方法と受け入れ試験(UAT)

流れ

記事 → 推進力の信号(Jev)→ 月ごとの割合と勢い → インサイトのカード(Claude が解釈、Jev が裏付けを判定)→ 1・3 年は数値の延長、5・10 年はシナリオ → 寄稿文(Claude)→ Jev の 8 基準で審査し、不合格は書き直し → 公開 → 台帳で期日に採点。

UAT の結果 2026-10-09 09:15

ID試験値水準(1〜5)判定
U1データの網羅0.9134 / 3合格
U2信号の正しさ(F1 の平均)0.8565 / 3合格
U3収集量の影響(|r| の最大)11 / 3不合格
U4過去での的中(3 か月先の向き)01 / 3不合格
U5カードの根拠(Jev)01 / 3不合格
U6寄稿文の質(Jev 8 基準の最低)01 / 3不合格
U7地平間の一貫性(Jev)01 / 3不合格
U8頁とメニュー(満たした数 / 4)11 / 5不合格

寄稿文の審査の尺度(Jev の 5 段階)

基準12345
根拠性Claims cite no insight cards or dataFew claims are tied to cards; most are assertionAbout half the claims cite cards or indicatorsMost claims cite specific cards or indicator valuesEvery substantive claim cites a card number or indicator value
具体性Only vague generalitiesMostly general, a few concrete itemsSome concrete actors, numbers or datesMostly concrete actors, mechanisms and magnitudesConcrete actors, mechanisms, magnitudes and dates throughout
反証可能性Nothing could be checked laterOne loosely checkable statementA few checkable statements without thresholdsSeveral checkable predictions with dates or thresholdsClear indicators, thresholds and dates that would confirm or refute the forecast
地平間の一貫性Contradicts itself or the dataSome internal tensions left unexplainedMostly consistentConsistent, with tensions explainedFully consistent and explains how this horizon follows from the shorter ones
不確かさの表し方States the future as certainRare hedges, no probabilitiesSome probabilities or rangesProbabilities or ranges for the main claims and alternativesExplicit probabilities summing sensibly, ranges, and what would change them
新規性Only restates common headlinesMostly common viewsSome non-obvious connectionsSeveral non-obvious cross-domain connectionsDistinctive, data-driven insight not found in typical commentary
読みやすさHard to followDense or disorganisedReadable with effortClear and well organisedVery clear, plain language, easy to follow
均衡One-sidedMostly one-sidedMentions both opportunities and risks brieflyWeighs opportunities and risksWeighs opportunities, risks and who gains or loses, fairly