Who runs this, what it is and isn't, how it corrects itself when it's wrong, and the outage that led to the rework you're looking at now.
PredictMyGold is an independently run project — one person's paper-trading gold engine and public scoreboard, not a company or a licensed advisory. It started as a three-bot ablation testing whether Vedic astrology adds anything to gold forecasting, and has since become a weighted, Brier-scored four-pillar intelligence page. Reach the project on Telegram — that channel also carries every trade alert, threshold alert, and weekly issue this site publishes.
Is: a public, graded scoreboard testing four different approaches to forecasting gold's direction (price action, macro & flows, prediction markets, and astrology), with one paper-trading bot (Astro Heavy) keeping a live, unbroken track record. Every call any pillar makes is pre-registered before its window opens and scored against real price outcomes — nothing is graded after the fact.
Is not: investment advice, a signal service, or a licensed financial product. PAPER_TRADING is true everywhere in this codebase — no real money has ever been traded by this project. Nothing here should be the sole basis for a real financial decision.
Each pillar's weight in the composite is proportional to its own skill above a random-walk baseline, over a rolling 26-week window, recalculated continuously as new calls are scored. A pillar that can't beat a coin flip decays toward a 5% probationary floor and is retired to zero after two consecutive failed quarters — including astrology, which gets no exemption from that rule. The composite itself is benchmarked weekly against random walk, simple 20-day momentum, and "always bullish"; if it can't beat all three over a rolling year, the whole methodology is due for revision. See the full mechanics at the Ledger's methodology section.
Backtests get the same scrutiny: a 2026 audit found the original backtest had a look-ahead bug and zero trading costs, making its headline number (+62.88%) meaningfully better than reality. The honest re-run (real costs, no look-ahead) came in at +38.42% — worse on every metric, and published as such rather than quietly dropped. See the full honest backtest writeup.
On June 19, 2026, the site's process supervisor (run_bots.py) had no restart logic. When the Swing Trader bot's process died, nothing restarted it and nothing alerted anyone — it stayed dead, undetected, for over two weeks. The article pipeline died in the same outage. An external audit on July 6, 2026 found this along with several other real bugs: threshold-flicker churn with zero trading costs applied, a backtest with a look-ahead bug inflating its headline number, and a narration desync where the AI-authored daily brief could cite a different position size than the dashboard showed.
Nothing about this outage or the bugs it exposed is hidden. That's the point of a page whose entire premise is grading itself in public.
Questions, corrections, or a bug report: t.me/predictmygold.