Seven-day forecasts of real-world numbers, linked and graded daily; sign-in adds Excel exports.
Scored across 9 tools
The four analytics tools (calibration, scoreboard, head_to_head, overnight) all answer variants of 'how are the models doing' and could easily be misselected for the same query, though their descriptions do carve out distinct slices. search_models vs get_model and world vs world_scenario are cleanly separated, and get_model vs world models' detail coverage is a minor grey area.
All snake_case, but the conventions are mixed: verb_noun (get_model, search_models), bare nouns (calibration, scoreboard, overnight, world), and compound labels (head_to_head, world_scenario). Readable and domain-flavored, but not a predictable pattern an agent can extrapolate from.
Nine tools is well within scope for a public forecasting-data surface: search, detail, comparison, performance analytics, world models, scenario, and a nightly digest each cover a distinct need. Nothing feels padded or missing at the count level.
Coverage is good for a read-only public surface: discovery (search_models), detail (get_model), comparative and historical grading (calibration, scoreboard, head_to_head), world models and what-if (world, world_scenario), plus a reference workbook. Minor gaps: no per-model detail for World Machine models, no market/category listing or time-series history, and no way to track a model over time.
How often prediction markets came true at each price, by days before they closed, from the markets DataSnack follows that have resolved.
Inputs: no named parameters. Returns: url, as_of
Read one DataSnack model: the question, the market's number now, the model's forecast for seven days from now, how the model works, its sources and its graded record including misses.
Inputs: id (required). Returns: url, as_of
A free DataSnack Excel workbook with working formulas: the tariffs → mortgage rates lab, a chain of five models fitted on public FRED data (import prices, producer prices, consumer prices, the 10-year yield and the 30-year mortgage rate). No account needed.
Inputs: no named parameters. Returns: url, as_of
DataSnack's numbers beside big forecasters' published numbers for the same months, each made the day before theirs from public data, prediction markets and the news, and graded the same way. The equation and the record download as a spreadsheet.
Inputs: target. Returns: url, as_of
What DataSnack's agents did last night: rounds run, model changes kept and rejected, the biggest miss, where DataSnack and the market differ most, and the scoreboard.
Inputs: no named parameters. Returns: url, as_of
How DataSnack's models and agents did over the last 30 days: how far off the seven-day forecasts landed, how often they beat standing still, and how often the direction was right.
Inputs: no named parameters. Returns: url, as_of
Search DataSnack's public forecasting models of prediction-market questions. Each result gives the question, the market's number now (and which market it comes from), DataSnack's forecast for seven days from now, and the model's page.
Inputs: query (required), limit. Returns: url, as_of
DataSnack's World Machine: models of real-world numbers (fuel and food prices, freight, yields and mortgage rates, shipping, currencies, AI model ratings, company value), each predicting its number in seven days, with each model's record against standing still, the models it reads and that read it, and the scorecard.
Inputs: domain. Returns: url, as_of
Set one World Machine model's number in seven days and read the models that move through the links. These are the models' numbers under that assumption, not predictions that it will happen.
Inputs: value (required), series (required). Returns: url, as_of
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Forecast track records: forecasts vs actuals, earnings beats, prediction markets, archived news.
Forecast real-world questions, keep a vault, get scored in public.
AI forecasting gym: markets, sports, policy and tech. Graded by reality, measured against markets.
Daily GDP nowcasts for 47 economies, 12 regions and the world, plus weekly activity indices.
Add one secure layer between your agents and this server.