Project summary

Philotica Forecast Studio is an open, experimental tool for probabilistic forecasting. I am developing this project to combine historical data, statistical models, and structured assessment (including verification against actual results), while also utilizing the Agentic Delphi panel (more on this below). This is an experimental beta version.

What are this project's goals? How will you achieve them?

Two pilot statistical projects are already up and running. One of them estimates probabilities related to natural gas reserves in the United States based on data from the Energy Information Administration. The other tracks typhoons approaching Japan using data from the Japan Meteorological Agency. For questions that fall outside the scope of these pilot projects, my forecasting panel has also been launched. The feature I’m most eager to develop and test is the Agentic Delphi panel (which is already operational, but requires expert validation), since assembling a panel of experts is time-consuming and requires access to individuals who may not always be available. Agentic Delphi offers an AI-based alternative inspired by the Delphi method. Users can analyze the line of reasoning and areas of disagreement before accepting the result as a new version of their forecast. The panel can also analyze questions from pilot statistical projects. Manual sessions use the user’s own AI chatbots; automated sessions use the user’s API keys, and usage fees are charged by the providers.

Next comes a module for armed conflict recurrence, using UCDP historical data. Alongside its statistical baseline, I plan to test whether hypotheses from my War-Peace Code research.

How will this funding be used?

Item / Amount (USD) / Notes
1) Hardware (new work computer) / 2000 / depends on needs: laptop or workstation; 2) Scientific AI access (approx. 6 months) / 1200 / approx. 200 USD/month; 3) AI model APIs for the Delphi panel / 1500 / tests, runs and model comparisons; 4) Hosting and data / 500/ servers, domains, paid data sources; 5) Expert reviewers (e.g. 3 people) / 3000 / circa 1000 USD per independent review; 6) Publications / 1500 / open access fees, conferences; 7) My working time / 5300 / approx. 3 months part-time = circa 1770 USD/month.

Total = 15000

Minimum = 3000 = hardware & scientific AI access

Who is on your team? What's your track record on similar projects?

I work alone on Philotica. My background is strictly related to EIA and due diligence for private equity funds, including at Mott MacDonald and ERM, and coordinating surveys and monitoring for the environmental impact assessment of Poland's first nuclear power plant.

Forecasting I have recently done and published:

1) "Probability of Tactical Nuclear Weapon Use in a US-Israeli Coalition Conflict with Iran" (assessment as of March 18, 2026; v0.1.1, CC BY 4.0). A simplified Bayesian model in odds form (prior × odds-ratio multipliers), with sensitivity analysis, a red team section and dated falsification conditions with a verification schedule. The model placed peak risk around week 6 of hostilities, with a key decision point around April 5. On April 7, the US President wrote that "a whole civilization will die tonight", and hours later a two-week ceasefire was agreed; nuclear weapons were not used. The timing of the escalation peak matched the model; the event itself did not occur. One case cannot validate a 25 to 55% range, which is why the next step is systematic calibration. DOI: 10.5281/zenodo.19186538

2) "The Clock Loses Its Grip" (UCDP Dyadic, 294 observations): the Weibull shape parameter of gaps between conflict episodes rose from about 0.47 in 1950 to about 0.90 in 2020, 0.195 per decade (95% CI 0.112 to 0.279, p = 0.001 against a corrected null). Rounding and truncation artefacts were simulated and both worked against the trend; the procedure's false-rejection rate was 4.95% against a nominal 5%. I report it as a candidate finding, because the test was designed after an exploratory look at the data. https://philotica.com/journal/the-clock-loses-its-grip

3) "The Contrast That Survives" (Correlates of War, 1816 to 2007): the post-1914 contrast in the timing of participant engagement is significant (M2 = 24.37, p = 0.019), and remains so after removing the five and ten highest years. Without the world-war years it does not reach significance, and I publish that limit as well.

All my publications: ORCID 0009-0003-7807-5651.

What are the most likely causes and outcomes if this project fails?

The Agentic Delphi method may not provide better forecasts than a single model, but it serves as an additional tool directly linked to forecasting panels (including the user’s own forecast form). It is worth remembering that a forecast is not a prediction, and that this tool is backed by a person with over 20 years of experience in applying scientific methods, such as in modeling noise and air pollution (for the purposes of environmental impact assessments). I realize that, within the allotted time, there may be too few resolved cases to reliably measure accuracy, as the Brier index requires a large number of cases. There may also be too few users and testers, which means a lack of feedback. At the moment, this is a one-person effort, with limited time and no support, but please check out my LinkedIn profile:https://www.linkedin.com/in/philotica-mariusz-wlodarczyk-16h9, and the website https://philotica.com/ - Philotica is the hub where I publish ideas, their statistical tests and findings.

If the project fails, the result will still be published, including negative results, and the code and statistical pilots will remain openly available.

How much money have you raised in the last 12 months, and from where?

I have not sought funding so far, because I believed I first needed to have something to show.