task_061026_20 — Contract Follow-ups: DINTaskAuditor Size (#201 A4), Dual-Role Guard (#180), Cross-Model S5 (#193) (Umer Majeed)

[umermjd11]

cc @umeradl. This is the tracking discussion for task_061026_20.

Full spec: Developer/tasks/task_061026_20.md. It lands with the forwarding PR. The source plan is task-plan-051026-1, reviewed in PR #219 (verification, decisions, decisions applied).

Timeline

Oct 6 – Oct 9, 2026 (about 3 working days). Parts A and C can start right away. Part B ships with Part A, as two commits in one PR, so the CI size gate proves both fit.

Parts

Part	Issue	PR	Summary	Estimate
A	#201 Part A item 4	PR 1, commit 1	DINTaskAuditor size review. Four getters with no outside reader become internal (−320 B → 22,398 B). A fold review of the dispute and reassignment paths is required	1 d
B	#180	PR 1, commit 2	Per-address dual-role guard in registerDINAuditor (+102 B), plus the stale DinValidatorStake NatSpec	0.5 d
C	#193	PR 2	Two-level S5: the per-slasher GI ring stays, and a per-validator timestamp ring is added (7 days / threshold 6, placeholders; 0 means off on upgraded proxies)	1.5 d

Decisions (resolved on PR #219)

1. Four getters, with one combined budget. The DINTaskAuditor budget across both tasks is: 22,718 → 22,398 (A) → 22,500 (A+B) → 22,575 B with task_061026_21's every-batch commitment check, a margin of 2,001 B. That's in the warn band but far from the 1,024 B fail line. The ways back out are a fold that actually saves bytes, or dropping Is_testdataCIDs_Assigned.
2. security: dual-role registration allows one operator to control both the auditor filter and T1 aggregator in the same GI #180: per-address guard, with three tests.
3. security: S5 recidivism is per-model, so a validator can spread missed votes across models and never escalate #193: option A, the two-level ring. The MECHANISM_DESIGN.md:87 discrepancy is recorded, not changed.

Interplay

• PR test(adversarial): P3Adversarial.t.sol — task_240926_17 Part 2 #218 (P3Adversarial.t.sol): its Row 6 and Row 11 test_knownGap_… tests flip to test_defended_… in whichever PR lands second.
• task_061026_21 (full-GI alignment) shares DINShared.sol, the DINTaskAuditor ABI and its byte budget. Both specs have a coordination section.
• PR feat(indexer): implement DIN Protocol subgraph — platform + task-level contracts #29 (subgraph) gets an ABI note for the internal getters.

Post progress updates, questions and PR links here. If anything in the spec is unclear or doesn't look right, comment here and cc @umeradl.

🤖 Generated with Claude Code

[umermjd11]

Spec PR: #232. It adds this task's spec, the companion task, and closes task_021026_19.

[umermjd11]

PRs raised:

PR	Covers	Status
#235	Parts A + B (#201 A4, #180)	DINTaskAuditor 22,718 → 21,290 B (3,286 B margin, out of the warn band). Getter trim plus three folds that save bytes; dual-role guard (+102 B). 494 tests pass
#236	Part C (#193)	Two-level S5 (7 days / 6, 0 = off on upgraded proxies), new ValidatorEscalatedS5Global event, S5_GLOBAL_* deploy keys. DinValidatorStake +749 B. 502 tests pass

• PR test(adversarial): P3Adversarial.t.sol — task_240926_17 Part 2 #218's Row 6 and Row 11 known-gap tests are not flipped. Both PR bodies say how to flip them when test(adversarial): P3Adversarial.t.sol — task_240926_17 Part 2 #218 rebases.
• The two PRs conflict on one line of DinValidatorStake.md (adjacent caveats No. 4 and No. 5). Keep both lines; whichever PR merges second rebases.
• ABI note posted on feat(indexer): implement DIN Protocol subgraph — platform + task-level contracts #29.

[umermjd11]

Update: all PRs for this task are raised and merged.

PR	Covers	Status
#235	Parts A + B (#201 A4, #180)	Merged 2026-10-06
#236	Part C (#193)	Merged 2026-10-06

No PRs remain for task_061026_20.

• Follow-up (outside this task's scope): #286 brings MECHANISM_DESIGN.md's S5 tier in line with the code (MIN_STAKE + jail, no blacklist), per the MECHANISM_DESIGN.md S5 penalty doesn't match code (entire stake + blacklist vs MIN_STAKE + jail) #274 decision. It's open.
• Still pending on test(adversarial): P3Adversarial.t.sol — task_240926_17 Part 2 #218: its Row 6 and Row 11 test_knownGap_… tests still need to become test_defended_… when it rebases, as both PR bodies describe.

