ESLint suggestion task registration

Registered before these task calls and after earlier outcomes were observed. ESLint #21122 is open and marked accepted. The manifest has SHA-256 e5a9a66a85a9218cc68056cb95f0a77bce9651cb185623f8aff38eae13939ce7 and binds the source, input snapshots, isolated tools, checks and paired random order.

The issue-specific check requires strict-null and, where safe, null-or-undefined suggestions for equality and inequality. It checks reversed operands and avoids a compound suggestion when repeating a call, property access, increment or shadowed undefined could change behavior. The public check exercises a single suggestion; separate regression controls check existing rule reports and cases that should not be reported. On ESLint commit 12cbff4125db3f6446f72fa0c48c46b92b120427, the baseline fails the issue and public checks and passes the regression controls. A reference derived from the closed, unmerged PR #21439, head bc55433a1e95d977affae246b676f04307382235, passes all three custom checks. This is a positive control, not an accepted upstream fix.

The original five-case upstream rule suite passes on the baseline. On the reference, three cases fail because the test expectations do not declare the newly introduced suggestions; the suite was not silently counted as passing or replaced with modified upstream expectations. The custom rule-reporting controls are partial and cannot establish full regression safety.

The restricted Linux VM, two randomized contexts, gpt-6-astra solver, 12-turn limit, deepseek-flash and gpt-6.1-sol blind AI reviewers, explicit rubric, solver-family adjudication and stopping rules match the mixed batch protocol. This is one additional issue chosen after prior results. Its history discusses a proposed solution, so leakage or prior exposure is possible. The cost target is below $1 but is not a hard cap. This cannot establish production correctness, cost savings, independent human review or autonomous memory management.