Philotica Forecast Studio

Philotica Forecast Studio brings together two statistical forecasting pilots, a workspace for personal forecasts, and an experimental module connecting the application with Mariusz Włodarczyk’s research on chaos and the limits of models.

The platform is open for public testing. Its purpose is to make forecasting assumptions explicit, compare alternative approaches, and learn from both forecast outcomes and user feedback.

P1 · US natural gas storage

P1 estimates the probability that working natural gas storage in the US Lower 48 will fall below a selected threshold within a supported forecast horizon. It uses data from the US Energy Information Administration (EIA).

The baseline method combines seasonal changes in storage with the historical distribution of forecast errors. Users can explore supported thresholds and horizons, inspect the underlying assumptions, and compare the baseline estimate with the optional χ experiment where an experimental result is available.

On the homepage, filtering to the Energy category also shows this pilot's live current probability together with retrospective probability estimates calculated at several selected historical dates, using the same formula as the current estimate.

P2 · Typhoon approaches to Japan

P2 estimates the probability that the number of typhoon approaches to Japan will meet or exceed a selected threshold in a specified month. It uses statistics from the Japan Meteorological Agency (JMA).

The baseline applies Bayesian estimation to historical monthly observations. An “approach” follows JMA’s definition, involving a typhoon center passing within 300 km of designated observation stations. It does not necessarily mean landfall or damage at a particular location.

P2 also provides access to the χ experiment. Historical sample sizes are limited, particularly when examining a specific month and threshold, so the strength of the available evidence is an important part of interpreting the results.

On the homepage, filtering to the Security category shows this pilot's live current probability together with retrospective probability estimates calculated at several selected historical dates, using the same formula as the current estimate.

My forecast · Your own questions

The personal forecasting workspace helps users turn an idea into a clearly defined forecasting question. It provides fields for the event, deadline, resolution criteria, evidence and counterarguments, together with a checklist for reviewing the question.

Users may enter their own probability estimate or leave the probability unestimated. Drafts can be saved in the browser and exported. This module supports the formulation and documentation of personal forecasts; it does not currently calculate probabilities for arbitrary questions or automatically monitor their outcomes.

My forecast also supports CSV import, scoring against outcomes, and comparing approaches. Existing forecasts can be imported from a CSV file, resolved against actual outcomes once they're known, and evaluated with standard scoring (log-loss, Brier score); a separate view compares different forecasting approaches side by side on the same questions.

Delphi · A structured, multi-source forecasting panel

Delphi runs a short, structured protocol — two rounds of independent estimates, a critique step, and a moderator synthesis — across several forecasters on the same question, then reports both the panel's mean estimate and the disagreement range between forecasters as two distinct numbers.

Manual relay is available now, at no cost charged by Philotica: you paste responses from your own chat assistants (e.g. ChatGPT, Claude) into each step yourself: any cost is whatever your own chat subscription already carries, not a separate charge from this application. An automatic version is also available, calling OpenAI, Anthropic or Gemini APIs directly with your own key, billed to your own provider account. OpenAI and Anthropic have each been confirmed with a real, complete session; Gemini has not — it's offered anyway because an unverified provider's risk (a response that fails to parse, or a less accurate cost figure) is bounded to the one session that tries it, never to anyone else. Open the Delphi panel directly.

Help guide · English, Japanese, Korean

Every tool has a contextual help guide, reachable from a help button on its own page or from one nav-level guide covering all ten: P1, P2, My forecast, CSV import, evaluation, action comparison, Delphi (manual and automatic, explained separately), the χ experiment, and your forecast register. Each topic answers the same six questions: what it's for, what to provide, the steps, what the result means, its limitations, and what's stored and whether it costs anything — with a labelled example throughout.

The χ experiment · From theoretical work to operational testing

The optional χ module draws on Mariusz Włodarczyk’s research programme developed in Screening Chaos: Impact Assessment of the Chaos Function χ and Its Link to the Non-Repeating Past Paradigm (PNP), its supplement on Converted χ, and the unpublished working draft Toward a Mathematics of χ ⇒ P Conversion.

Within this framework, χ denotes what lies beyond a model’s current expressive capacity. The conversion programme asks whether observable traces of missing structure can help a model extend what it can represent and forecast.

In Toward a Mathematics, Włodarczyk is developing an operational mathematical formulation of this process, including what he describes as “chaotic equations.” That formulation remains under development and is not yet implemented in the P1 and P2 experimental modules. The current implementation instead applies standard statistical methods, inspired by the research programme’s question rather than a realization of its chaos-equation apparatus, to study whether patterns in a model’s past errors can inform subsequent estimates.

The practical question is whether patterns in a model’s previous errors contain information that can improve subsequent probability estimates. The experiment provides an alternative calculation alongside the baseline, allowing the two approaches to be examined and compared. Its connection to the author’s theory is one of inspiration and motivating question, not an implementation of the theory’s proposed equations.

The broader conceptual background includes the Non-Repeating Past Paradigm (PNP) and the Implosive Edge Hypothesis (IEH), particularly the associated idea of Δ as accumulated change. These provide context for the research programme. The forecasting experiment does not constitute a test of their physical or philosophical claims, and its current implementation does not draw on the equations proposed in the draft. For the code and notes behind IEH and Δ, see the ieh-change-measure-delta repository.

Why this is an experimental release

Both the application and its research methods remain under development. Public testing allows users to examine the questions, assumptions, explanations and results while the methods undergo further evaluation.

A calculated probability is an estimate conditional on data and assumptions. The presence of a χ result does not establish that it is more accurate than the baseline. Comparisons may show improvement, no clear difference, or worse performance. Where the data or calculation do not support an experimental estimate, the application should state that limitation rather than supply a substitute percentage.

The unpublished mathematical draft is exploratory. The implementation offers a way to investigate, with standard statistical methods, the broader question the draft raises; it does not implement the draft’s proposed equations, and the draft itself is not a completed or empirically established theory of chaos.

Try the platform and share feedback

No account is required to explore the public forecasts or submit feedback. Comments on usability, unclear explanations, data issues, forecasting assumptions and the χ experiment are welcome. Feedback helps guide development; predictive performance must be assessed separately against observed outcomes.

Open Philotica Forecast Studio

---

Financial support for this project is welcome and much appreciated.

Support via PayPal.me ↗

Supporters who let us know via the Contact page will receive a discount on the eventual paid tier, proportional to their support.