Session Log

One entry per significant decision: what was done, what was considered, what was rejected and why.

Date: 2026-09-21
Done: Web-researched sourced principles (DRY, KISS, parse-dont-validate, hexagonal architecture, etc.) and committed the first 15 principles, 8 architectures and 6 practices as a private GitHub repo (grenudi/design-canon).

Date: 2026-09-21
Done: Tested the format against this repo’s real data specifically (not a toy example): schema changes, ALTER TABLE limits, a real semicolon-in-SQL bug, BLOB corruption, and 300k-row determinism on a real multi-core CI runner.
Considered: Doltgres, Dolt, gitsheets, GlueSQL.
Rejected: See tad-engine’s and text-as-data-template’s own session logs for the full comparison - this repo was the proving ground.

Date: 2026-09-21
Done: Moved dc.py, render.py and the invariant checks into a new template repo once they had been battle-tested here, then adopted that template’s engine back into this repo via .tad/, keeping this repo’s own render.py local since its output (grouped principles, “Watch out:”, “Also known as:”) is genuinely specific to this content.

Date: 2026-09-22
Done: Followed the same migration as text-as-data-template once subrepo was tested and adopted there.

Date: 2026-09-22
Done: Renamed the original repo to software-engineering-canon-legacy, generated a new software-engineering-canon from the template via GitHub’s actual template-generation API (not simulated), bootstrapped .tad/ with bin/adopt-engine.sh, and carried the real content and this custom render.py across - a battle test of the whole template pipeline end to end, at the request of the person running this project.
Considered: Keep using the original repo as-is.
Rejected: The point was specifically to prove a fresh repo generated from the template works for real, not simulated - this repo became that test.

Date: 2026-09-22
Done: git subrepo pull brought the generic engine’s definition-list render.py update cleanly, but this repo keeps its own local tools/render.py (deliberately, for its richer presentation), so the pull alone did nothing here - the same bold-to-definition-list edit had to be hand-applied to all 8 spots in this repo’s own renderer. Added the matching _config.yml, _layouts/default.html and assets/css/style.css by hand too, since those are not vendored.
Considered: Assuming the engine pull would cover the whole change here, since it covered it cleanly elsewhere.
Rejected: Checked rather than assumed: grepped for bold markers in this repo’s render.py right after the pull and found 8 still there, confirming the two repos needed genuinely different amounts of work for the same visual change.