-
Notifications
You must be signed in to change notification settings - Fork 3
v0.3
Anchor issue: #75
Status: 🔄 In progress — PR #74
There is a principle in engineering that is so obvious one hesitates to state it, and yet, judging by the state of a great many publicly-available repositories, it plainly requires stating: a feature built on a broken foundation is not a feature; it is a liability dressed as progress. GödelOS, at the time v0.4 begins, will have a green test suite or it will have nothing worth calling a foundation.
PR #72 achieved something significant: zero collection errors across 1,299 tests. But collection and execution are different things, and 167 tests that collect without error still fail at runtime — every one of them a pre-existing divergence between the tests' expectations and the source modules' actual behaviour, accumulated before the current development cycle began. PR #74, currently executing, will resolve them all.
| Metric | Value |
|---|---|
| Total collectible tests | 1,299 |
| Collection errors | 0 ✅ |
| Passing | 925 (861 + 24 + 40) |
| Failing | 167 |
| Skipped | 139 |
These are not mysterious failures. Each one represents a stale expectation — a test written against an interface that was subsequently refactored, a return format that changed, an attribute that was renamed. The fix is not clever; it is thorough. Copilot is working through them module by module.
| Module Group | Failures |
|---|---|
| Simulated environment | 9 |
| Meta-control RL module | 9 |
| Parser enhanced | 8 |
| Modal tableau prover (enhanced) | 8 |
| Type system enhanced | 8 |
| Explanation-based learner | 10 |
| Resolution prover (×2) | 14 |
| Belief revision | 6 |
| ILP engine | 6 |
| CLP module | 4 |
| Perceptual categoriser enhanced | 4 |
| Frontend modules | 7 |
| ... and approximately 46 additional | varies |
- All 167 failures resolved (PR #74)
- Full suite:
pytest tests/ -v→ zero failures, zero errors - No weakened assertions; no deleted tests; no mocked-away behaviour
-
mainbranch remains green on every subsequent push
The rules on test integrity are not negotiable. A green suite achieved by deleting inconvenient tests is not a green suite; it is a fraud, and one that will compound interest at a most unfavourable rate.