Skip to content

Revert calibration post-processing (measured regression on gate rounds) #172

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
lukeinglis merged 1 commit into from
Aug 24, 2026
Merged

Revert calibration post-processing (measured regression on gate rounds) #172

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
34 changes: 0 additions & 34 deletions eval.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,38 +93,6 @@ def is_async_forecaster(forecaster: Forecaster) -> bool:
return result


_MARKET_ANCHOR_WEIGHT = 0.94
_EXTREMITY_FLOOR = 0.04
_EXTREMITY_CEIL = 0.96
_DATASET_SHRINKAGE = 0.06


def _apply_calibration(
forecasts: dict[str, float],
questions: list[Question],
) -> dict[str, float]:
logger.debug("apply_calibration_start", n_forecasts=len(forecasts), n_questions=len(questions))
q_by_id: dict[str, Question] = {q.id: q for q in questions}
calibrated: dict[str, float] = {}
for key, prob in forecasts.items():
base_id = key.rsplit("_", 1)[0] if "_" in key else key
q = q_by_id.get(base_id) or q_by_id.get(key)

if q is not None:
is_market = q.source.lower() in MARKET_SOURCES
if is_market:
fv = getattr(q, "freeze_datetime_value", None)
if fv is not None and 0.0 <= fv <= 1.0:
prob = _MARKET_ANCHOR_WEIGHT * fv + (1.0 - _MARKET_ANCHOR_WEIGHT) * prob
else:
prob = (1.0 - _DATASET_SHRINKAGE) * prob + _DATASET_SHRINKAGE * 0.5

prob = max(_EXTREMITY_FLOOR, min(_EXTREMITY_CEIL, prob))

calibrated[key] = prob
return calibrated


_PROVIDER_PREFIXES = (
"vertex_ai/", "openai/", "anthropic/", "google/",
"litellm/", "azure/", "bedrock/",
Expand Down Expand Up @@ -413,8 +381,6 @@ async def run_eval(
multi_forecaster=multi_forecaster, # type: ignore[arg-type]
)

forecasts = _apply_calibration(forecasts, questions)

has_composite = any(
"_" in k and k != q_id
for k in forecasts
Expand Down
Loading