Research Repository and Knowledge Curation

What should a research repository actually contain?

Findings, with studies as parent records. Nobody searches a repository for a report, so a store of documents pushes the work back onto the searcher. Curate individual claims that stand on their own, each carrying its base, population, fieldwork period, method and confidence, and link them to the study they came from. Extracting findings is the main cost of a repository and it is also the entire value.

How should research findings be tagged?

Use a small number of independent facets rather than one deep hierarchy, because a finding is about a topic, a population, a market and a decision at the same time. Keep the vocabulary small and controlled: aim for roughly fifteen to thirty topic terms, write a definition for each including the boundary that separates it from its neighbour, and require the curator's approval to add a new one. A small controlled vocabulary beats a large uncontrolled one every time.

How do you stop a repository serving out-of-date findings?

Give every finding a status rather than every study, since findings inside one study age at different rates. Four statuses work: current, dated, superseded (with a link to what replaced it) and retired. Set review periods by volatility, add event triggers such as a product, pricing, policy or market change, and make a missed review downgrade the entry to dated automatically rather than leaving it current by default.

What do you do when two studies in the repository contradict each other?

Keep both, keep both discoverable, and link them with a record that states what each found, the diagnosed cause of the difference (different population, wording, period, method, or genuine disagreement), which is better evidenced and why, and what would settle it. Showing only the more recent or more convenient finding makes an analytical judgement invisibly on behalf of everyone who will ever search. The contradiction record is frequently the most useful entry in a repository.

Why do research repositories fail?

Almost always on maintenance rather than design. The pattern is predictable: contribution falls once the founding projects close, tags drift because nobody reviews them, a stale finding is served as current, trust breaks, and people go back to asking the longest-serving researcher. The fix is a curator with allocated hours and contribution written into project close, not a better schema.

How far back should we curate our existing research?

Only as far as people are actually asking. Comprehensive backfill is where repository projects consume their budget and never launch the forward workflow. Curate forward from today, and backfill the specific studies that appear in your request log, which is usually ten to twenty rather than everything.

Can old research be deleted once it is superseded?

No. Keep superseded and retired entries, linked to whatever replaced them. The historical record is the only thing that makes change over time visible, and an organisation that deletes its old answers ends up believing it has always known what it currently knows.

What about participant data and confidentiality in a searchable repository?

Treat searchability as a new purpose that consent given for a report does not automatically cover. Decide per class of material what may be entered, what is entered but access-restricted, what is redacted on entry, and what retention period applies. A verbatim that was safe in an aggregated report can identify someone once it is tagged by market and segment and retrievable organisation-wide. Restricted entries should be visible as restricted, so a searcher knows material exists rather than concluding nothing was found.

Do we need a repository at all?

Below roughly fifteen studies, probably not. A single well-maintained index listing study, date, question, sample, method, headline findings and file location will outperform a repository, cost almost nothing and need no curator. And if nobody has time allocated to maintain it, do not build one at all: an unmaintained repository is worse than none, because it looks authoritative while serving stale evidence.