A small interview that turns a theory into a tournament-ready submission bundle.
Pick one of the seven tasks, write a sentence or two on the mechanism you think drives cross-country variation,
and predicto proposes candidate predictors from the canonical panels (World Bank, OWID, V-Dem, WVS).
Tick the ones to keep, optionally upload your own country-level columns, and the agent fits a joint linear model and writes the §4.6 artefact bundle for you.
v1 scope.
Outcome data is bundled for all seven tasks. Predictors are drawn from the canonical panels and from researcher-uploaded columns; a small set of common predictors (GDP, internet use, schooling, urbanisation, fertility, female labour-force participation, manufacturing share) ships bundled, others appear as
stub, no data yet
until live-panel APIs are wired in v2.
Step 1 of 5
Pick a task
One submission per task. The seven tasks are listed below.
Step 2 of 5
Describe the mechanism
Write a sentence or two on what you think drives the cross-country variation.
Be specific enough that the agent can map your words onto measurable indicators.
Example: "Countries with higher income inequality have lower life satisfaction, because relative deprivation undermines well-being."
Step 3 of 5
Review and confirm predictors
Tick the predictors you want in the fit.
Optionally upload one or more CSVs with your own country-level columns: each file needs the columns
country_iso3
,
year
, and
value
.
The agent fits a single joint linear model on whatever you select.
Upload your own column(s) (optional)
Step 4 of 5
Fit and inspect the bundle
The agent fits a pooled linear model on the historical panel, predicts each country at the target wave year,
and writes a §4.6 artefact bundle (predictions.csv, causal_model.md, verbal_account.md, data_sources.md, code/run.R, README.md).
Diagnostics are shown first so falsifications of the theory are visible before submission.
Once the diagnostics look defensible, file this run as your submission.
The bundle on disk is stamped with a submission timestamp and added to the predicto queue.
You can still re-fit and re-submit; the most recent submit timestamp wins.