--- license: apache-2.0 pretty_name: typesafeAI/evalsafe-invoice-processing (mirror) source_datasets: - typesafeAI/evalsafe-invoice-processing tags: - laya - laya-finetune - mirror - domain-business - domain-finance - use-conditional - genre-other - label-choice - annot-llm - lang-en language: - en configs: - config_name: cases default: true data_files: - split: test path: data/cases.parquet - config_name: questions data_files: - split: test path: data/questions.parquet - config_name: run_results data_files: - split: test path: data/run_results.parquet --- # typesafeAI/evalsafe-invoice-processing(複製) ## データの内容 請求書の処理(invoice processing)の判断を評価するための、テスト専用のデータ。構成 `cases` では 1 行が 1 ケースで、`input_json` に完全な入力、`consensus` などにラベルセットが入る。参照ラベルは OpenAI と Anthropic のラベルを合わせた consensus で、モデルが生成したものであり人手による正解ではない。ほかに `questions` と `run_results` の構成があり、質問ごとの回答と実行ごとの採点を記録している。 ## データの概要 | 項目 | 値 | |---|---| | 行数 | 8,374(3 構成・3 分割の合計) | | 列数 | 6〜12(構成による) | | 容量 | 16.09 MB(ファイル 7 本) | | ファイルの形式 | parquet 3、json 1 | | 言語 | en | | 公開 | public | ### 構成と分割 | 構成(config) | 分割(split)ごとの行数 | 列数 | |---|---|---:| | `cases` | `test` 150 | 6 | | `questions` | `test` 6,874 | 10 | | `run_results` | `test` 1,350 | 12 | ### 列(構成 `cases`・分割 `test`。構成ごとに列が違う) | 列 | 型 | 先頭の行の値 | |---|---|---| | `case_id` | `large_string` | `ap_00000` | | `input_json` | `large_string` | `{"schema_version":"1.0","case_id":"ap_00000","co…` | | `metadata_json` | `large_string` | `{}` | | `openai` | `struct{status: large_string, model: struct{name: large_string, provider: large_string, reasoning_effort: large_string, question_mode: large_string}, used_opus_fallback: bool, decisions: list}` | `{status, model, used_opus_fallback, decisions}` | | `anthropic` | `struct{status: large_string, model: struct{name: large_string, provider: large_string, reasoning_effort: large_string, question_mode: large_string}, used_opus_fallback: bool, decisions: list}` | `{status, model, used_opus_fallback, decisions}` | | `consensus` | `struct{status: large_string, method: large_string, contributors: list, used_opus_fallback: bool, decisions: list}` | `{status, method, contributors, used_opus_fallback, …}` | 行数と列は HF のデータセットビューアの集計(元のデータセットの現在の版で数えたもの)。全データセットの一覧は [HYUKI5/dataset-catalog](https://huggingface.co/datasets/HYUKI5/dataset-catalog) にある。 ## 複製について [`typesafeAI/evalsafe-invoice-processing`](https://huggingface.co/datasets/typesafeAI/evalsafe-invoice-processing) をリビジョン `6beeb2d2acd65c086c835022f5f4d7434114cafc` の時点でそのまま複製したもの。中身は変えていない。判断モデル Laya(`convaiinnovations/laya`)の微調整に使う候補として、調査メモ(`Laya 学習用データセット調査`)に載せたデータを集めている。 ## 出典 - 元のデータセット: https://huggingface.co/datasets/typesafeAI/evalsafe-invoice-processing - 複製したリビジョン: `6beeb2d2acd65c086c835022f5f4d7434114cafc`(2026-09-29 16:00:56+00:00) - 元のライセンス表記: `apache-2.0` ## 論文 - 元のカードに論文へのリンクは見つからなかった ## アノテーションの方法 (元のカードに該当の見出しが無いので、ラベルの付け方に触れた段落を抜き出した) スナップショット: 2026-09-28。150 件のケースと 6,874 件の質問インスタンス。 既定の参照: **consensus**。ラベルはモデルが生成した参照である。 ## 微調整に使うときの注意 - 調査メモの台帳での区分: B(継承条件・gated・モデル由来の条件・上流の規約などが付く)。区分は記載の文言で機械的に付けたもので、法的な判定ではない。学習に入れる前に元のカードとライセンスの原文で確かめる。 - Laya の入力は 2,048 token まで(ModernBERT-large / mmBERT)。長い本文は先頭から黙って切られるので、切れる行は事前に除くか、要約・切り出しをしてから入れる。 - 分割は記事・文書単位で行い、同じ本文から作った別の問いを学習側と較正側に分けない。温度と閾値は学習に使っていない分割で別々に当てる。 ## 調査メモの記述 | メモ | 節 | ライセンス(記載) | 確認 | 注記 | |---|---|---|---|---| | `multitask_weak_synthetic.md` | Q1. Laya と同形式の合成・変換済み「typed decisions」データは、公開されているか (L40) | apache-2.0(invoice-processing、HF API。onet も HF API で apache-2.0。customer-service | 確認済 | 上流/条件注記 / 原文注記: 参照ラベルが LLM の合意で、人手ではない。test のみ | ## 元のカード(日本語訳) 以下は元のデータセットのカードの本文を日本語に訳したもの(機械翻訳)。