What’s coming next?
Probabilistic 30/60/90-day risk forecasts per AI vendor. We publish them here. We Brier-score them against subsequent reality. We publish that too. A model that forecasts and never grades itself is a guru. A model that forecasts and grades itself in public is a forecaster.
Graded out-of-sample over 131 historical 30-day windows - each forecast uses only receipts dated before it, then we check what actually happened. 0 is perfect. This is the honest v0 baseline; live forward windows accumulate from here, and wrong calls get posted in red.
§ 01LIVE FORECASTS
| VENDOR ⇅ | BAND ⇅ | RECEIPTS (6MO) | 30D ▼ | 60D ⇅ | 90D ⇅ | LIKELY NEXT | SEVERITY ⇅ |
|---|---|---|---|---|---|---|---|
| Gemini (Google) | ON WATCH | 95% | 97% | 98% | other | minor | |
| OpenAI API | CLEAN | 57% | 82% | 92% | price-increase | major | |
| DeepSeek | CLEAN | 49% | 74% | 87% | model-swap | minor | |
| Cursor | CLEAN | 43% | 67% | 81% | other | major | |
| Midjourney | ON WATCH | 39% | 63% | 78% | tier-removed | critical | |
| Lovable | CLEAN | 39% | 63% | 78% | rate-limit-cut | minor | |
| Runway | CLEAN | 39% | 63% | 78% | feature-gated | major | |
| Grok (xAI) | NO RECEIPTS | 39% | 63% | 78% | rate-limit-raise | info | |
| Devin | CLEAN | 39% | 63% | 78% | billing-model-shift | major | |
| Claude (Anthropic) | ON WATCH | 33% | 56% | 70% | feature-gated | minor | |
| Perplexity | CLEAN | 24% | 43% | 57% | rate-limit-cut | major | |
| v0 | CLEAN | 24% | 43% | 57% | billing-model-shift | critical | |
| ElevenLabs | CLEAN | 24% | 43% | 57% | rate-limit-cut | major | |
| Claude Code | ON WATCH | 17% | 32% | 44% | tier-removed | major | |
| Replit Agent | ON WATCH | 16% | 29% | 40% | billing-model-shift | major | |
| GitHub Copilot | ON WATCH | 15% | 28% | 39% | billing-model-shift | minor | |
| ChatGPT | ON WATCH | 15% | 27% | 38% | billing-model-shift | minor | |
| Notion AI | CLEAN | 14% | 26% | 36% | tier-removed | major |
§ 02 PUBLIC SCOREBOARD
These grades are backtested over our receipt archive - the model is scored against history it didn’t get to see (each forecast uses only receipts before its date). Live forward windows accumulate on top from here. Right calls stack green. Wrong calls stack red. No cherry-picking, no quiet edits.