Forecasting with Accountability Loops | Generated by AI
Question: You set a calendar event a year ago as a future prediction/reminder, and now you’re thinking you should build more habits like that — using calendar events to predict the future and check back on them. Is this a good system?
Answer:
Yes, and you’re independently discovering something that serious forecasters do systematically. It has a name: forecasting with accountability loops.
Why This Works
The core mechanism: prediction → time capsule → review → calibration.
Without the review step, your brain rewrites history (“I knew that would happen” = hindsight bias). The calendar event forces an honest audit.
Superforecasters (Tetlock’s research, Superforecasting) do exactly this — write predictions with explicit probabilities, date them, review them, and track their calibration score over time. The review loop is what separates people who get better at predicting from people who just have opinions.
What to Capture in Each Event
Don’t just set a reminder. Write the full prediction at creation time:
Date: 2025-05-23
Prediction: Anthropic will cut Opus pricing by >30% within 12 months
Confidence: 75%
Reasoning: DeepSeek pressure, inference cost curves, API volume incentives
Review date: 2026-05-23
When you review, record:
- What actually happened
- Was your reasoning correct even if outcome was wrong?
- What would have changed your prediction?
Suggested Categories for You
Given your trajectory:
AI/Tech
- Model capability jumps (GPT-5, Claude 4, Qwen 4 timeline)
- Price drops (your Anthropic prediction)
- Which labs survive/consolidate in 2 years
Career
- Where your income comes from in 12 months
- Whether you’re still contracting or full-time AI role
- Token consumption rate (you’re at 1.5B/yr now — what in 2026?)
Personal projects
- Which of your tools (ww, iclaw, zz, Hermes plugins) you’re still using in 1 year
- Blog post count, notes count milestones
Markets/macro
- CNY/USD, China tech regulation, GPU export controls impact
Simple System (Terminal-first for you)
Since you’re already blog/GitHub-native:
# predictions.md in your blog repo
## 2025-05-23 | Confidence: 75%
**Prediction**: Anthropic cuts Opus price >30% by May 2026
**Review**: [fill in 2026-05-23]
## 2025-05-23 | Confidence: 60%
**Prediction**: Still using Hermes over Claude Code in 6 months
**Review**: [fill in 2025-11-23]
Calendar event = just the trigger to open predictions.md and fill in the Review field.
The Real Compounding Value
After 2-3 years of this, you’ll have:
- A calibration score — do you actually know what you think you know?
- A map of where your reasoning was systematically off
- Intellectual honesty that’s rare — most engineers have opinions, few have evidence about the quality of their opinions
You’re already thinking in this direction. Formalizing it even slightly — just adding confidence % and written reasoning — makes it 10x more valuable than a bare calendar reminder.
References: