クイック回答
6 controlled steps：decision contract、eligibility/response coverage、segmentation前のidentity protection、frozen specificationによるscore reproduction、uncertainty付きdistribution/trend/subgroup/comment interpretation、owner/follow-up付きtested action。response rateはparticipationでrepresentativeness証明ではなく、averageはcompressionでworkforce voiceそのものではなく、themeはcoded signalでdiagnosis/critic identificationの許可ではない。
resultを開く前にsurvey analysis contractを定義
permitted purpose
one named decision/learning question、affected process、owner、prohibited use。
population and instrument
eligibility date、roster logic、survey version、language、channel、window、response unit。
analysis specification
item coding、denominator、missingness、composite、weight、comparison、uncertainty、suppression。
release and action
audience、access、reviewer、disclosure、correction、action owner、follow-up、retention、deletion。
evidence-led pulse survey analysisの6 steps
順番に実行。later dashboardではundefined population、unsafe privacy promise、undocumented scoring ruleを修復できない。
decisionとanalysis contractを定義
business question、eligible population、collection window、instrument version、response unit、approved comparison、action owner、review date、privacy promise、permitted/prohibited useを明記。
- release evidence
- resultを見る前にpopulation、denominator、wording、reporting rule、owner、useを再現可能。
- failure to prevent
- decisionなしのtemperature check、versionなしのmid-wave変更、sentimentのindividual performance・discipline・attendance・health inference転用を禁止。
frame・invitation・response coverageを検証
eligible roster、exclusion、delivery、start、partial、complete、duplicate、late、missing groupを照合。named numerator/denominatorでrateを算出しnonresponse集中をunit、shift、location、tenure、channel、language、accessで確認。
- release evidence
- frame date、eligibility logic、disposition count、rate formula、coverage gap、weighting decision、uncertainty noteをrelease recordへ。
- failure to prevent
- high response rate=representative、異なるeligibility group比較、synthetic response補完、weighted/unweighted混同を禁止。
segmentation/text analysis前にidentityを保護
direct identifier、quasi-identifier、free text、timestamp、rare role、small team、intersectionをmap。approved access、minimum reporting、suppression/aggregation、secure review、retention/deletion、re-identification testをdashboard/export前に適用。
- release evidence
- audience、smallest reportable unit、suppressed cell、redaction、export control、retention、reviewer、residual riskを各viewに表示。
- failure to prevent
- 実現不能なanonymity約束、tiny denominator/distinctive commentで特定、raw commentのmanager default送付、name削除=de-identificationを禁止。
reproducible specificationでscore
item ID、wording、option、scale direction、reverse key、valid denominator、missing/NA、composite membership、minimum coverage、weight、rounding、version crosswalkをfreezeしhand caseでcode検証。
- release evidence
- second analystがapproved raw extractから全item/indexを再現し同じvalue、state、version、exception logを得る。
- failure to prevent
- approved interpretationなしのordinal average、neutral=negative、skip=zero、crosswalkなしのscale混合、proseからAI score推定を禁止。
distribution・trend・subgroup・commentを統合分析
headline average前にcount/distribution。compatible period/groupのみ比較しdenominator、uncertainty、coverage change、contextを表示。commentはdocumented taxonomy、sample evidence、contrary example、reviewer agreement、redactionでcode。
- release evidence
- findingごとにmeasure、population、period、n、method、comparison、uncertainty、support/contrary evidence、privacy state、interpretation、owner。
- failure to prevent
- tiny team ranking、changeからcausation、comment cherry-pick、silence=agreement、model confidence=statistical certainty、多数cutからalarm選択を禁止。
findingをaccountable action/learning loopへ
evidence boundary、affected process、employee input、option、selected action、owner、resource、due date、success/harm measure、communication、escalation、cadence、stop/correction ruleを記録しrespondentを露出せずreport back。
- release evidence
- heard/not concluded、change、owner、follow-up、impact test、correction/concern routeをemployeeが確認可能。
- failure to prevent
- dashboard=action、one pulseからoverpromise、low-score group処罰、critic特定、inconvenient result抑制、comparable follow-upなしのimprovement claimを禁止。
