要点
eligible population定義後100 ticketをcensus/documented sample。minimum necessary redaction、traceable issue unit、audience向けsearch replay、5 primary outcome、reusable task/resolution cluster、counterexample review、specific action priority、release後measure。
5 knowledge coverage outcomes
| outcome | evidence | typical action |
|---|---|---|
| covered and found | accessible article found / answer adequate | reuse/monitor/flag or fix |
| covered but not found | adequate article、findability failure | discoverability/metadata/link/search/translation |
| partial or outdated | relevant but incomplete/inaccurate/stale/inaccessible/version-wrong | update/split/merge/validate/translate/restrict |
| missing article | no suitable reusable knowledge after search/review | create with owner/audience/evidence/acceptance |
| not suitable for self-service | identity-specific/live/dangerous/regulated/specialist | agent-assisted; safe routing/preparation only |
gap判定前の3 validity test
selection validity
別analystが100 eligible/selected ticketを再現可能か。不可ならfrequency/coverageは不安定。
search validity
audience、query、permission、language、index、result、article version、answer test記録。one failed queryでは不足。
decision validity
actionがreusable needを解決しcounterexample、risk、ownership、effort、acceptanceをreviewしたか。volume≠priority。
100 support ticket分析の9 steps
later judgmentはscope/privacy/unit/search evidenceに依存するため順番に実行。100はauditabilityを高めるがsampling error/biasを除去しない。
decision・population・review windowを定義
method
分析がcreate、improve、merge、archive、re-index、translate、restrictの何を決めるか明示。例を見る前にeligible channel、product、language、customer segment、ticket state、fixed received/resolved windowを定義。
evidence
decision question、population、included/excluded channel、product/version、language、ticket state、start/end/timezone、owner、selection rule、blind spot。
acceptance gate
source systemからscope再現可能。excluded/unavailable channelをlimitationとして残しdefensible sampling basisなしに100 selected recordをall customerへ一般化しない。
cherry-pickせず100 ticket選択
method
scope該当が100ならcensus、その他はdocumented representative/stratified sample。selection order/inclusion probabilityを保持。escalation、long ticket、recent failure、easy example、記憶ticketだけを選ばない。
evidence
population count、sampling frame/hash、method、random seed/order、strata/allocation、replacement、duplicate、reason、restricted crosswalk ticket ID。
acceptance gate
exactly 100 distinct eligible ticketをaccount。少ない場合理由。known dimensionでsample/population比較しimbalance開示。result後のweighting創作禁止。
sensitive ticket dataをminimize・redact・separate
method
knowledge questionに必要なfieldだけのanalysis view。name、contact、credential、payment、secret、health/employment、security artifact、不要free-textをremove/tokenize。re-identification keyは別authorized system。
evidence
field inventory/purpose、data owner、basis、redaction rule、access role、crosswalk、retention/deletion、quality check、security/privacy route。
acceptance gate
analyst/modelはminimum necessary view、redaction sample test、source unchanged、restricted caseはspecialist reviewしnarrativeをcluster/articleへ漏らさない。
requester languageを保ちissue normalize
method
multi-issue ticketをanswerable issue unitへsplit。original symptom wordsとnormalized conceptを分離しproduct、version、environment、task、expected/observed、resolution、substantiationを記録。similar wordingだけでmergeしない。
evidence
ticket/issue-unit ID、original phrase、normalized intent/entity、product/version/environment、expected/observed、resolution/workaround、verified root cause、tag、model version、rationale、confidence/counterexample。
acceptance gate
unitはticketへtrace、customer wording保持、separate issue分離、uncertainty明示。low-confidence/multilingual/specialist/high-impactはhuman review。
各issueでknowledge searchをreplay
method
relevant timeにintended user/agentがaccess可能なsurfaceをsearch。exact query、filter、language、role、product context、index version、ranked result、opened article、findabilityを記録。customer phraseとinternal term双方test。
evidence
search actor/permission、surface、query/language、filter/context、time、result ID/rank、article version/state/audience、open evidence、ticket link、failure、judgment。
acceptance gate
coverage judgmentにreproducible search evidence。one query failだけでmissing、internal-only articleでcustomer coveredとしない。findability/adequacyを分離。
5 coverage outcomeの一つにclassify
method
covered-and-found、covered-but-not-found、partial/outdated、missing article、not suitable for self-service。title similarityでなくaudience、findability、task completion、accuracy、freshness、permission、riskで判定。
evidence
outcome/definition version、article/version、audience/access、query evidence、tested step、accuracy/freshness、missing element、risk restriction、reviewer、rationale、confidence、action。
acceptance gate
primary outcomeはmutually exclusive、related actionは共存可。not suitableはidentity-specific、dangerous、regulated、live-state、specialist理由必須。
evidence clusterとcounterexample test
method
same customer task/reusable resolutionでissue unitをgroup。keyword、sentiment、team、product areaだけではない。似るが異なるanswer/permission/version/ownerを要するcounterexampleで検証。
evidence
cluster ID/version、inclusion rule、customer phrase、task/resolution、member ID、similarity signal、counterexample、split/merge、language/product/version boundary、reviewer、unresolved。
acceptance gate
clusterにcoherent actionability/documented boundary。one ticketはmultiple issue unit可だがticket coverage denominatorはticket。AI suggestionはreviewer確認前proposal。
