prompt-atlas-ecl

Chapter 11 · Preparing for Collapse and Renewal — Expansion

Canonical: PROMPT_ATLAS.md#ch11-preparing-for-collapse-and-renewal · Prompts: prompts/ch11.yaml · Part: VI

Worked Example — The Pandemic Oracle

Original: “Model a system where AI predicts and repairs global supply chain collapses before they cascade.” (worked through pandemics)

1. Aggregate weak signals — Genomic uploads, hospital admission rates, sewage assays, flight logs; declare each source’s chain-of-custody.
2. Forecast, don’t decide — The oracle outputs probability bands, not policies.
3. Pre-committed playbooks — Each band corresponds to a pre-negotiated policy bundle (mask supply pre-position, hospital surge prep, surveillance scope) ratified before crisis.
4. Public scoreboards — Predictions, calibration, and policy triggers are public; private actors cannot quietly arbitrage them.
5. Renewal arc — After the crisis, retrospective: what was foreseen, what was acted on, what was renewed.

Prompt Templates

# Sentinel design
"For  (e.g., pandemic, supply-chain, cyber, climate),
 specify: (1) data sources + custody, (2) forecast metric + calibration target,
 (3) pre-committed policy bands, (4) the named human ratifier per band,
 (5) public scoreboard format."

# Resilient-collapse rehearsal
"Run a tabletop exercise where  fails. After failure,
 identify the smallest design change that would convert the next failure
 into a graceful degradation rather than a cascading collapse."

# Renewal retrospective
"Six months after , generate a renewal retrospective:
 what foresight existed, who heard it, what changed, what is now
 designed in for the next cycle."

Anti-patterns

• Sentinel without ratifier. A forecast nobody is empowered to act on is a press release.
• Improv during crisis. Inventing the policy when the wave is breaking guarantees worst outcomes.
• Black-box oracle. Forecasts whose calibration cannot be audited will be ignored — or trusted disastrously.
• Renewal-as-PR. A retrospective with no design change is theatre.

Try This

1. Tabletop a Failure — Pick one critical system. Simulate its failure for one hour. Document the graceful-degradation gap.
2. Pre-commit a Playbook — Write the 3-band policy ladder for one shock you fear.
3. Sentinel Calibration — For one metric you forecast, publish your last 12 months of calibration.
4. Renewal Loop — Add a “renewal retrospective” item to your post-incident template.
5. Atlas-of-Renewal Contribution — Document one practice your team uses that helps you bend rather than break; share it.

Repo Cross-Links

• MACP bus — sentinel-agent topics; in-memory fallback lets you prototype without infra.
• Z3 Tester — encode “every band must have a named ratifier” as an invariant.
• tracking/__init__.py — log forecasts and outcomes for calibration.

Guide for AI & Humanity

• Listen-or-lose. Foresight without listening is wasted.
• Pre-commit. Crisis is the worst time to invent policy.
• Calibrate publicly. Trust is earned in calm, spent in storm.
• Renew on the record. A retrospective without a design change is grief, not learning.

Citations & Further Reading

• Charles Perrow, Normal Accidents (1984).
• Joseph Tainter, The Collapse of Complex Societies (1988).
• Donella Meadows, Thinking in Systems (2008).
• Andy Stirling, Keep It Complex (Nature, 2010).

This site is open source. Improve this page.