Cormorant Foraging Framework

Cross-Reference to Established Fields

Purpose

This document maps the Cormorant Foraging Framework to established decision-making, control theory, and intelligence frameworks. It serves to:

1. Position CF within existing knowledge
2. Demonstrate completeness through parallel structure
3. Clarify what is borrowed vs. unique
4. Provide familiar entry points for practitioners from different fields

---

Executive Summary

Aspect	Status
Feedback loop structure	Standard (borrowed)
Three orthogonal dimensions	Unique formulation
Biomimetic grounding	Unique
Layered derivation (0 → 1 → 2)	Less common
Specific formulas	Unique
Observable anchoring principle	Unique

The loop is borrowed. The bird is yours.

---

Framework Comparison Matrix

Framework	Domain	Loop Structure	CF Equivalent
OODA	Military Strategy	Observe → Orient → Decide → Act	Sense → Measure → Act → Loop
PID Controller	Engineering	Measure → Compare → Correct	3D → DRIFT → Fetch
Cybernetics	Systems Theory	Input → Process → Output → Feedback	Foundation → Derivation → Action → Loop
Reinforcement Learning	Machine Learning	State → Action → Reward → Update	Sense → Fetch → Outcome → Re-sense
Scientific Method	Science	Observe → Hypothesize → Test → Revise	Sense → DRIFT → Fetch → Learn
PDCA (Deming)	Quality Management	Plan → Do → Check → Act	Measure → Fetch → Outcome → Adjust
System 1/2	Cognitive Psychology	Fast → Slow → Decision	Chirp (fast) → DRIFT (slow) → Fetch
Double-Loop Learning	Organizational Theory	Action → Result → Reflect → Reframe	Fetch → Outcome → DRIFT → Foundation
Bayesian Inference	Statistics	Prior → Evidence → Update → Posterior	Wake → Chirp/Perch → DRIFT → Updated Wake

---

Detailed Cross-References

---

1. OODA Loop (Military Strategy)

Origin: Colonel John Boyd, US Air Force Domain: Combat decision-making, competitive strategy

The OODA Structure

Observe → Orient → Decide → Act
    ↑                        |
    └────────────────────────┘

Mapping to Cormorant Foraging

OODA	CF Layer	CF Component	Function
Observe	Layer 0	Chirp + Perch + Wake	Gather signals, structure, context
Orient	Layer 1	DRIFT	Assess position relative to goal
Decide	Layer 2	Fetch calculation	Determine if/how to act
Act	Layer 2	Fetch execution	Perform action

Key Differences

OODA	Cormorant Foraging
"Orient" is implicit/cultural	DRIFT is explicit/calculated
No formula for decision	Fetch = Chirp × |DRIFT| × Confidence
Binary (decide or don't)	Threshold-based (execute/confirm/queue/wait)
No dimensional decomposition	Three orthogonal dimensions

What CF Adds

• Quantification: OODA describes; CF calculates
• Thresholds: CF provides explicit decision gates
• Decomposition: "Observe" breaks into Chirp (signal), Perch (structure), Wake (memory)

---

2. PID Controller (Control Engineering)

Origin: Control theory, early 20th century Domain: Industrial automation, robotics, process control

The PID Structure

          ┌─────────────────────────────────┐
          │                                 │
Setpoint ─┼─→ [Error] → [P + I + D] → Output
          │      ↑                     │
          │      └─────────────────────┘
          │           (Feedback)
          └─────────────────────────────────┘

Components:

• P (Proportional): React to current error
• I (Integral): React to accumulated error
• D (Derivative): React to rate of change

Mapping to Cormorant Foraging

PID	CF Layer	CF Component	Function
Setpoint	—	Goal/Target	What you're aiming for
Current State	Layer 0	Chirp + Perch + Wake	Current measurement
Error	Layer 1	DRIFT	Gap between current and goal
P (Proportional)	Layer 2	Chirp in Fetch	Immediate response to signal
I (Integral)	Layer 0	Wake	Accumulated history
D (Derivative)	Layer 1	DRIFT change over time	Rate of gap closure
Output	Layer 2	Fetch decision	Action taken

The PID ↔ CF Formula Parallel

PID Output:

u(t) = Kp·e(t) + Ki·∫e(t)dt + Kd·de(t)/dt

CF Fetch:

Fetch = Chirp × |DRIFT| × Confidence
      = Signal × Gap × Readiness

Key Differences

PID	Cormorant Foraging
Continuous numerical control	Threshold-based decisions
Single error signal	Three-dimensional sensing
Abstract variables	Biomimetic grounding
Tuning via Kp, Ki, Kd	Weighting via dimension scores

What CF Adds

• Confidence gating: PID always outputs; CF can choose not to act
• Dimensional decomposition: Error isn't monolithic—it has structure
• Semantic grounding: "Wake" vs "Integral" carries meaning

