Repository navigation
test(network): verify opentype 75/25 sealed payouts - #314
Conversation
Exercise runtime-only, shared and separate hotkeys, plus unused and filtered runtime mass through signed leaves, HTTP seals and validator verification. Keep production trust and algorithm configuration unchanged. Co-Authored-By: Claude Code <noreply@anthropic.com>
|
Greptile (@greptileai) review — please review current revision 77542ba, particularly 75/25 fixture coverage, signed leaf/HTTP seal assertions, residual burn, and validator recomputation. Test-only; no production configuration change. |
|
Greptile (@greptileai) review — the previous check for final revision 77542ba ended neutral with "Review was cancelled" and produced no review. Please retry this unchanged test-only revision. All CI checks pass; review completion remains required. |
This comment has been minimized.
This comment has been minimized.
Cover the shared-rounding blind spot identified in review with literal expected vectors for all five payout cases. A one-unit UID0 mutation now fails every case. Co-Authored-By: Claude Code <noreply@anthropic.com>
|
Greptile (@greptileai) review — latest revision adds independent literal final_vector expectations for all five cases, fixing the u16 rounding coverage gap. All five fail under the one-unit UID0 mutation; baseline passes. Please review the new HEAD. |
|
Greptile (@greptileai) review — current final SHA 7e102c1 needs a completed review. The post-push review was cancelled/neutral again. All CI is green and the prior u16 finding is fixed, mutation-tested, replied to and resolved. Please retry review on this unchanged revision. |
Problem and behavior
Expand the existing multi-container reward-path regression into five OpenType fixture cases: runtime-only, quality/runtime paid to the same hotkey, different hotkeys (75/25), unused runtime mass, and filtered runtime mass. Assert signed leaf scores, independent literal chain-ready u16 vectors, exact participant coverage, sealed algorithm-3 HTTP weights, residual UID0 burn, and independent validator verification without chain submission. Oversized out-of-metagraph mass cannot dilute or inflate eligible rewards.
Test-only change: container weights are fixture inputs, not proof of OpenType's own allocation implementation, live GPU execution, or on-chain payment. Existing synthetic 40% OpenType challenge share is unchanged. No production algorithm configuration, signed trust document, deployment, or public API changes.
Validation
scripts/check_repo.py --final--check-examplesalso passesFull suite has two existing third-party deprecation warnings. Generic skill
repo-audit.shcannot recognize.gitfiles in worktrees; repository-specific final hygiene gate passes.Public contract
BASE_*/domain compatibility remain intactGreptile
@greptileai reviewRisk
Test-only. No production emission, signature, VM isolation, migration, or rollback changes. Real local signing, gateway HTTP and validator recomputation are exercised with fake external challenge/chain boundaries. Production algorithm-3 activation still requires the independent offline owner ceremony.
🤖 Generated with Claude Code
No outstanding findings prevent merging.
Summary
The network test now checks exact sealed u16 vectors for all five OpenType payout cases. No outstanding findings remain.
Reviews (3) · Last reviewed commit: "test(network): pin independent sealed u1..."