River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

Research Repository Structure Template

Somebody types the question eighteen months later. The study that answers it is named after a project, so the search returns nothing.

Free download  ·  No account needed

A repository filed by study answers a question nobody asks. This pack re-indexes what you already have into one row per question, phrased the way somebody who does not know the answer would type it, filed under a product area, with the studies and the participant count behind it. Nielsen Norman Group's survey of repositories describes two shapes, a document library and a database of findings, and neither is indexed by the thing people search with.

The second half is the expiry. Every answer records the specific change that would invalidate it: a screen redesigned, a segment entering, a regulation landing. Never a date. A calendar rule bins good answers about things nothing has touched and keeps bad answers about things that moved last month. In the worked audit an eighteen-month rule and the change rule disagreed about 36 of 139 answers, and the 13 the calendar would have kept are the dangerous ones, because they read as fresh.

Three years at a freight-forwarding platform: 68 studies holding 149 distinct questions, only 41 reachable from a study title or tag, and 31 asked again while a live answer already sat in the folder, worth 81 researcher-days. Built for a team whose research keeps getting re-run. What a study found still comes out of research synthesis, and a standing weekly habit belongs in the continuous discovery pack; this is where both end up so the next person can find them.

Three years of research, re-indexed by question

The Study Register, the Insight Index with both expiry rules side by side, and the Reuse Log.

Study Register

Illustrative, for a fictional freight-forwarding platform called Larkspur. Eight of its 68 studies, each judged on one question: would somebody who needed one of its answers have found it from the title and the tags alone?

StudyNamed afterMethodQuestions it answeredReachable from its own title
Post-Brexit GB-EU flowsa trade corridorInterviews7No
ICS2 readiness discoverya regulationInterviews6No
Operator daily diarya methodDiary9No
Customs broker handoff researchan internal handoffInterviews4No
Small forwarder onboarding interviewsa segmentInterviews7No
Quote turnaround time studya thing people doAnalytics5Partly
Commodity code entry studya thing people doInterviews5Yes
Booking flow usability round 2a flowUsability3Yes

68 studies, 214 question-instances, 149 distinct questions, and only 41 of the 149 reachable from a title or a tag. The other 108 answers sat inside a study whose name gave no indication they were in there. The split is not random: the unreachable studies are named after regulations, corridors, segments and methods, and the reachable ones are named after something a person does. Usability tests are the exception that misleads, because their titles genuinely do carry their questions, which is why a team believes the archive returns more than it can.

Insight Index

One row per question, phrased the way somebody who does not know the answer would type it. Each answer carries the named change that would invalidate it, shown against what an eighteen-month calendar rule would have said instead.

QuestionAreanExpiry conditionChange ruleCalendar rule
Who actually enters the commodity code?Customs17Classification assistant shipsLiveExpired
What makes a quote take more than a day?Quoting14Partner network or rate APIs changeLiveExpired at 26 mo
Who chases a missing document?Customs11Chasing gains any automated elementExpired in MarchLive, 14 mo
How is a quote priced on a lane with no history?Quoting24Rate suggestion moves into the formExpiredLive
Do operators forward the automated ETA?Track and trace151The ETA model changesLiveLive
Is the customs checklist used as a checklist?Customs5Any change to the checklistLive, lowLive
How do forwarders pick a broker for a lane?Customs2not applicableOpenOpen

139 of the 149 questions have an answer. The change rule expires 51 of them and the calendar expires 61, and the two agree on only 103. Of the 36 they disagree about, 23 are answers the calendar would bin that nothing has invalidated, including a twenty-six-month finding about quote turnaround on a flow nobody has touched. The other 13 are the dangerous ones: a fourteen-month-old answer about chasing missing documents, on a screen where the automated reminder shipped in March, which the calendar still calls fresh and somebody will act on.

Reuse Log

Every incoming research request, checked against the index before anything is commissioned, and recorded either way. Quoted as it arrived rather than tidied into research language, because that is what the index has to match.

