Nine-test comparison
Public benchmark resultUnity · one frozen request

Shrine Exploration v1

The environment shows the upside of a seven-hour long-horizon Claude Opus 5 Max run. Animation, character quality and collision remain material weaknesses.

Visual acceptance failedThe failed gate does not suppress this public result.
Shrine Exploration — Claude Opus 5 Max presentation recording poster
Exact Launch 004 capture · audio removedshrine-approach-unity-opus-5-max
Final visual score
14 / 40
Wall-clock
7h 41m 36.5s
Recorded tokens
328,930,377
Cost boundary
$204.17+
Public visual review

12 → 14 → 14 / 40

Three recorded reviews improved the result once, then held at 14. The final state remained below the visual acceptance gate.

  1. Review 112/ 40
  2. Review 214/ 40
  3. Review 314/ 40
What passed

A public benchmark record, three release-player identities and the captured environment exist. Functional or packaging evidence is not a substitute for visual quality.

What failed

The final visual review identified material animation, character-quality and collision debt. Visibility does not turn the 14/40 into a pass.

Cost, token and method boundary

Exact recorded values before conclusions.

One-request workflow

Claude Code · Opus 5 Max

One frozen request, seven separately billed subagents, internal retries allowed, and no human follow-up or manual correction after submission.

Parent session

327,042,040

Recorded parent-session processed tokens.

Subagents

1,888,337

Recorded subagent tokens. Their input, cache and output mix is unavailable.

Qualified cost

$204.17+

Parent-session Claude API list-price equivalent plus an unknown additional subagent amount; not a Claude Code invoice, subscription charge, or exact complete run cost.

Reliability boundary: This single run does not establish a reliability rate.

Fixed benchmark input

The Samurai package was supplied unchanged.

The Shrine run remained one request. The samurai was a separately completed AI-generated fixture—produced iteratively with Opus 5, Meshy and Blender—and supplied unchanged. Its production time, tokens and cost are not counted in the Shrine run receipt.

It is an included RemakeBench benchmark fixture under a benchmark-use scope, not contestant output or an unrestricted standalone asset-store download.

SAMURAI CHARACTER FIXTURE V1
  • 53-bone rig
  • 6 authored animation clips
  • Katana + saya
  • 5 PBR texture maps
  • 15 files · 80,410,681 bytes
Fixed input · not contestant output

Builder fulfillment verified: all 15 fixture members are present; exact bytes or privacy-sanitized path equivalents are hash-bound to the approved Unity source.

Inspect the Samurai fixture provenance
Public player identities

Three exact archives. Three different evidence states.

All three public archives passed streamed storage read-back. The WebGL build also has a verified hosted-startup path; macOS evidence still does not transfer to WebGL or Windows.

macOS ARM64Public artifact · download-eligible
Hosted · read-back passed

shrine-approach-v1-macos-arm64.app.zip

Exact bytes
190,194,015
SHA-256
f045ee61380dfe18d144d13827758cd80f89bc066f52be50b684b76b0078dd81
Gameplay
Verified
Physical input
Verified
Audible output
Verified
Evidence basis
operator-attested
Trust statead-hoc signed, not notarized

Gameplay, physical input, footsteps and ambience are operator-attested. The app is ad-hoc signed, is not notarized, and macOS may require right-click → Open.

WebGLPublic artifact · download-eligible
Hosted · read-back passed

shrine-approach-v1-webgl.zip

Exact bytes
157,918,695
SHA-256
aa518976838751c4a9a1bc32d5357c369ab2c36929d99f3e63760bc07f4def4b
Gameplay
Not verified
Physical input
Not verified
Audible output
Not verified
Evidence basis
rendered startup only
Trust statebrowser-sandboxed, unsigned

Browser startup verified. Gameplay, physical input and audible output are not yet verified on WebGL.

Rendered startup, WebGL 2.0, and zero recorded console errors are verified. Gameplay, physical input and audible output are not yet verified on WebGL.

