NORMATIVE REGISTRYBest RLHF Data
DOCUMENT CONTROLRLHF-2026.11 // NORMATIVE
ALIGNMENT PIPELINESSFT · DPO · PRM · RLVR
[00]
NORMATIVE GOVERNANCE // LAB VERIFICATION & BENCHMARK AUDIT

Post-Training Lab Verification & Normative Benchmark Slot Protocol

Independent post-training data labs, RLVR environment builders, and alignment tooling maintainers may submit anonymized trajectory manifests and telemetry schemas to verify or update their specification record on bestrlhfdata.com.

[01]
REGISTRY ALLOCATION TIERS

Record Verification, Partner Badges & Featured Entity Lease

Every commercial record listed in this registry undergoes cap-table neutrality screening and sample-task turnaround verification prior to badge issuance.

Registry verification and sponsorship structure
TierAllocation LimitIncluded Scope
Level 01 // Canonical Open-Source & Baseline Ecosystem NodeZero Commercial Fee — Public Repository & Schema AuditCovers Apache-2.0 / AGPL alignment frameworks (TRL, Argilla, Label Studio, Cleanlab), public benchmarks (SWE-bench, RewardBench), and baseline market comparators indexed from public documentation.
Level 02 // Verified Independent Post-Training NodeQuarterly Technical Audit — 100% Lab-Independent Cap Table MandatoryVerifies PhD/SWE onboarding protocols, Shadow-DOM anti-LLM telemetry fields, containerized execution support, and verified matrix routing (capped at 3 verified partner nodes).
Level 03 // Featured Normative Benchmark Entity (1 of 1)Single Registry Slot — Full 50-Trajectory, Rubric Alpha & Sandbox AuditReserved for a single 100% independent post-training lab passing blind 50-trajectory JSONL validation, ordinal Krippendorff's alpha >= 0.85 reproduction, eloquent-wrong canary testing, and Docker exit-code verification.
[02]
VERIFY_01 // TECHNICAL_AUDIT_GATES

Four-Gate Normative Post-Training Lab Inspection

Commercial sponsorship never overrides technical failure or weakens normative thresholds in `SPEC_STD RLHF-2026.11`. Every lab applying for Level 02 or Level 03 verification submits an anonymized 50-trajectory sample artifact inspected across four deterministic gates: corporate cap-table independence from competing foundation model developers, raw multi-rater rubric logs confirming ordinal Krippendorff's alpha >= 0.84, cryptographic keystroke telemetry schemas (`iki_cv`, `revision_ratio`, untrusted paste gap), and reproducible OCI container or formal kernel exit verification.

Vendors carrying > 5% equity ownership by a frontier model lab or failing anti-synthetic canary audits are ineligible for Featured or Verified Node status regardless of sponsorship budget.

[03]
VERIFY_02 // STATUS_REVOCATION

90-Day Re-Verification Cadence and Immediate Demotion Triggers

Verified post-training records undergo 90-day re-certification and face immediate reclassification if an acquisition introduces frontier-lab equity conflicts, if substantiated prompt/holdout leakage occurs, or if audited sample deliveries exhibit unflagged crowdworker LLM contamination.

Submit Record Verification or Featured Placement Inquiry

NORMATIVE CITATION STRING:Best RLHF Data Normative Specification. “Post-Training Lab Verification & Normative Benchmark Slot Protocol.” Revision November 2026. https://bestrlhfdata.com/sponsor-or-verify/