# TIBO RESET — Codex Reset Forecasts, Original Evidence & Evaluation

An independent, unofficial forecast and verification project about **Tibo Sottiaux's publicly announced global Codex resets**. This project is not operated by Tibo, OpenAI, or Codex.

Official site: https://tiboreset.pages.dev/

## What is being predicted?

The primary forecast target is the **first public post by Tibo confirming completion of a global reset**, not the moment an individual user's quota changes. We track three different clocks:

1. **Owner declaration** — Tibo publicly announces or promises a reset.
2. **Account observation** — a platform or account reflects changed allowances.
3. **Owner completion** — Tibo first publicly states that the global reset is complete.

Personal quota renewals, banked allocations, unrelated posts, and product releases are not counted as owner-completion events. A declaration and its replies/quotes may form one contextual graph. Multiple completion posts for the same declaration are counted as one event. An explicit withdrawal affects the relevant declaration context; "not yet complete" does not amount to withdrawal.

## Live evidence and forecast interfaces

- [Service and model health](https://tiboreset.pages.dev/api/health): current public runtime information, with timestamped status.
- [Latest forecast](https://tiboreset.pages.dev/api/forecast): the public current forecast and its origin time. The live record, not this static document, is authoritative for current probabilities and model version.
- [Evaluation records](https://tiboreset.pages.dev/api/evaluation): time-window outcomes and scores, including provisional or incomplete records when applicable.
- [Competitor observations](https://tiboreset.pages.dev/api/competitors): independent comparison observations; competitors' probabilities are **not** inputs to the standalone model.
- [Prediction method](https://tiboreset.pages.dev/method.html): the model's public explanation, assumptions and limitations.
- [History and verification](https://tiboreset.pages.dev/history.html): the human-readable track record.

Each endpoint has its own data freshness and missingness limits. Issuing a forecast, collecting a post, classifying that post, and scoring a matured forecast are separate steps. Source collection is not guaranteed to be comprehensive. Silence in an incomplete source window is not a confirmed negative outcome.

## Forecast and evaluation protocol

- Forecasts are recorded **before** results are known; published forecast origins are not rewritten in hindsight.
- The standalone forecasts report separate probabilities for completion within 24, 48, and 72 hours after an origin, plus timing-distribution information where available.
- For an origin `t` and horizon `h`, the evaluation window is `(t, t+h]`. The outcome must be eligible and known under the protocol. Future-publication and future-collection information must not enter an earlier forecast.
- Brier score and log loss measure probability forecasting. Time-distribution metrics require the correct event clock; no-event windows have no completion-time error.
- Missing probabilities are null. Provisional absence is reported as provisional, not upgraded to strict truth without source coverage.
- Overlapping forecast windows are not independent events. A comparison with another forecaster requires matching target definitions, origin times, source freshness and expiry conditions.
- The latest currently deployed model/runtime versions are given by the live `/api/health` endpoint; research and shadow candidates are distinct from the main public forecast.

## Sources and provenance

The primary source of owner declarations and completions is Tibo's own public [X account](https://x.com/thsottiaux). Original post identifiers, public timestamps, first-seen timestamps and links are material to classification. Published account-observation sources and competing forecasts are kept separate. Some public X search methods have delays or omissions, and verified late parent posts may require corrected event history.

## 日本語での概要

TIBO RESETは、Tibo本人によるCodex全体リセット**完了の最初の公表**を対象とした非公式の予測・検証サイトです。本人の予告、運営や利用者の反映観測、本人の完了公表を別の時計として扱います。事前の予測原票を維持し、24・48・72時間後に検証します。取得できていない情報や網羅性未確認の陰性は暫定扱いです。最新の確率・時刻・モデル版は公開APIを参照してください。

Document reviewed: 2026-11-02 JST. The linked live APIs supersede static examples for current values. Website: https://tiboreset.pages.dev/