finding stateを区別
measured estimate
population、denominator、period、method、uncertainty付きreproducible value。
directional signal
coverage、sample size、design、context、disagreementでinferenceが限定されるsignal。
not reportable
privacy、minimum-cell、access、quality、compatibility ruleによりこのlevelでrelease不可。
unresolved or corrected
source、coding、denominator、privacy、interpretation、action disputeがopenまたはversioned correction済み。
worked example：lower scoreはまだcauseではない
hypothetical pulseはeligible 240、complete 168でcomplete response rate 70%。approved clarity itemはcompatible prior wave 72% favorableから61%へ11 percentage-point difference。redacted comments 46件がpriority changeに言及。7-person teamは3 responseでsuppressed。leadership causationや全employee dissatisfactionは証明しない。
unsafe conclusion
「leaderのpriority changeでengagementが11 point低下、lowest teamをfix」ではcoverage、uncertainty、denominator、privacy suppression、item scope、contrary comment、alternative explanationを隠す。
evidence-led action
clarity itemを両wave denominator/method付きdirectional signalとしてreport、small teamはnot reportable。coded support/contrary commentをreview。30日間single priority owner/weekly change logを公開しunchanged itemでfollow-up、coverage/operational outcomeをtest、correctionをinviteしharm/no improvementならstop/revise。
OpenMaxによるsurvey evidence/action coordination
OpenMaxはapproved survey contract、roster snapshot、invitation disposition、consent/confidentiality notice、permissioned extract、scoring version、suppression check、reproducible table、comment redaction/coding、independent review、action assignment、employee feedback、correction、follow-up、monitoringをcoordinate。purpose、employment authority、statistical judgment、privacy/legal approval、sensitive handling、interpretation、resource、stop decisionはhuman。
privacy・employment・statistical・trust boundary
- qualified local HR、ER、privacy、security、legal、labor、accessibility、statistics、research、localization、records、AI-governance ownerがactual useをapprove。
- architecture、contract、access、threshold、free-text、export、retention、incident responseで実現できないanonymous/confidential/de-identified promiseを禁止。
- qualified ownerがlawful、necessary、disclosed processをapproveしない限りsurveyをindividual employment decisionやprotected complaint/accommodation channelから分離。participation、nonparticipation、criticism、correction、protected activityへのretaliation禁止。
- response rate、favorable score、average、theme、model confidence、benchmark、dashboard、manager approvalはrepresentativeness、causation、fairness、privacy、improvement、validityを証明しない。actual outcomeをtestしharmful useをstop。
情報源、編集方法、制限
OpenMax編集部はOPM FEVS explanation、technical report、result interpretation、U.S. Census questionnaire construction、NIST de-identification/re-identification riskを確認し6-step workflow、evidence state、240-invite hypothetical exampleを独自構成。2026年9月3日再確認。
- U.S. OPM — About the Federal Employee Viewpoint Survey
- U.S. OPM — 2024 FEVS Technical Report
- U.S. OPM — Understanding results
- U.S. Census Bureau — Survey Questionnaire Construction
- NIST SP 800-188 — De-Identifying Government Datasets
よくある質問
十分なresponse rateは？
universal cutoffなし。formula/disposition、missing group/reason、design/weighting、uncertaintyを示す。high rateでもsystematic nonresponseはあり得る。
small teamへscoreを出す？
approved privacy/quality ruleで許可される場合のみ。otherwise suppress/aggregate、raw access制限しnot reportableはperformance judgmentでないと説明。
AIでopen-text commentをsummary？
validated taxonomy/controlled access下でassist可能だがhumanがredaction、sample fidelity、contrary comment、uncertain classification、sensitive routing、release languageをreview。
score changeはintervention効果を証明？
no。causal claim前にinstrument/population compatibility、coverage、uncertainty、concurrent change、implementation fidelity、operational outcome、harm、alternative explanationを確認。
OpenMaxで安全にautomateできることは？
approved input、calculation、privacy gate、evidence-linked finding、review、action ownership、feedback、correction、follow-upをcoordinateしpurpose、statistical judgment、sensitive handling、employment authorityはqualified people。