be manufactured merely to reduce a warning counter.
That rule is correct and is not being overridden here.
The defect is different. At the source revision docs/findings/ held 15
documents with no directory index, and the collection is not homogeneous
archive material. It mixes documents that still state a working project
thesis — distributed-agent-coding.md, cited by issue #2, and
2026-08-28-emergence-from-vague-goals.md, cited by issue #22 — with dated
landscape studies, bounded repository-health evidence, and external source notes
that explicitly disclaim endorsement. A reader arriving at the directory listing
cannot tell which is which, and the flat date-prefixed filenames do not help.
Distinguishing current thesis from retained history is the navigation gap. The warning delta below is a consequence of closing it, not the reason for the pass.
All 15 documents were read and classified into four groups. The index links each document exactly once.
| Group | Count | Basis |
|---|---|---|
| working theses and program framing | 4 | still states a position the project builds on |
| repository-health and navigation evidence | 5 | bounded issue #152 review with a measured delta |
| growth, discovery, and free-compute landscape | 3 | dated diagnosis of the project’s external position |
| historical records and source notes | 3 | retained memory or external citations, not guidance |
No document was moved, renamed, deleted, reclassified in its own text, or given new authority. No detector rule changed. No warning was suppressed.
Only three cross-references to open issues are stated, and each is a citation that already exists in the issue body — #2, #10, and #22. Associations that looked plausible but were not present in repository or issue text were dropped rather than asserted.
The canonical observatory was run on the source revision and on the proposed tree, with output written outside the scanned repository:
python tools/idkgraph_observatory.py . --output-dir /tmp/idkgraph-observatory --pretty
| Finding | Before | After |
|---|---|---|
orphan_document_candidate |
26 | 16 |
document_referenced_only_by_non_markdown_artifact |
11 | 9 |
docs/findings/ findings |
12 | 0 |
accepted_decision_without_document_link |
0 | 0 |
| unexpected deterministic errors | 0 | 0 |
The two resolved notices and ten resolved warnings are exactly the twelve
docs/findings/ findings; no finding outside this directory changed.
Reading and classifying 15 documents took roughly 30 minutes of review effort
in this pass; those are not independent-human minutes. The
detector was accurate on this cohort: every flagged document genuinely had no
inbound Markdown link, and the two document_referenced_only_by_non_markdown_artifact
notices correctly identified documents reachable only from non-Markdown
artifacts. Zero false positives were observed on this rule at this revision.
That is a statement about the detector, not about the repository. The repository quality question — whether these documents needed an index — was answered separately above, and for a genuinely homogeneous archive the honest answer would have been no.
This pass does not claim independent-human review minutes. Issue #167 remains the separate independent-review gate for issue #152, and the author of this pass is not an independent reviewer.