Windows x86-64Public artifact · download-eligible
Hosted · read-back passed

shrine-approach-v1-windows-x86_64.zip

Exact bytes
198,547,544
SHA-256
3c07e2df476e7650a3d0cbf2fa2c53e22b62ad496d5e6351f01fb64f1a86602e
Gameplay
Not verified
Physical input
Not verified
Audible output
Not verified
Evidence basis
PE32+ GUI x86-64 structure only
Trust stateunsigned; Windows runtime untested

The genuine PE32+ x86-64 player was structurally verified but not launched on Windows. Gameplay, physical input and audio remain unverified, and the executable is unsigned.

Licence and exclusion boundary

Redistributable project assets are cleared. Five references stay private.

The Builder source projection may include all redistributable project assets. It deliberately excludes the five internal concept references below.

Cleared asset classes

  • character, equipment, rig and animations
  • Meshy-generated Shrine textures
  • ElevenLabs-generated and derived audio
  • Opus/Blender-authored environment and props
  • Unity package dependencies
  • fonts, sky and custom shaders; no font files were found

Named private exclusions

  1. reference/A_hero_view.pngSHA-256 bfff498e9929f0bdfc27ec65cd3dbc21490bd2996894358801bc092c3ae587fc
  2. reference/B_overview_aerial.pngSHA-256 aef9a04beed4f1fdd2b657d04d5f762efd4f4ace6a817f8769647a888765a52b
  3. reference/C_sheet_a_hero_reverse_overview.pngSHA-256 32754deccfb381d599eb6030f5646629779faefe07e96746688451abd402b1d0
  4. reference/D_sheet_b_elevations_details_materials.pngSHA-256 db2051803a5ec3c5cf3a90fb78bdf16f024e15814056795dde827a20ffd966a9
  5. reference/E_detail_eave_paving_materials.pngSHA-256 bea2b890469573dcf4a8e23178956e1bb7692a62eb355df089f2394b0fc5ad0e

Exclusion proof · 373f279b1f61fbe5c1ef85a1fe3aa062a3eae5c66028c06408082b601e332b29

Public evidence · exact frozen receipt

Inspect the manifest before the package.

The public snapshot exposes checksummed result, player, licence, exclusion, fixture, prompt and test records. The exact prompt preserves 7 home-relative path examples. Local paths are preserved because this is the exact submitted prompt. The portable fixture binding resolves the supplied Samurai package for reproduction. Raw operator evidence, private object-store credentials and internal release material stay out.

Activated · owner-approved, storage-verified. Activated from owner approval and exact storage read-back. Ordinary-customer preactivation retrieval was waived and not performed; no customer-retrieval receipt exists.

Benchmark record SHA-256
1c37028cc513331bdbd86e3408f43fbcd01a07dd5e250dd7d86fbd3b5e9f8dc0
Prompt SHA-256
7f557c5a9cbdcd6d33565cc7c9262f63e4055abc8d507f3f2d704e6a396550ed
Verified resultPublic result
Claude Opus 5

Shrine Exploration v1 — Claude Opus 5 Max

Build a walkable third-person Unity shrine approach with a complete environment, character, rig, animations, audio, collision, validation, and release output in one frozen long-horizon request.

Max reasoningHeadline result
Workflow cost
$204.17+
Wall-clock
7h 41m 36.5s wall-clock
Processed tokens
327,042,040 parent processed + 1,888,337 subagent tokens; subagent token mix unavailable
Record state
verified
Public summary

A 7h 41m 36.5s one-request Unity workflow with 328.93M recorded parent-plus-subagent tokens, a $204.17+ parent-session API list-price lower bound, and a visible 14/40 visual-acceptance failure.

Cost basis
  • Only the parent session has the token mix required for exact API list-price arithmetic.
  • Seven separately billed subagents used 1,888,337 tokens, but their input, cache and output split is unavailable.
  • The exact complete run cost is unknown and higher than $204.17.
Visible evidence gaps
  • public player hosting and signed-out read-back verification
  • WebGL gameplay, physical-input and audible-output verification
  • Windows runtime, gameplay, physical-input and audio verification
  • complete subagent token mix and exact total API-equivalent cost
  • repeat-run reliability evidence
Separate harness record

Builder access opens the available full ledger for this run. This record is not part of either currently published immutable prompt package, so no exact prompt or package export is implied.

RemakeBenchResearch console