Repository navigation
test(roadmap): a synthesized chain edge is authored and must gate #235
Description
Problem
PAL-205 makes inferred position advisory — enumeration order (toc, #223) and filename collation between equal ranks (tie, #234). palee roadmap --auto-chain also derives depends_on from position: it chains topics by their order field, appending unordered ones after ordered ones (INV-47).
That similarity is misleading, and nothing in the suite said so. A roadmap is a file the learner typed whose only purpose is to declare a prerequisite plan, and INV-47 already lets an explicit depends_on override the synthesis. So the ruling recorded here: a synthesized roadmap chain edge is an authored edge and it gates. It carries no depends_on_source label, and an absent label is the fail-closed authored case.
The risk is the opposite direction: a future refactor that routes synthesized edges through the advisory labels "for consistency with the chain tiers" would pass the entire suite while turning every roadmap into a suggestion.
Fix
A test that asserts the consequence, not the intent — test/cli-roadmap-wikilink.test.ts, "a roadmap-synthesized chain edge gates, because the learner wrote the list":
- the imported note's frontmatter has
depends_on: [T-g-one]and nodepends_on_source; palee plan --jsonoffers onlyT-g-onewhile mastery is 0.
Proven load-bearing by making doImport label synthesized edges toc: the test fails on the missing label, and restoring the file byte-identical returns it to green.
Acceptance criteria
- Synthesized roadmap edges are written unlabeled.
- The dependent topic stays off the ready list until its predecessor is mastered.
- The test fails if the synthesis is ever demoted to an advisory source.
- INV-47's wording records the ruling, so the goal sentence ("nothing derived from list position gates") does not read as a bug report against
roadmap --auto-chain.
Environment
Repo Kuldeep2822k/cli, branch test/roadmap-order-gates stacked on test/pal205-gate-hygiene (#226). Node 24, @kuldeep2822k/palee 0.5.2. Test-only: no source file changes.
Activity
Kuldeep2822k commentedon Sep 30, 2026
Verification round - criteria 1-3 hold at
feat/73-tier0-hygiene@0d222bfsrc/cli/roadmap.ts:466-475writes a synthesized edge without a label, and:487-489removes a stale one first (deletes precede sets -src/storage/frontmatter.ts:125-130), so the dependent gates as authored:test/cli-roadmap-wikilink.test.ts:461asserts the label is absent (:495) and the dependent is withheld (:498-505, exactly['T-g-one']). Two independent routes make that test fail if the synthesis is ever demoted: adding adepends_on_sourcesurvives the removal, and turningundefinedinto an advisory default falls throughsrc/types.ts:514.I have left criterion 4's tick alone rather than change it, but flagging that the tree does not support it.
planning/invariants.md:75(INV-47) says nothing about labels or gating, and:73states affirmatively thatdepends_on_source"never affects gating" over a two-value domain - both now false againstsrc/types.ts:500andsrc/engine/dependency.ts:771. Nothing in the repo yet states that a synthesized roadmap edge gates because it is unlabeled, so a reader of INV-46's goal sentence ("nothing derived from list position gates") can still readroadmap --auto-chainas a bug. #238 owns the invariant rewrite; the question for you is whether #216's INV-47 restatement plus this issue's code comments count as recording the ruling, or the invariant text itself has to change.