Work now moves to the companion task, task_061026_21 (Discussion #231).

[umermjd11]

@Santiagocetran, could you run the din-task-close skill on this discussion (din-task-close 230)? All PRs for task_061026_20 are merged: No. 235 (Parts A + B) and No. 236 (Part C). The task is assigned to me, so under Developer/MAINTAINING.md No. 1 a peer has to close it.

Neither No. 286 (the S5 doc follow-up) nor the No. 218 test renames are deliverables of this task.

[umeradl]

Closing task_061026_20. All Part A, B and C deliverables are verified on develop at e25df81.

Forced close, not a peer check. This task is assigned to Umer, and din_peer_check.py returned REFUSE for umeradl, which is the same person as umermjd11. Umer ran din-task-close with an explicit forced override rather than waiting for the peer pass requested above, so this departs from Developer/MAINTAINING.md No. 1 and No. 5.3. The verification below was still run in full, and a peer can reopen this if they disagree.

Merged PRs (both MERGED into develop, both ancestors of e25df81):

PR	Covers	Merged	Merge commit	Peer review
No. 235	Parts A + B (issue No. 201 A4, No. 180)	2026-10-07	e66d7bc	Blocking items No. 1 to No. 3 fixed in 5022604, re-reviewed
No. 236	Part C (issue No. 193)	2026-10-25	55c4b52	Double-count finding fixed in d63ed98, reviewed by @Santiagocetran and @robertocarlous

The earlier thread update gives 2026-10-06 for both merges. The dates above are from GitHub.

Part A

Deliverable	Evidence
Four getters internal	DINTaskAuditor.sol:95, :103, :227, :262. All four are gone from dincli/abis/DINTaskAuditor.json; rewardClaimed, testDataDisputes and giRewardSnapshot stay public
Size table and folds	In the No. 235 body. Three folds kept; each one shrinks the contract
ABI, docs, No. 29 note	ABI regenerated, DINTaskAuditor.md updated, note on PR No. 29
Gate and tests	See Verification below

Part B

Deliverable	Evidence
Guard, interface entry, error	DINTaskAuditor.sol:681-682, DINShared.sol:119, DINShared.sol:376
Three tests	StakingEnforcement.t.sol: test_registerDINAuditor_aggregatorSameGI_reverts, _aggregatorOfEarlierGI_registers, _secondAddressWithOwnStake_registers
NatSpec	DinValidatorStake.sol:121-122 names releaseGIRegistrationSlots; no "not yet enforced" remains
Threat model Row 6	DEFENDED (single address), with the Sybil limit noted

Part C

Deliverable	Evidence
Two-level S5	_partialSlashTimes at DinValidatorStake.sol:189, escalation at :367, setS5GlobalParams at :672, defaults at :240-241 (7 days / 6)
Event	Sibling event ValidatorEscalatedS5Global at :195; ValidatorEscalatedS5 keeps its signature
Deploy keys	S5_GLOBAL_WINDOW / S5_GLOBAL_THRESHOLD in DeployPlatform.s.sol and DeployPlatform.md
Tests	test_s5Global_* in SlashingInvariants.t.sol, override tests in DeployPlatform.t.sol and DeployPlatformScript.t.sol; test_crossModelGICollision_slashPartialDoesNotUnderflow retained
Docs	DinValidatorStake.md, MECHANISM_DESIGN.md:81, threat model Row 11 (DEFENDED above the global threshold). The :87 discrepancy is recorded at :88, unchanged, as specified

Verification on e25df81 (clean detached worktree, real via_ir profile):

• npm ci && forge build succeeded.
• contract_size_gate.py: DINTaskAuditor 21,290 B (3,286 B margin, ok), DinValidatorStake 10,921 B (ok). DINTaskCoordinator is 23,497 B (warn), unchanged by this task.
• forge test: 516 passed, 0 failed.
• pytest -m "not integration": 606 passed.

Deviations from the spec:

• __gap: stays at uint256[50] rather than shrinking. The spec was amended at PR No. 236 under the storage_layout.md core rules.
• S5 escalation epoch: s5EscalationEpoch is not in the spec. It was added in d63ed98 so that no miss counts toward two escalations.
• Auditor budget: the folds took DINTaskAuditor well below the budgeted 22,500 B.

Left open, not deliverables of this task:

• Issue No. 201 stays open for its Part B, which the spec puts out of scope.
• PR No. 218 is open. Its diff already has test_defended_dualRoleRegistration but still has test_knownGap_perSlasherS5Evasion, so its Row 11 flip is still due when it rebases. No known-gap test is on develop.
• PR No. 286 (MECHANISM_DESIGN.md S5 tier) is open.