1
# Stop Conditions
2
3
Check these after every iteration (Step 6 of `loop.md`). Stop the loop and report as
4
soon as any of these is true — do not keep iterating past them.
5
6
## Hard stops (always stop)
7
8
- `max_iterations` reached.
9
- The full test suite is red and the harness cannot get it green again after a
10
  reasonable attempt (e.g. 2 focused retries) — stop and report the failure rather
11
  than leaving the repo in a broken state.
12
- A behavior requires infrastructure not available in this environment (a live
13
  database, an external API with no test double, network access) and no reasonable
14
  mock/fixture can be constructed — record it as a deferred gap, don't fake a pass.
15
- A behavior gap turns out to require a product/design decision the harness can't
16
  infer from the code, docs, or issue tracker (e.g. undefined behavior on conflicting
17
  input with no spec) — stop and ask the user rather than guessing.
18
19
## Soft stops (stop and report, this is success)
20
21
- Coverage for the targeted module reaches a clear plateau: two consecutive iterations
22
  with no meaningful coverage increase (e.g. < 1%) because remaining uncovered lines
23
  are trivial (simple getters, defensive code that can't realistically be reached).
24
- All identified untested behaviors in the targeted scope have been covered — don't
25
  invent artificial tests just to keep the loop running.
26
- Diminishing returns: the last 2-3 iterations added tests for increasingly contrived
27
  edge cases with low real-world likelihood. Note these as candidates but stop.
28
29
## Explicitly not a stop condition
30
31
- Reaching a specific coverage percentage number (e.g. "80%") — coverage percentage is
32
  a signal, not a goal; a module can be well-tested at 70% (all the real branches
33
  covered) or poorly tested at 95% (assertion-free tests padding the number). Judge by
34
  behavior coverage, not the raw metric.
35