Phase 3: Prove monotonicity and realized-loss bounds F3.2/F3.3 #227
Labels
No labels
bug
cli
core
docs
event
experiment
figure
invariant
metrics
oracle
phase-0
phase-1
phase-2
phase-3
phase-4
phase-5
phase-6
provenance
revocation
tests
workload
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Blocks
Depends on
#222 Phase 3: Implement and prove the exact decision oracle
erikinkinen/AES
#194 Phase 2: Execute phase 2 and release evidence
erikinkinen/AES
#226 Phase 3: Prove adaptive oracle optimality F3.1
erikinkinen/AES
Reference
erikinkinen/AES#227
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Plan ID:
P3-15-ERecord type: Formal leaf
Scientific protocol: 2.2 · Backlog revision: 2.2.1
Specification: sections 7.4; parent requirements and common section-11/17.4 gates apply.
Source:
AES_PHASES_2_6_PLAN.md, issue heading at line 5396.Plan SHA-256:
8acdaeb52d6a94fcad7de64d64521cd465f608344d63560b45f9202f5af71212Protocol snapshot: Download the complete governing plan.
Tracking parent: P3-15 (#222).
Goal
Calculate certified adaptive and nonadaptive minimax risks.
Applicable parent requirements
Prerequisites
Steps
Definition of Done
Unsound corruption is explicitly outside the assumptions and failed premises have fixtures.
Execution and acceptance contract
Apply the governing semantic/data/statistical contracts in sections 3–5 and the relevant phase section, plus the section-11 checks applicable to this task. Section 17.4 defines the required acceptance evidence. Store the outcome and immutable prerequisite/input/output hashes at
artifacts/backlog/2.2.1/P3-15-E/acceptance.json.Only explicit dependency and parent-closure edges govern readiness; numbering adds no predecessor dependency. Checkboxes are work/acceptance items, not commit quotas. Keep implementation changes independently reviewable using the existing PR template. Execution, review and gate tasks close with their evidence/decision records and a PR only when tracked artifacts change.
An unsupported scientific effect is a valid result. Missing correctness validation, invalid inference or absent required review is not completion. Do not change endpoints, matrices, margins, inference rules or sample sizes outside the protocol change-control process.