SelfDoc Feed Pipeline

The website should be fed from the documentation system, not hand-maintained from memory.

Source of truth

The authority order is:

1. Runtime/source behavior in D:\code\ccode.
2. HELP, metadata, command registration, comments, @dottalk.usage, @dottalk.contract, contract scans, and canary evidence.
3. SelfDoc reports and manualgen manifests.
4. Reviewed manual sections.
5. Website pages in D:\dev\x64base-site.

When a DotTalk++ command or function is added, the expected flow is:

source change
  -> command/function registration
  -> source header usage contracts and source comments
  -> HELP / metadata / comments evidence / contract scans
  -> SelfDoc and CMDHELPCHK validation
  -> manualgen inventory / validate / dry-run
  -> reviewed manual section
  -> website command/function/API pages

Current report-only feed

The current repeatable feed uses:

cd D:\code\ccode
python tools\manualgen\manualgen.py --repo-root D:\code\ccode --manual developer inventory
python tools\manualgen\manualgen.py --repo-root D:\code\ccode --manual developer validate
python tools\manualgen\manualgen.py --repo-root D:\code\ccode --manual developer build-dry-run
python tools\selfdoc\build_engine_feature_crosswalk.py --repo-root D:\code\ccode

Generated outputs include:

• labtalk/reports/selfdoc/x64base_engine_feature_crosswalk_v1.csv
• labtalk/reports/selfdoc/x64base_engine_feature_crosswalk_v1.md
• docs/manuals/developer/generated/x64base_engine_feature_crosswalk_manual_section_v1.md
• docs/manuals/developer/manualgen/generated/manualgen_build_dry_runs/<run>/developer_manual_build_dry_run.md

Promotion rules

• Do not auto-promote uncertain scanner output into public copy.
• Preserve status labels: runtime-evidenced, source-evidenced, help-catalog-evidenced, planned, canary, review-needed.
• Do not claim a lane is complete when the evidence says active beta, canary, or review-needed.
• Keep generated reports local until reviewed, then summarize them on the website.
• Command and function pages should cite the extraction snapshot and should be regenerated when HELP/CMDHELP/usage contracts change.
• Commands that launch external apps, perform network requests, write files, write HELP DATA, or write DBFs must carry explicit mutates: and risk: fields before being promoted as stable docs.

Procedural curation

The repeatable curation loop is:

scan source headers
-> harvest comments and usage contracts
-> rebuild or inspect HELP/CMDHELP surfaces
-> run CMDHELPCHK or report-only validators
-> run manualgen inventory/validate/dry-run
-> review generated sections
-> update website derivatives

CMDHELP, CMDHELPCHK, MAINT, DDICT, MANUAL, BBOX, and contract tools all belong in this loop. Their documentation should state mutation boundaries explicitly so a report-only inspection command does not get mistaken for a promotion command.

Current limitation

Manualgen currently validates the manual structure but requires Python >= 3.12. The latest run used Python 3.11.0, so validation failed on that environment check only. Boundary failures were 0, and the dry-run manual artifact was still generated for review.