feat(gsdc2023): add TripWhitelistDDGate (default-deny DD-carrier gate)#103
Open
rsasaki0109 wants to merge 1 commit into
Open
feat(gsdc2023): add TripWhitelistDDGate (default-deny DD-carrier gate)#103rsasaki0109 wants to merge 1 commit into
rsasaki0109 wants to merge 1 commit into
Conversation
Phase 80 showed the broad DDAnchorGate (anchor coverage >= 0.6) is net-negative on Kaggle: ~84% of DD rows fell in p95_delta>=3m trips, with a 105.72m spike on ebf-y/pixel5. Broad DD gating cannot be rescued by a single floor. TripWhitelistDDGate keeps fgo_dd_carrier only for trips that are explicitly whitelisted, or that clear conservative internal-consistency thresholds (anchor>=0.8, dd_epochs>=1, dd_pairs_mean>=4.0, no_tdcp absent) -- stricter than the 0.6 floor. A deny-list always wins. It reads only internal signals (never truth), so an offline A/B audit can emit the whitelist and production replays it deterministically. It is a duck-typed drop-in for CombinedGate(dd=...). Adds 8 unit tests (incl. the marginal trip the broad gate kept but this denies, allow/deny precedence, DD-density floor, and CombinedGate substitution). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
rsasaki0109
force-pushed
the
feat/gsdc2023-dd-trip-whitelist-gate
branch
2 times, most recently
from
6679b69 to
5250a11
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Phase 80 showed the broad
DDAnchorGate(anchor coverage ≥ 0.6) is net-negative on Kaggle: ~84% of DD rows fell inp95_delta≥3mtrips, with a 105.72 m spike onebf-y/pixel5. Broad DD gating cannot be rescued by a single floor.Changes
TripWhitelistDDGateinexperiments/gsdc2023_ab_gates.py— keepsfgo_dd_carrieronly for trips that are explicitly whitelisted, or that clear conservative internal-consistency thresholds (anchor≥0.8, dd_epochs≥1, dd_pairs_mean≥4.0, no_tdcp absent) — stricter than 0.6. A deny-list always wins. Reads only internal signals (never truth), so an offline A/B audit can emit the whitelist and production replays it deterministically. Duck-typed drop-in forCombinedGate(dd=...).tests/test_gsdc2023_ab_gates.py— 8 new tests (the marginal trip the broad gate kept but this denies, allow/deny precedence, DD-density floor, CombinedGate substitution). 24/24 pass.Scope: this lands the gate logic at the A/B-audit seam (where the whitelist is produced). Production wiring into
build_gsdc2023_bridge_submission.py+ whitelist-YAML generation is a follow-up (needs config plumbing + the full candidate data, not present locally).🤖 Generated with Claude Code