Shaduf.
Union Alpha/R5 final status and routine-monitoring closure
Research report · R5 · final routine status

Final status and routine-monitoring closure

Run run:62ac219d-07c9-40e4-821d-f07cdeec9e53 · final cutoff 2026-09-24T00:36:00Z UTC · bounded public, lawful, read-only primary-source sweep. Direct inference/access requests: n=0.

Final status
No reopening trigger was found. The former OpenRouter Union Alpha preview remains historical and retired. Its historical page says the preview was revealed as Pareto by Unbiased. The current catalog does not list stealth/union-alpha. The former Go alias is absent from the three inspected current Go surfaces. Current paid unbiased/pareto is separate and not shown to be a fixed continuation of either former alias.

Evidence gates

Evidence questionRequired evidenceR5 result
Former-preview attributionDirect first-party languageMet only for historical OpenRouter preview → Pareto by Unbiased. R5-S01
Former Go alias → ParetoDirect source language naming the Go alias and ParetoNot met. Go-source omission is not mapping evidence. R5-S05
Fixed equivalenceVersion, configuration, and date pin joining a former alias to ParetoNot met. Current catalog metadata and a canonical slug do not meet this gate; terms permit composition changes. R5-S02 · R5-S07
Benchmark resultStable raw artifact, fixed model/version/date, dataset/task IDs or n, harness/settings, and result/cost methodNot met for Union Alpha or an independently reproducible Pareto result. Provider claims and configurable harness remain separate. R5-S06 · R5-S08–09

No Pareto score can be called a Union Alpha benchmark without the fixed-linkage gate.

R4 → R5 decision table

FieldR5 decisionBoundary
Old-slug / re-listing statusUnchangedHistorical page says the free stealth period ended; public catalog has no stealth/union-alpha match. This is not an old-ID request result. R5-S01 · R5-S02
Former Go alias statusUnchangedNo alias match in inspected Go docs, public catalog, or changelog. Absence applies only to those sources. R5-S05
Explicit alias → Pareto mappingUnchanged / partialOpenRouter explicitly maps the former OpenRouter preview, not the former Go alias, to Pareto. R5-S01
Fixed Pareto version/configuration mappingUnchanged: none foundunbiased/pareto-20260917 is current route metadata, not mapping language; terms permit mutable composition. R5-S02 · R5-S07
Current Pareto route, price, declared limits/capabilities, and data termsUnchangedCurrent listing metadata and Unbiased terms are separately scoped. None transfers to former aliases. R5-S03 · R5-S07
Numeric route-specific quotaUnchanged: none foundLimits material is generic; no inspected source publishes a numeric cap for former aliases or Pareto. R5-S04
Immutable raw benchmark/result packageNot inspectable / no qualifying inspected artifactLanding and method files are configurable; release/results URLs had browser-cache misses. This does not prove absence. R5-S08–09
Independent reproducible resultUnchanged: none establishedThe provider card is provider-published; its n=30 comparison is preliminary and not independently validated. R5-S06

Route-specific rate-limit conclusion

RoutePublished route-specific numeric limit at cutoffDirect observationConclusion
Historical stealth/union-alphaNone found in inspected current primary sources0Unknown. Catalog omission is not a quota, error, or entitlement result.
Former OpenCode Go aliasNone found; alias absent from inspected current sources0Unknown. Do not reuse current Go estimates for other named models.
Current unbiased/paretoper_request_limits: null; no numeric quota found0Unknown. null does not mean unlimited. Generic policy is not a route-specific quota.

Exact R5 answer: no inspected current primary source publishes a numeric route-specific cap for either former alias or current Pareto, and R5 made zero direct requests.

Performance record boundary

Unbiased’s model card publishes a Pareto 26.9 headline slate: DeepSWE 74, Terminal-Bench 4.0 51, MMMU-Pro 78, HLE no-tools 49, and ArXivMath 88. It also publishes a preliminary DeepSWE 30-task point: 70.0%, USD 0.29/task, Sep. 20. The card says that point was screenshot-transcribed, not independently validated, and needs denominator and cost-method confirmation. The public pareto-evals repository provides configurable method code and documents local output; it does not provide a fixed former-alias mapping or qualifying immutable run package on the inspected surfaces. These are not Union Alpha benchmark results. R5-S06 · R5-S08–09

Decision and reopen triggers

Decision: close routine monitoring. Keep R1 through R5 as dated records. This is a status decision, not a new model test, access test, benchmark, or performance claim.
  1. R6 — fixed-linkage verification: a first-party source explicitly maps a former Union Alpha alias to a named Pareto version, configuration, and date.
  2. R7 — benchmark-artifact audit: an immutable raw Pareto result package or independently reproducible result appears with the required method fields.
  3. R8 — official re-listing/access observation: a former alias is officially re-listed or explicitly mapped to an active route, with a separately approved low-risk protocol and pre-existing authorized credential.
  4. R9 — material route-policy delta: a first-party source changes Pareto price, a numeric route-specific limit, retention/data terms, or retirement/replacement status.

Capture limits

Source ledger, capture notes, collection method, request count, and checksums are retained with the R5 run. Browser captures are excerpt and absence notes, not raw HTTP bodies. Release/results sub-surfaces were not inspectable because of browser-cache misses; that limitation does not establish absence.

Current pages: summary · availability · benchmarks · attribution. Earlier records: R4 · R2 · R1.

Search published pools, pages, reports, and evidence.