| """Generator knobs for the difficulty ladder (SPEC 9.2, benchmark review F1/BM-1). |
| |
| decoys healthy rules and one route that use the same constructs the planted defects use (an average |
| of per-pod values, per-pod grouping on a channel), each with the reason it is right; changing |
| one costs preservation (on by default: it closes the "odd construct means bug" shortcut) |
| e3-evidence E3 tickets come with alert-channel and PagerDuty history whose explanation needs the route or |
| inhibition match (on by default: SPEC 2.2 wants 1-3 packets per task) |
| strip-locators the comments a planted defect leaves on its rule or route are dropped, except the TODOs a |
| ticket refers to (off by default: the next rung of the ladder if E1-E3 come in above the band) |
| label-rule README says outright that labels beyond the required ones are fine when they don't change where |
| an alert goes, as the expert grader reads it (off by default: it changes the workspace of every |
| task, so it waits for the next build) |
| |
| The build workers read ACTIVE; `alertforge expert --knobs a,b` sets it for a run. |
| """ |
|
|
| from __future__ import annotations |
|
|
| ALL = ("decoys", "e3-evidence", "strip-locators", "label-rule") |
| DEFAULT = frozenset({"decoys", "e3-evidence"}) |
| ACTIVE: set[str] = set(DEFAULT) |
|
|
|
|
| def set_active(names) -> None: |
| bad = set(names) - set(ALL) |
| if bad: |
| raise ValueError(f"unknown knobs: {sorted(bad)}") |
| ACTIVE.clear() |
| ACTIVE.update(names) |
|
|
|
|
| def on(name: str) -> bool: |
| return name in ACTIVE |
|
|
