Research Repositories
Date: 2026-08-17
A searchable store of research findings, traceable to the evidence they came from. Without one, every study is a slide deck read once and forgotten, and the same question gets researched every eighteen months by someone who didn’t know it had been.
A research repository holds findings, the raw evidence behind them, and enough metadata to find both later.
The unit is the finding, not the study. A repository organised by project answers “what did we do in Q2”; a repository organised by finding answers “what do we know about delivery expectations”, which is the question people actually have.
What a record needs
| Field | Why |
|---|---|
| The finding | One sentence, in plain language |
| Evidence | The clip, quote or observation it came from |
| Confidence | How many participants, which method, how strong |
| Context | What was being tested, when, with whom |
| Tags | Topic, page, journey stage, segment |
| Date | Findings decay |
The evidence link is what separates a repository from a wiki. A finding you can’t trace back to a session is an assertion, and after six months nobody can tell which is which — Qualitative Coding.
Tag by the question, not the project
The single decision that determines whether it gets used:
BY PROJECT (fails)
/2026-Q2-checkout-study
/2025-navigation-research
→ only findable if you remember the
project existed
BY QUESTION (works)
#delivery-expectations
#trust #returns #mobile
#first-time-buyer
→ findable by someone who wasn't there
Tag with the words people will search for, including the ones from your own domain — “P&P”, “postage”, “shipping” and “delivery” should all reach the same findings.
Why it matters more than it sounds
- Research gets repeated. The same question researched twice is a straightforward waste, and it happens constantly across team changes
- Findings compound. Three small studies pointing the same way is stronger evidence than any one of them, and only visible if they’re in one place — Triangulation
- It answers questions without new research. “Do we know anything about why people abandon at delivery?” answered in five minutes is the repository paying for itself
- It survives people leaving. Research understanding is otherwise held entirely in the heads of whoever ran it
The failure mode
It becomes a graveyard. Studies get uploaded, nobody searches, quality declines, and it stops being worth searching — which was the prediction.
What prevents it:
- Search that works — on the words people actually use
- Low friction to add — if logging a finding takes an hour, it won’t happen
- Someone who answers — a person who responds to “do we know about X?” by searching it
- Linked from decisions — a brief citing repository findings makes it load-bearing
The pattern that works: answer questions from it publicly. Every time someone asks a research question and gets a link rather than a shrug, its value becomes visible to people who’d otherwise ignore it — the same social problem as Design System Adoption.
What to store
Not everything — a repository holding every raw recording is unsearchable.
STORE findings, clips, quotes,
summaries, the tags
LINK full recordings, transcripts,
decks
DON'T STORE PII, faces, names, anything
a participant didn't consent
to being retained
Consent and retention are a genuine obligation, not a formality. Recordings contain identifiable people, often discussing personal circumstances, and a repository is exactly the kind of long-lived store where retention periods get forgotten — Research Ethics, PII in Analytics.
Tooling
Purpose-built tools exist and a well-structured shared document works. The structure and the habit matter far more than the tool — a disciplined folder of tagged markdown beats an expensive platform nobody updates.
Start with the tagging scheme, not the software. Changing tools later is easy; retrofitting a taxonomy onto 200 untagged findings is not — Taxonomy and Labelling.