gap validateとevidence-backed backlog優先付け
method
knowledge-domain reviewerがsource ticket、search evidence、candidate article、counterexample、specialist restrictionを確認。reach、task impact、recurrence、risk、strategy、confidence、effort、dependency、owner capacityを別入力で評価しone AI scoreに隠さない。
evidence
validated cluster/outcome、distinct ticket/denominator、task、impact/risk、window、confidence、effort/dependency、article action、owner/reviewer、acceptance、rationale、review date。
acceptance gate
create、improve、merge、re-index、translate、archive、product fix、agent-assistedを指定。high volumeだけで優先せずconflictはhuman adjudication。
ship・use/effectiveness measure・repeat
method
normal content standard、technical review、accessibility、permission、translation、versioning、rollbackでpublish/update。articleをissue clusterへlinkしsearch/useをmonitor、task completionをsample、flag/correctionをcaptureしcomparable windowでrepeat。
evidence
article/action ID/version、approval、checks、publish/audience/index state、cluster link、search query、use/reuse、success method、failed use、correction、next window。
acceptance gate
action delivered/verifiedでbacklog close、draftだけでは不可。search/click/reuse/deflection/time/ticket changeをdenominator/confounder付き解釈しdesignなしにcausal claimしない。
illustrative case：「login help」は2つの異なるgap
hypothetical exampleでcountはOpenMax resultではない。selected 100中12がlogin/SSO。modelはone authentication article missingを提案するがhuman reviewがreject。
- search replay。 8 issue unitはnew phone/lost codes等。recovery articleはあるがinternal MFA resetでのみrank。covered-but-not-found。
- counterexample check。 4 unitはSCIM post-deployment error。recovery articleでは解決不可、tenant-specific investigation。3はverified config defect、1はdifferent IdP mapping。
- action split。 Aはrecovery title/customer phrase/link/search synonym改善。Bはverified defect向けrestricted runbookとsafe customer route。mappingはexisting integration article update。
- denominator/uncertainty保持。 12 of 100 selected ticketsと報告しall customer 12%とはしない。selection scope、issue-unit count、search evidence、confidence、next windowを記録。
learning loopの運用・measurement
role / independence
sample owner、privacy/security、issue coder、search evaluator、knowledge-domain reviewer、content owner、accessibility/localization、measurement ownerを指定。cluster modelをsole evaluatorにしない。
denominator付きmetric
selected ticket、issue unit、cluster、outcome、disagreement、action、article use、task-success、correction、repeat resultを報告。ticket/issue/search/user/article denominator分離。
failure / recovery test
biased sample、duplicate、redaction leak、multilingual split、wrong merge、inaccessible/internal-only、stale、search outage、disagreement、owner absence、publish fail、index delay、rollbackをtest。
minimum auditable analysis record
analysis_id · decision_question · population · window · sampling_frame · selection_method · ticket_crosswalk · redaction_version · issue_unit_id · customer_phrases · normalized_issue · search_actor · query · result_ranks · article_version · coverage_outcome · cluster_id · counterexamples · reviewer · priority_inputs · backlog_action · owner · acceptance · publish_version · effectiveness_method · next_window
OpenMaxによるknowledge-gap analysis coordination
OpenMaxはsampling frame freeze、minimized view、issue split/normalize、customer phrase保持、permitted search replay、outcome/cluster proposal、counterexample、domain review、owned backlog、delivery/repeat evidenceをcoordinate。privacy/specialist/cluster/priority/publicationはhuman。
sampling・privacy・causal interpretation boundary
- sampling designなしに100 ticketをall customer/channel/language/product/timeへ一般化しない。
- identity、credential、payment、sensitive narrative、security artifact、restricted investigation、不要raw textをmodel/analyst/articleへ漏らさない。
- similarity scoreで異なるpermission、version、root cause、remedy、audience、personをmergeしない。
- not suitableはsafety/service-design decisionでcustomerがdifficult、issueがlow valueの証拠ではない。
- search、click、reuse、deflection、time、volume、satisfactionは別measure。comparison/confounder analysisなしにcausal claimしない。
よくある質問
100 ticketはstatistically representative？
自動的ではない。population、selection、strata、missing channel、imbalance、inference次第。100はtransparent review unit。
no result=missing article？
いいえ。customer phrase、permission、language、filter、index health、navigation、candidate articleをtest。
one ticketからmultiple issue unit？
はい。source ticketへlinkしissue coverageとticket coverageのdenominatorを分ける。
frequent issueは全てself-service？
いいえ。identity-specific、live、dangerous、regulated、specialistが必要、またはproduct/process fixが適切。
gap priorityは？
reach、impact、recurrence、risk、strategy、confidence、effort、dependency、capacityを可視化しvolume/opaque scoreだけにしない。
OpenMaxで何をautomate？
sampling、redaction、issue unit、search evidence、proposal、review、backlog、publishing handoff、monitoring、repeatをcoordinateしconsequential judgmentは人。
情報源、編集方法、制限
OpenMax編集部はKCS v6 reuse/improve/article structure/knowledge domain analysis、NIST AI RMF selection/representativeness/measurement/monitoring/independent review、WCAG 2.2を確認しoriginal 9-step workflow、5-outcome model、illustrative login caseを作成。2026年9月3日再確認。
- Consortium for Service Innovation — Knowledge Domain Analysis
- Consortium for Service Innovation — Reuse is Review
- Consortium for Service Innovation — KCS Article Structure
- NIST — AI RMF Core
- W3C — WCAG 2.2