Request as it arrivedRow matchedOutcomeWould have commissioned
Can we find out who types in the HS code, I think it's the brokerQ-018Answered from the indexYes
Do operators actually trust our ETAQ-034Answered from the indexYes
Why do small forwarders drop out of onboardingQ-063Answered from the indexNo, would have guessed
We need to understand quote turnaround before we scope the redesignQ-052, Q-055Partly answered, study narrowed to one questionYes
How do people chase missing docs todayQ-093Expired in March, study commissionedYes
What is on the customs checklist that people skipQ-089Five participants, study scoped to the gapYes
Nobody has looked at partner rate latencynoneGenuinely new, study commissionedYes

56 requests over two quarters and 24 answered from the index outright, which is 43%. Report the denominator every time, because a hit rate without one is a number somebody chose afterwards. Only 17 of the 24 would have become studies, and at the register's eleven-day average that is 187 researcher-days. The other 7 would have been settled by guessing, which is a quality change rather than a saving and must not be added to it. The narrowed, expired and thin rows are what make the 43% believable.

What is in the pack

01

Insight Index

One row per question, phrased the way somebody who does not know the answer would type it, filed under one product area with the studies and the participant count behind it.

02

Expiry as a named change

Every answer records the screen, integration, segment or regulation whose movement would invalidate it, so the index can be checked against a release log rather than a calendar.

03

Study Register that points rather than stores

One row and a link per study, with the reports and transcripts left wherever they already live, because a repository project that begins with a migration ends during the migration.

04

The redundancy audit

Question-instances against distinct questions, with the repeats split into avoidable, expired and thin, and a cost attached to the avoidable ones only.

05

Reuse Log

Every incoming request checked against the index before anything is commissioned, recorded whether it hit or missed, with an honest column for whether it would really have become a study.

06

Disagreements kept on one row

Where two studies contradict each other, both answers stay on the same question with their dates, methods and segments, rather than becoming two rows nobody knows about.

How it works

  1. 1

    Send everything

    Reports, decks, transcripts, survey exports, the wiki pages, the folder nobody has opened. Partial is fine and duplicates are the finding.

  2. 2

    Extract the questions

    Each study is read for the questions it turned out to answer, including the ones nobody set out to ask, which are the ones that get searched for.

  3. 3

    Merge, never duplicate

    Each question is checked against the index first, so a repeat raises the participant count instead of opening a second row.

  4. 4

    Set the expiry and start logging

    Every answer gets the change that would kill it, and every incoming request gets checked against the index and recorded either way.

Frequently asked questions

What do I need before this is useful?

Whatever exists, however scattered. Reports, decks, transcripts, survey exports, the wiki pages, the research tool export, the folder nobody has opened. Partial is fine and duplicates are useful, because the duplicates are the finding. Nothing gets migrated or reformatted: each study becomes one row and a link.

Why index by question instead of by study?

Because a question is what somebody types. Information-foraging theory says people judge whether a source will answer them from its representation rather than its contents, and a study title represents the project. A study called Post-Brexit GB-EU flows answers who enters the commodity code on page four, and nobody will find it.

Do I have to migrate everything into a new tool?

No, and a repository project that begins with a migration ends during the migration. The Study Register points at wherever reports and transcripts already live. Filing a study is one row and a link, about twenty-five minutes, and what gets written down is the questions it answered rather than the study itself.

Why not just expire answers after eighteen months?

Because the calendar is wrong in both directions. Run both rules over the same 139 answers and they disagree on 36. Twenty-three are good answers about things nothing has touched. Thirteen the calendar would keep have already been invalidated by a product change, and those are the ones somebody acts on.

Is this a replacement for a research tool?

No. Nielsen Norman Group describes two repository shapes, a document library and a searchable database of findings, and a tool can be either. This changes the unit rather than the software: one row per question, an expiry tied to a named change, and a log of what the index deflected.

How do I know it is working?

The Reuse Log, which is the sheet most repositories skip. Every incoming request is checked against the index first and the outcome is recorded either way. In the worked audit's first two quarters, 24 of 56 requests were already answered and 17 of those would otherwise have become studies.

Does it decide what to build?

No, deliberately. The index holds what was found, never a recommendation, because a row that recommends becomes the thing people argue with while the evidence underneath goes unread. Requests from customers belong in feedback triage, and a finished usability test report gets filed here rather than written here.

Find out what your archive actually holds

Send whatever exists. The first thing back is the count: how many distinct questions, how many findable, and how many were asked twice.

Index what I already have