BuilderBuilder
Versioned benchmark collection

Red Sands v1

Harness v1 · Builder projects v1 · 4 measured runs across 1 target. Each entry is a single recorded attempt; repeat-run reliability is not established.

Inspect the run ledgersOpen Builder reproduction← All collections
Western frontier town with a player character and saddled horse in Red Sands v1
Western frontier town with a player character and saddled horse in Red Sands v1
4Measured runs
1Frozen targets
4Model families
4Configurations
Collection scope

Recorded evidence, without a synthetic score.

Runs share the same frozen target and prompt within each test, with disclosed provider-specific stacks. Costs are recorded estimates as labelled in each ledger, not subscription invoices.

Targets

  • Red Sands v1

Model coverage

  • GPT-5.6 Sol · Max
  • Kimi K3 · Unreported
  • Claude Opus 5 · Max
  • Qwen3.8 Max · xhigh

Evidence available

  • Four selected measured run ledgers
  • Exact frozen prompt identity
  • Per-run quest source and generated voice artifact identities
  • Four reconstructed projects with clean install, production build, and browser smoke receipts
  • Prompt SHA-256 7a2e77ceec4154ceac38bf0293d7b10f3e89857bd44a60894d9fbc956898a859
  • Recorded prompt path benchmarks/fixtures/red-sands-quest-threejs.v1/PROMPT.md

Still missing

  • Human start-to-finish playthrough for every run
  • Blind evaluation and quality scoring
  • Repeat-run reliability evidence
Public receipts

4 measured run ledgers

Open any row for its recorded workflow cost, timing, token usage, stack disclosure, artifact integrity, and explicit evidence gaps.

01GPT-5.6 SolMax reasoning · Red Sands v1artifact_operator_static_checks_ledger_no_playthrough_evidence · Prompt version v1$19.89 API-equivalent estimate from official OpenAI API Standard pricing for the measured token mix; NOT the actual subscription-backed Codex chargerecorded estimate1h 0m 43.1s wall-clockEnd-to-end workflowOpen 02Kimi K3Unreported reasoning · Red Sands v1artifact_model_playthrough_complete_ledger_partial_window_metrics · Prompt version v1≥$5.68 Independently recomputed API-list-price equivalent for the measured token mix; not an itemized subscription charge (lower bound)recorded lower bound1h 0m 6.6s (measured window only) wall-clockEnd-to-end workflowOpen 03Claude Opus 5Max reasoning · Red Sands v1artifact_operator_static_checks_ledger_no_playthrough_verification · Prompt version v1$129.47 First-party Claude API list-price equivalent for the measured token mix; not an itemized subscription chargerecorded estimate1h 30m 54.0s wall-clockEnd-to-end workflowOpen 04Qwen3.8 Maxxhigh reasoning · Red Sands v1artifact_model_playthrough_complete_ledger_partial_metrics · Prompt version v1¥235.23 Independently recomputed first-party API-list-price equivalent for the measured token mix; NOT a billed amount. This run used an Alibaba Model Studio token-plan subscription, under which usage is not itemized as cash. (upper bound)recorded upper bound1h 5m 52.3s wall-clockEnd-to-end workflowOpen
Prompt provenance

Exact prompt receipts

A path is shown only when the published record includes the prompt hash that identifies it. Campaign evidence and projections keep the hash gap visible.

Separate records

Related records with unknown prompt provenance

These published records stay separate from the prompt receipts above. Unknown provenance is why they are listed here.

No separate unattached records are published for this collection.

Limits

Evidence and material gaps

  • Repeat-run reliability evidence is not established.