訳に疑問があるときは、末尾の原文を確かめる。 --- # 請求書の処理 スナップショット: 2026-09-28。150 件のケースと 6,874 件の質問インスタンス。 既定の参照: **consensus**。ラベルはモデルが生成した参照である。 ## データ 構成 `cases`、`questions`、`run_results` のいずれかを読み込む。いずれも `test` の分割を持つ。 - **cases:** `case_id` ごとに 1 行。完全な入力は `input_json` に、説明的なメタデータは `metadata_json` に入り、`openai`、`anthropic`、`consensus` のラベルセットを持つ。判断は `policy_id` ごとにまとめられ、`status`、`actions`、`primary_action` を含む。 - **questions:** ケース、ノード、質問、入力の組み合わせごとに 1 行。`question_instance_id` で識別される。`question_json` と `state_json` はモデルの入力を表す。各ラベルセットには、`answer_json`、`probabilities`、`confidence`、`confidence_method` が含まれ、スコアの質問では `expected_score` も含む。`answer_json` は最頻回答である(同点の場合は null)。期待スコアは別の数値である。 独立したラベルには、実際のモデル、推論の努力度(reasoning effort)、質問のモード、`used_opus_fallback` が含まれる。欠けているラベルには、明示的なステータスと null の値が入る。ケースは残す。 質問のステータス `not_answered` は、その参照がその質問インスタンスに回答しなかったことを意味する。 ## 採点 `dataset.json` には、ポリシー ID と該当する比較の規則が列挙されている: - `exact_actions`: 引数を含むアクションの集合全体を比較する。順序と重複は無視する。 - `primary_action`: 指定された主要なアクションを、その引数を含めて比較する。 ## 実行結果 `run_results` には、最終的なプロットで選ばれた 9 件のコードルートの実行が含まれる。1,350 行あり、`run_id` と `case_id` の組ごとに 1 行。記録された質問への回答は 61,866 件である。`test` の分割には、書き出されたすべての行が含まれる。分割はしていない。 各行には、モデル名、提供元、推論の努力度、質問のモードが含まれる。さらに、実行の `status`、`cost_usd`、`wall_time_s`、`summed_call_time_s`、トークン数、呼び出し回数と、2 つの入れ子のリストを含む: - `questions`: `node_id`、`question_id`、`kind`、実際の `question_json` と `state_json`、`status`、`answer_json`、`probabilities`、`expected_score`、`confidence`、`confidence_method`。入力は繰り返し含めるので、各結果は自己完結している。回答は記録された実行の値である。`expected_score` は、スコアの質問について分布から計算される。記録された質問だけが現れる。選ばれなかった分岐は、作り出された回答を生まない。動的な選択肢は、参照の質問の表にないインスタンスを含めて、そのまま保持している。 - `decisions`: `policy_id`、`status`、`actions`、`primary_action`、`scores`。各スコアは `metric_id` と、**consensus** の参照に対する真偽値の `value` を持つ(該当しない場合は null)。指標は `exact_actions` または `primary_action` の比較だけを使う。 実行は、`cases` の中の consensus の採点対象ケースをそのまま使う。予測の誤りは採点に残り、誤答として数える。測定値はケースごとで、ポリシーの判断をまたいで共有される。欠けた測定値は null のまま残す。コストは、各実行の要約に記された基準を使って USD で記録されている。 `dataset.json` には `runs` が含まれ、`run_id` をキーとして、プロットされた集計スコア、分母、平均コストと平均時間、測定値の件数を持つ。モデルの構成について選ばれた実行は、プロットに使われたもの(主要な正解率が最も高いもの)であり、必ずしも最新の実行ではない。全体のプロット点は、4 つのワークフローすべてに存在する構成について、ワークフローを等しく平均したものである。 ## Consensus 質問単位の consensus は、同じ質問と入力について利用可能な確率分布の平均を取る。`contributors` と `weights` は、2 つの出典の場合はそれぞれ 0.5、単一の出典の場合は 1.0 を指定する。consensus の確信度は、混合後の確率の最大値である。 ケース単位の consensus には、混合したシグナルにワークフローを適用して得られた、最終的な参照のアクションが含まれる。これらは評価に使われる参照である。異なる動的な質問インスタンスは、質問の表の中で別々のまま残る。 ## 対象範囲 | Labelset | Cases with decisions | Opus fallback cases | |---|---:|---:| | openai | 150 | 0 | | anthropic | 150 | 0 | | consensus | 150 | 0 | ## エンコード `_json` で終わるフィールドは JSON でエンコードされたテキストである。`json.loads` でデコードする。確率分布は `option`/`probability` の組のリストである。アクションの引数は `arguments_json` に保持される。説明的なメタデータは、モデルに見える状態とは分けてある。 ## ライセンス このデータセットは [Apache License 2.0](LICENSE) のもとでライセンスされている。
原文(英語) # Invoice processing Snapshot: 2026-09-28. 150 cases and 6,874 question instances. Default reference: **consensus**. Labels are model-generated references. ## Data Load configuration `cases`, `questions`, or `run_results`; all have a `test` split. - **cases:** one row per `case_id`, with the complete input in `input_json`, descriptive `metadata_json`, and `openai`, `anthropic`, and `consensus` labelsets. Decisions are grouped by `policy_id` and contain `status`, `actions`, and `primary_action`. - **questions:** one row per distinct case/node/question/input combination, identified by `question_instance_id`. `question_json` and `state_json` describe the model input. Each labelset contains `answer_json`, `probabilities`, `confidence`, `confidence_method`, and `expected_score` for score questions. `answer_json` is the modal answer (null for ties); the expected score is a separate numeric value. Independent labels include the actual model, reasoning effort, question mode, and `used_opus_fallback`. Missing labels have explicit status and null values; the case is retained. Question status `not_answered` means that reference did not answer that question instance. ## Scoring `dataset.json` lists policy IDs and the applicable comparison rules: - `exact_actions`: compare complete action sets including arguments; ignore order and duplicates. - `primary_action`: compare the designated primary action including its arguments. ## Run results `run_results` contains 9 code-route runs selected by the final plots: 1,350 rows, one per `run_id` and `case_id`, with 61,866 recorded question answers. The `test` split contains every exported row; nothing is partitioned. Each row includes the model's name, provider, reasoning effort, and question mode; execution `status`, `cost_usd`, `wall_time_s`, `summed_call_time_s`, token counts, and call count; and two nested lists: - `questions`: `node_id`, `question_id`, `kind`, the actual `question_json` and `state_json`, `status`, `answer_json`, `probabilities`, `expected_score`, `confidence`, and `confidence_method`. Inputs are repeated so each result is self-contained. Answers are the recorded run values; `expected_score` is calculated from the distribution for score questions. Only recorded questions appear: branches not taken do not produce fabricated answers. Dynamic choice options are preserved exactly, including instances absent from the reference question table. - `decisions`: `policy_id`, `status`, `actions`, `primary_action`, and `scores`. Each score has `metric_id` and a boolean `value` against the **consensus** reference (null when inapplicable). Metrics use only `exact_actions` or `primary_action` comparisons. Runs use exactly the consensus scoring cases in `cases`. Prediction errors remain in scoring and count as wrong. Measurements are per case, shared across policy decisions; missing measurements remain null. Cost is recorded in USD using the basis named in each run summary. `dataset.json` contains `runs`, keyed by `run_id`, with plotted aggregate scores, denominators, mean cost/time, and measurement counts. The selected run for a model configuration is the one used by the plot (highest primary accuracy), not necessarily its newest execution. Overall plot points average workflows equally for configurations present in all four workflows. ## Consensus Question-level consensus averages the available probability distributions for the same question and input. `contributors` and `weights` specify 0.5 each for two sources or 1.0 for a single source. Consensus confidence is the maximum blended probability. Case-level consensus contains the final reference actions from applying the workflow to its blended signals. These are the references used for evaluation. Distinct dynamic question instances remain separate in the question table. ## Coverage | Labelset | Cases with decisions | Opus fallback cases | |---|---:|---:| | openai | 150 | 0 | | anthropic | 150 | 0 | | consensus | 150 | 0 | ## Encoding Fields ending in `_json` are JSON-encoded text; decode with `json.loads`. Probability distributions are lists of `option`/`probability` pairs. Action arguments are preserved in `arguments_json`. Descriptive metadata is separate from model-visible state. ## License This dataset is licensed under the [Apache License 2.0](LICENSE).