SecondRun / Post-release observation

What does the medal establish?

ClusterMAX 3.0 is released. This ledger records 17 selected provider assignments and the scope of the evidence reviewed. It is not a 77-provider audit or a customer-outcome result.

What the report actually tests

The inspected article describes managed clusters, configuration audits, compute/network/storage tests, training and inference workloads, extended load, and fault recovery. It also uses customer feedback and separates short hands-on observations from longer-run questions. The published testing is real evidence; this challenge asks a different, narrower question.

What our test adds

Hold the workload, ordinary customer treatment, baseline and mapping fixed before new jobs. Compare the baseline against the same predictor with the medal added, including a medal-blind reference adjustment. Retain all terminal attempts and report uncertainty across providers. No real prospective outcome packet has been submitted.

The report explains different recovery expectations for different hardware generations. A useful comparison therefore carries the actual service, hardware, permissions, support and observation period. A provider logo alone does not encode those conditions.

What remains unresolved in this review

The ledger records partial global scope and test conditions; exact provider-specific dates and service configurations have not been normalized. An out-of-sample comparison with a frozen specs-and-price baseline was not identified in the reviewed sections. Provider appendix entries are not individually adjudicated here. These are limits of this review, not claims that evidence exists nowhere.

Complete 77-provider transcription · Changes from 2.1

Selected release assignments

Transcribed from the supplied release graphic and cross-checked against named changes in the announcement. Seventeen selected entries, not the full roster.

Provider3.0 classificationScorer treatment
CoreWeavePlatinumEligible tier; a real plan and outcomes are still required
NebiusPlatinumEligible tier; a real plan and outcomes are still required
OracleGoldEligible tier; a real plan and outcomes are still required
Google CloudGoldEligible tier; a real plan and outcomes are still required
AzureSilverEligible tier; a real plan and outcomes are still required
LambdaSilverEligible tier; a real plan and outcomes are still required
FirmusSilverEligible tier; a real plan and outcomes are still required
TensorWaveSilverEligible tier; a real plan and outcomes are still required
GMISilverEligible tier; a real plan and outcomes are still required
CrusoeBronzeEligible tier; a real plan and outcomes are still required
HyperstackBronzeEligible tier; a real plan and outcomes are still required
DigitalOceanBronzeEligible tier; a real plan and outcomes are still required
VultrParticipation RibbonOutside original transform
RunpodParticipation RibbonOutside original transform
FluidstackUnavailableNot a quality score
ScalewayUnavailableNot a quality score
OVHcloudUnderperformingEligible tier; a real plan and outcomes are still required

Per-provider, per-dimension review status

From claims-3.0.observed.json (schema secondrun.launch-claims.v2), read directly -- no status below is asserted anywhere else. Every cell carries a hover title with the exact limitation text recorded for that field; hover or inspect the source JSON for the full wording. PARTIAL means a value is supplied but this scoped review has not normalized it to the provider-specific claim named in the limitation; NOT_REVIEWED means this ledger did not extract or verify the field; NOT_IN_REVIEWED_SOURCES means the reviewed sections do not supply it. None of these three statuses assert that the missing material does not exist elsewhere.

SUPPLIED cited value, fully reviewedPARTIAL value supplied, not fully normalizedNOT_REVIEWED not extracted/verified hereNOT_IN_REVIEWED_SOURCES absent from the reviewed sectionsOMITTED claim explicitly not evaluable
ProviderMedalScopeObservation conditionsEvidenceTest datesPredictive check
CoreWeavePlatinumPARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
NebiusPlatinumPARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
OracleGoldPARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
Google CloudGoldPARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
AzureSilverPARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
LambdaSilverPARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
FirmusSilverPARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
TensorWaveSilverPARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
GMISilverPARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
CrusoeBronzePARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
HyperstackBronzePARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
DigitalOceanBronzePARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
VultrParticipation RibbonPARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
RunpodParticipation RibbonPARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
FluidstackUnavailablePARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
ScalewayUnavailablePARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES
OVHcloudUnderperformingPARTIALPARTIALPARTIALNOT_REVIEWEDNOT_IN_REVIEWED_SOURCES

Citations: every cell above except the two Unavailable-tier Scope cells is sourced to the 3.0 article (Results, Scope, testing methodology, reliability and agentic-coding sections), retrieved 2026-09-23T21:54:18Z. The two Unavailable-tier Scope cells (Fluidstack, Scaleway) are sourced to the launch announcement, retrieved 2026-09-23T21:54:17Z. Medal-table source (the release graphic): retrieved 2026-09-23T22:03:49Z, hash in the ledger JSON. Rubric source: NOT_VERIFIED_FOR_3.0 -- no rubric binding is possible from this ledger yet.

Version changes stay visible

Participation Ribbon is a new ordinal category. The original five-tier transformation remains unchanged and refuses it. Unavailable is recorded separately from quality. A future six-tier model must be identified as a new, post-release choice. The exact 3.0 grading rubric is not yet verified in this ledger; the article is a methodology source, not a substitute for unobserved rubric bytes. Historical R1 inputs and results stay unchanged; the readout explains their limits.

No pre-medal registration of a real study is claimed. A design frozen now can still predict future jobs. Code publication, study registration, rating binding and outcomes are separate records.

Disclosure

Hot Aisle provided $200 in compute credits for my own testing.

Sources inspected

Primary release ranking graphic

3.0 article (Results, Scope, testing methodology, reliability and agentic-coding sections) · Launch announcement · Response hashes and retrieval conditions · Ledger contract.

The fetched homepage still showed older content; it was not used to contradict the announcement. Exact source bytes remain in author-controlled custody; the full copyrighted article is not republished.