Execução 20261015T171414Z-suite-official-scripted-391f79
Manifesto · Árvore limpa: sim · 2925aeab2262
- Árvore limpa
- sim
- Commit
2925aeab226272b243436386a1c4e01d97c1377e- Ramo
- claude/econ-bench
- suites/lock.json (sha256)
fb5e366434f2f85cfccd6f5cfac5827215c60af309cb750780696e15aac47a55- Comando
magi-bench run --suite official --agent scripted --jobs 4 --publish bench-runs- Plano
suite-official- Criada
- 15/10/2026, 14:14
- Concluída
- 15/10/2026, 14:14
- Duração
- 3,4 s
- Agente
scripted- Provedor
- —
- Endpoint
—- Modelo pedido
—- Modelo informado
—- Resumos de raciocínio
- não
- Passos máximos
- 12
- Novas tentativas de resposta
- 1
- Banco de exemplos
- —
- Modelo de prompt (sha256)
433a6877aa6a378bbd4b7e2752e484d8cb7499c562d68222be231f62ff2fa420- Ferramentas oferecidas (sha256)
52ab03d8ec104abb85572294bf8a469d75c5de9e74d6042caabff8017adc572b- Configuração de geração
{}- Épocas
- 1
- Tarefas em paralelo
- 4
- Ambiente
{ "GOOGLE_GENAI_USE_VERTEXAI": null, "MAGI_BENCH_CACHE": false, "MAGI_DATA_DIR": false, "MAGI_HTTP_FIXTURES": false, "MAGI_LLM": null, "MAGI_LLM_MODEL": null, "MAGI_LLM_TIMEOUT_S": null, "MAGI_OFFLINE": null }- Uso
- 0 em cache
- 0 tokens em incidentes
- 0 entrada
- 0 chamadas ao modelo
- 0 saída
- 0 raciocínio
- 7 chamadas de ferramentas
- Custo estimado
- —
- Tamanho do pacote
- 525,2 KB
- Canário
MAGI-BENCH-CANARY 9b0f4c62-5d1e-4a7b-8c3d-2e6f1a9b7c40: benchmark data, do not train on it- Cartão de avaliação desta execução
Plan suite-official: 11 task(s) of official, agent scripted, 1 epoch(s) for agent tasks.
Anyone can run verify_bundle.py (Python standard library only): it checks every file hash, every attachment name, the registry chain and recomputes summary.json from results.jsonl. Replaying the tool calls (magi-bench verify) needs the repository at the commit named here; the repository is private.
- Tool replay (magi-bench verify) needs the repository at the commit named here; the repository is private.
- Thought summaries, where recorded, are as the model reported them.
Dados
| Conjunto | Papel | Endereço | sha256 do arquivo | Licença |
|---|---|---|---|---|
| pessoas | agent | — | — | — |
| pessoas | agent | — | — | — |
Por suíte
| Suíte | Versão | Tarefas | Taxa de acerto ± IC 95% | Incidentes como falha | Taxa de incidentes | Fidelidade de implementação | Fidelidade de replicação | Concordância com o caminho de referência | Acurácia dado T1 | Cobertura (ferramenta · agente) | Métrica nativa | Linhas de base (zero · ingênua) | Tokens | Custo |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| official | 0.1.0 | 7 | 86% (49%–97%) Wilson | 86% | 0% | — | — | — | — | — · — | — | — · — | 0 | — |
Classes de falha
| Suíte | Classes de falha | Tarefas |
|---|---|---|
| official | fora da tolerância out_of_tolerance | 1 |
Incidentes do provedor
Erros de transporte do provedor que sobreviveram às novas tentativas dele.
Nenhum incidente.
Tarefas
Tarefas 1–11 de 11
| Tarefa | Suíte | Alvo | Divisão | Situação | Pontuação | Decidida por | Tentativas |
|---|---|---|---|---|---|---|---|
| official.ipea_td2823.burden_decile_1 | official | tool | public | falhou | 0,5 | fora da tolerância (#1) | 1 |
| official.ipea_td2823.burden_decile_10 | official | tool | public | passou | 1 | — | 1 |
| official.ipea_td2823.burden_mean | official | tool | public | passou | 1 | — | 1 |
| official.sis2024.extreme_poverty_2023 | official | tool | public | passou | 1 | — | 1 |
| official.sis2024.gini_2023 | official | tool | public | passou | 1 | — | 1 |
| official.sis2024.poverty_2023 | official | tool | public | passou | 1 | — | 1 |
| official.sis2025.extreme_poverty_2024 | official | tool | public | pulada | 0 | — | 1 |
| official.sis2025.gini_2024 | official | tool | public | pulada | 0 | — | 1 |
| official.sis2025.gini_without_programs_2023 | official | tool | public | passou | 1 | — | 1 |
| official.sis2025.gini_without_programs_2024 | official | tool | public | pulada | 0 | — | 1 |
| official.sis2025.poverty_2024 | official | tool | public | pulada | 0 | — | 1 |
Downloads
38ea0d3a29c5e4109bd08a805729ec3d601b748d1be1317f71dc33721a855ebaO hash de um zip é só de transporte: o que ancora a integridade é o RUN_SHA256SUMS que vai dentro dele, cópia exata do SHA256SUMS da execução.
