Framework Architecture
The Adaptive Decision & Analysis Protocol for Thinking is a lifecycle-based decision framework designed for complex, high-stakes environments. It synthesizes insights from decision science, intelligence analysis, behavioral economics, and organizational theory into a single structured protocol.
ADAPT draws on established traditions in decision science and structured analysis. Its multi-phase lifecycle approach builds on work by Kahneman and Tversky on cognitive biases, Heuer's structured analytic techniques from the intelligence community, Klein's recognition-primed decision model, Taleb's work on antifragility and decision-making under uncertainty, and the Cynefin framework's domain classification for complexity.
The framework's distinguishing contribution is its integration of these traditions into a single protocol with built-in triage routing, iteration rules, and an AI-native execution layer. Where existing frameworks tend to address either the analytical structure or the execution mechanism, ADAPT bridges both — providing a complete lifecycle from problem framing through implementation and integrity review.
The AI execution layer uses a two-stage prompting architecture: the system prompt encodes the complete framework knowledge, while stage prompts trigger specific analytical steps. This design ensures consistent methodology application while preserving human oversight at every checkpoint.
Ground Zero triage classifies every decision across five dimensions and routes it to the appropriate track. Chaotic environments get rapid action, not false precision.
Steps proceed in order by default, but 13 iteration rules allow looping back when later steps invalidate earlier ones. Sequential is the default; iteration is the safety mechanism.
Analysis is delivered in stages with human review at each checkpoint. The AI executes; the human decides. Control is never delegated.
Step 15 forces an honest reckoning with the quality of the analysis itself. Was any part softened, avoided, or made performative? If so, go back.
Low consequences and fully reversible decisions.
Direct recommendation · ~2 min
Meaningful but manageable decisions. 3 guided stages covering 11 analytical steps including Ground Zero.
Steps: GZ, 1, 2, 4, 6, 9, 10, 11B, 12, 14, 15 · 30 min – 2 hours
Irreversible or defining decisions in non-chaotic environments. 4 thorough stages covering all 15 steps.
All steps GZ through 15 · Days to weeks
Each step shows which analysis tracks include it. Steps marked only with Comprehensive are exclusive to the Comprehensive Analysis track.
Answer three questions: consequences, reversibility, and time sensitivity. Routes to Quick Decision (low + reversible), Focused Analysis (moderate or partially reversible), or Comprehensive Analysis (high stakes or irreversible).
Define the decision, deadline, constraints, and what is explicitly out of scope.
Present the default framing and at least two alternatives. Show how each produces a different strategy.
Identify foundational requirements and flag assumptions being treated as facts.
Rate confidence (High/Medium/Low), specify what breaks each assumption. For any "High" rating, identify a specific downgrade trigger.
Assess completeness and reliability of the information base. Identify what is missing or potentially misleading.
Generate at least three plausible alternatives with trade-offs and discriminating evidence.
Apply at least three disciplinary lenses (finance, legal, engineering, etc.). Domain tools (Porter, NPV, SWOT, etc.) plug in here as needed.
Assume the favored conclusion is wrong. Build the strongest case against it without softening.
Score options across short (0–6 months), medium (6–24 months), and long term (24+ months) with weighted criteria.
Premortem at 12 months: what caused failure? Define specific, enforceable exit triggers.
Who is affected, who benefits, who bears costs, and what shapes their behavior.
Who approves, who can block, who controls resources. Identify shadow incentives and unstated fears.
What signals change confidence? Calendar-committed review points for continue, adjust, or abandon.
Resources → activities → outputs → outcomes. What will be learned after action?
Three most important audiences. What does each hear first? How is uncertainty communicated?
Which steps were rushed or skipped? Was any part shaped by discomfort, politics, or attachment rather than genuine irrelevance?
Invoked after Step 8 (or after Step 6 in the Focused track) when their trigger conditions are identified during Ground Zero or Step 1. These are not embedded in the framework — they plug in when needed.
Trigger: Harms/benefits distributed unequally across groups
Examines who bears disproportionate costs and obligations beyond legal compliance.
Trigger: Deliberate manipulation plausible, OR significant information asymmetry, OR any party has incentive to shape the information environment
Investigates whether any party could be shaping the information environment and how analysis can be insulated.
Trigger: Probabilistic forecasting or financial modeling required
Produces explicit probability estimates with calibration checks and scoring rules as evidence arrives.
ADAPT is sequential by default but iterative by design. These 13 rules govern when and how the framework loops back:
Always start with Ground Zero. No exceptions.
If a later step invalidates an earlier one, go back.
If stakes increase, upgrade the track.
If Step 2 reframe changes the problem fundamentally, restart from Step 1.
If Step 4 "High" confidence cannot identify a downgrade trigger, downgrade to Medium.
If Step 5 reveals unreliable evidence, invoke the Adversarial Deception module.
If Step 5 data collection is adding noise rather than reducing uncertainty, stop collecting and decide.
If Step 8 red-team is stronger than the lead option, return to Step 6.
If Step 9 shows a tie, add criteria or increase time horizon granularity.
If Step 10 reveals a catastrophic failure mode, add mitigations or eliminate the option.
If Step 11B reveals a blocking gatekeeper, redesign Step 14 around that gatekeeper.
If Step 12 review fires a negative indicator, execute exit triggers. Do not renegotiate.
If Step 15 reveals a step was skipped due to discomfort, go back and complete it.