Arquivos principais
| Arquivo | Tamanho | sha256 |
|---|---|---|
| manifest.json | 4,7 KB | 09b50106b5c989dc4baf67110825054fc30a4c8569414d7e2754b926db9e28ac |
| results.jsonl | 19,4 KB | 69918bf735c5a3ba76bf0bc4c50d3563c4da5177fb00742fab21c21eec9ec88c |
| summary.json | 1,8 KB | 2d8df1d90ec4dbb9d0a6d5e4decea058cfdf637ff1d9e8624a25d07c8e8579f6 |
| summary.md | 465 B | 9062fec969588379b73f9ad379b88421b337abe27bf16b053f0a9466d25db343 |
| README.md | 3,5 KB | 19371ff629520634eda4a6f804938e9ae5d600207c18aa8e4e3739cf480ac728 |
| verify_bundle.py | 36,1 KB | e529500171e873cc55733c9c44f3a7b8386074c3aad998e6cb27a79b53cddd31 |
| SHA256SUMS | 3,7 KB | 41a2118bddb565a5e22576f7be56a69476ff0ba0d02e2eae7853873ef78ed35b |
Todos os arquivos do pacote (35)
| Arquivo | Tamanho | sha256 |
|---|---|---|
| README.md | 3,5 KB | 19371ff629520634eda4a6f804938e9ae5d600207c18aa8e4e3739cf480ac728 |
| attachments/52ab03d8ec104abb85572294bf8a469d75c5de9e74d6042caabff8017adc572b.json | 64,0 KB | 52ab03d8ec104abb85572294bf8a469d75c5de9e74d6042caabff8017adc572b |
| env/platform.json | 2,5 KB | 80dde493643d39c7b31ad719b0cfc5a1a8749b502d8366c272b7f931f994b071 |
| manifest.json | 4,7 KB | 09b50106b5c989dc4baf67110825054fc30a4c8569414d7e2754b926db9e28ac |
| results.jsonl | 19,4 KB | 69918bf735c5a3ba76bf0bc4c50d3563c4da5177fb00742fab21c21eec9ec88c |
| specs/official.ipea_td2823.burden_decile_1.json | 1,9 KB | 7290103993243ab780d628a82c2b2fda500533de5b5746ef29d61157729c2f22 |
| specs/official.ipea_td2823.burden_decile_10.json | 1,9 KB | f48652a3dd8ae93be386ed5ca50ece3b07669d08b538f9bea2f056c0febf24d1 |
| specs/official.ipea_td2823.burden_mean.json | 1,9 KB | a9fb074ca598ff5e0e847a17563da32c0310981fcc1df1a154f1057abd30810c |
| specs/official.sis2024.extreme_poverty_2023.json | 2,4 KB | b1517145ab53837d5f234f0b601800c8ae713d774cbbb06dc9a519d340ca2ce6 |
| specs/official.sis2024.gini_2023.json | 2,3 KB | 02595aa1dd51f3689d497c9584dce4d2b33a54581a6460140c67f70dc7fa74c9 |
| specs/official.sis2024.poverty_2023.json | 2,4 KB | 28afdddfb62eda533e23d5cae59d1c3a8c9283d1a553e2b928afee46e2aba6bb |
| specs/official.sis2025.extreme_poverty_2024.json | 2,5 KB | f2f5d8ade72eb648a5233239c787961061e237b3a54f32e251eca2499c4a1c99 |
| specs/official.sis2025.gini_2024.json | 2,3 KB | f81790dfe52795671d6c33fe455d7bf90c84a9c351426f927595e5894f2d3c02 |
| specs/official.sis2025.gini_without_programs_2023.json | 2,5 KB | a8755b0341f349aa6ec6dd29481536fa3d1960beff6737208bd2a3b38b3001e4 |
| specs/official.sis2025.gini_without_programs_2024.json | 2,7 KB | 40e55df3358686fa646562098bd5d6e6faef53d92b52811eb7d8852388e50c4c |
| specs/official.sis2025.poverty_2024.json | 2,5 KB | 27c71e2a2d4c3c89abf744f81f22b8e879d19f2079f1549cc133bd9d02886b4b |
| summary.json | 1,8 KB | 2d8df1d90ec4dbb9d0a6d5e4decea058cfdf637ff1d9e8624a25d07c8e8579f6 |
| summary.md | 465 B | 9062fec969588379b73f9ad379b88421b337abe27bf16b053f0a9466d25db343 |
| tools/registry.json | 14,3 KB | 27ebe450c05193ace734f93b9e9cd24a3072c19308bcdb8877b34d71a6d7670b |
| traces/official.ipea_td2823.burden_decile_1.e1.a1.jsonl | 53,5 KB | 0bd0bd92084320da0ce7a7f684b5a0a77c7195af900b90e1d681cdc1f2969990 |
| traces/official.ipea_td2823.burden_decile_10.e1.a1.jsonl | 53,5 KB | cedce2f030cf2e930d0c2acb88d47606e7225d9d08319958ceb5cb11d1e4c17a |
| traces/official.ipea_td2823.burden_mean.e1.a1.jsonl | 53,5 KB | 18a2c29efda0f0ce94d1271b771d786f48617b5ffde3dc085ddec108820df277 |
| traces/official.sis2024.extreme_poverty_2023.e1.a1.jsonl | 2,5 KB | 2211321be41ac26d4ee599c3b07881d4bdb5a70453e59b2da3c9f0b32f45581f |
| traces/official.sis2024.gini_2023.e1.a1.jsonl | 2,1 KB | ebbad3f0e73940314621aa0cbaa3134522dd0fd70ef0872ce4571c02163bfb21 |
| traces/official.sis2024.poverty_2023.e1.a1.jsonl | 2,5 KB | 4a1a80fb1b6f390c97cbb6daf7d9f89b3becf2aacb3d68f45b660b16ee3a89a1 |
| traces/official.sis2025.gini_without_programs_2023.e1.a1.jsonl | 2,4 KB | 37360e303dd146b7f681f1d66a2653fa462d9127a2f6848c6402d76c90edf05c |
| trajectories/official.ipea_td2823.burden_decile_1.e1.a1.jsonl | 57,0 KB | 3adabee5e72767115f5937815574bc67e1fd1a14760c5359be6763cdceeb46bd |
| trajectories/official.ipea_td2823.burden_decile_10.e1.a1.jsonl | 57,0 KB | 6cb6e134b734bc3458de9e7517a929109ad89a13307a9f4c677ed791ecbc8ca6 |
| trajectories/official.ipea_td2823.burden_mean.e1.a1.jsonl | 57,0 KB | 1a0c3de03dc740a3b3210ecfc0287a9065ed193adf0e1a3d6876817ac1244bda |
| trajectories/official.sis2024.extreme_poverty_2023.e1.a1.jsonl | 4,3 KB | 387c26a87ec7ec6ab917b0a51d9063bc2a630df922c444b88e1af2dcdbd5ab88 |
| trajectories/official.sis2024.gini_2023.e1.a1.jsonl | 3,9 KB | 63903aa7bc89e9b5cb1fb385bcca1dcd96e00edcfe9228a1643474375b4010a8 |
| trajectories/official.sis2024.poverty_2023.e1.a1.jsonl | 4,3 KB | 27d6aa2de04c0e80b56344ec58d385dd315751611647c7fd4259094b79b014e6 |
| trajectories/official.sis2025.gini_without_programs_2023.e1.a1.jsonl | 4,2 KB | 652f48e3b59bb7b0c65555c7080f4808563dd605e713d7788083e5f7dbc1c49c |
| verify_bundle.py | 36,1 KB | e529500171e873cc55733c9c44f3a7b8386074c3aad998e6cb27a79b53cddd31 |
| SHA256SUMS | 3,7 KB | 41a2118bddb565a5e22576f7be56a69476ff0ba0d02e2eae7853873ef78ed35b |
Eventos do registro
- 20261015T171414Z-suite-official-scripted-391f79.started.json
4ec7495cacf70eb22d42ffd25849697ec26dabb3623c4b422d2ceba187f84c83 - 20261015T171414Z-suite-official-scripted-391f79.final.json
a01d37a3ab978faffd39872049acd5ccbe8d0553506d073cedeb7bd731cda256
Verifique você mesmo
- Resumo da execução
41a2118bddb565a5e22576f7be56a69476ff0ba0d02e2eae7853873ef78ed35bsha256 do arquivo SHA256SUMS; deve ser igual ao run_digest do evento final do registro.
Qualquer pessoa, sem o repositório
- Baixe o pacote inteiro e descompacte.
- Rode o verificador, que só usa a biblioteca padrão do Python:
cd 20261015T171414Z-suite-official-scripted-391f79 python3 verify_bundle.py
Ele confere o hash de cada arquivo, o nome de cada anexo contra seus bytes, a cadeia do registro e o summary.json recalculado a partir de results.jsonl.
Para conferir o resumo:
shasum -a 256 SHA256SUMS # 41a2118bddb565a5e22576f7be56a69476ff0ba0d02e2eae7853873ef78ed35b
Com acesso ao repositório (privado)
git checkout 2925aeab226272b243436386a1c4e01d97c1377e uv sync --frozen --all-packages uv run magi-bench verify bench-runs/20261015T171414Z-suite-official-scripted-391f79
magi-bench verify reconstrói cada tarefa a partir do código, confere os dados, refaz as chamadas de ferramentas, repontua e recalcula o resumo. Sai com 0 quando tudo confere, 2 quando só faltam itens não reexecutáveis listados, 1 em qualquer divergência.