---

3. Cybernetics (Systems Theory)

Origin: Norbert Wiener, 1948 Domain: Systems theory, communication, control

The Cybernetic Loop

Input → Processor → Output
  ↑                   │
  └───── Feedback ────┘

Core principle: Goal-directed behavior through negative feedback

Mapping to Cormorant Foraging

Cybernetics	CF Layer	CF Component
Input	Layer 0	Chirp + Perch + Wake
Comparator	Layer 1	DRIFT
Effector	Layer 2	Fetch
Output	—	Action/Outcome
Feedback	Loop	Re-measurement
Goal	—	DRIFT = 0

Cybernetic Concepts in CF

Cybernetic Concept	CF Implementation
Negative feedback	Fetch reduces DRIFT toward zero
Homeostasis	Loop continues until DRIFT ≈ 0
Variety	Three dimensions provide requisite variety
Black box	Each layer can be treated as black box

Ashby's Law of Requisite Variety

"Only variety can absorb variety."

CF's three dimensions provide variety to match environmental complexity:

• Sound (Chirp) → Temporal/urgency variety
• Space (Perch) → Structural variety
• Time (Wake) → Historical variety

---

4. Reinforcement Learning (Machine Learning)

Origin: Sutton & Barto, computational learning theory Domain: AI, robotics, game playing

The RL Structure

         ┌────────────────────────┐
         │                        │
         ▼                        │
State → Agent → Action → Environment
                           │
                           ▼
                        Reward
                           │
                           └──→ Update Policy

Mapping to Cormorant Foraging

RL	CF Layer	CF Component	Function
State	Layer 0	Chirp + Perch + Wake	Current perception
Policy	Layer 2	Fetch formula	Decision rule
Action	Layer 2	Fetch execution	What the agent does
Reward	—	DRIFT reduction	Feedback signal
Value function	Layer 1	DRIFT	Expected distance to goal

Key Parallel: Value as Distance

RL Value Function:

V(s) = Expected cumulative reward from state s

CF DRIFT:

DRIFT = Distance from current state to goal state

Both measure "how far from goal" — RL in reward space, CF in methodology space.

Key Differences

Reinforcement Learning	Cormorant Foraging
Learns policy from experience	Policy is explicit formula
Requires training	Works immediately
Optimizes for reward	Optimizes for gap closure
State is abstract vector	State is three semantic dimensions
Exploration vs exploitation	Threshold-based confidence gating

What CF Adds

• Interpretability: CF dimensions are human-readable
• No training required: Formulas work out of the box
• Explicit confidence: RL explores blindly; CF gates on confidence

---

5. Scientific Method

Origin: Ancient, formalized 17th century Domain: Knowledge generation

The Scientific Loop

Observation → Hypothesis → Experiment → Analysis → Revision
      ↑                                              │
      └──────────────────────────────────────────────┘

Mapping to Cormorant Foraging

Scientific Method	CF Layer	CF Component
Observation	Layer 0	Chirp + Perch + Wake
Hypothesis	Layer 1	DRIFT ("I think the gap is X")
Experiment	Layer 2	Fetch (test the action)
Analysis	Loop	Measure new DRIFT
Revision	Loop	Update foundation

Parallel: Observable Anchoring

Scientific principle:

Claims must be testable against observation

CF principle:

"Every measurement ties to observable behavior, not speculation"

Both reject unfalsifiable assertions.

---

6. PDCA / Deming Cycle (Quality Management)

Origin: W. Edwards Deming, 1950s Domain: Quality management, continuous improvement

The PDCA Structure

Plan → Do → Check → Act
  ↑                  │
  └──────────────────┘

Mapping to Cormorant Foraging

PDCA	CF Layer	CF Component
Plan	Layer 0 + 1	Sense + DRIFT calculation
Do	Layer 2	Fetch execution
Check	Loop	Re-measure DRIFT
Act	Loop	Adjust approach

Key Difference

PDCA is prescriptive (tells you to plan). CF is descriptive (tells you where you are).

---

7. Kahneman's System 1/2 (Cognitive Psychology)

Origin: Daniel Kahneman, "Thinking Fast and Slow" Domain: Cognitive psychology, behavioral economics

The Dual System

System 1	System 2
Fast	Slow
Automatic	Deliberate
Intuitive	Analytical
Low effort	High effort

Mapping to Cormorant Foraging

System	CF Component	Reasoning
System 1	Chirp (high)	Fast signal detection, urgency
System 2	DRIFT + Confidence	Deliberate gap analysis
Override	Fetch threshold	System 2 can block System 1 impulse

The Override Mechanism

High Chirp (impulse to act)
        ↓
Low Confidence (Perch or Wake weak)
        ↓
Fetch score below threshold
        ↓
System 2 blocks action

CF provides explicit System 2 override through the confidence multiplier.

---

8. Double-Loop Learning (Organizational Theory)

Origin: Chris Argyris, 1970s Domain: Organizational learning, management

Single vs Double Loop

Single Loop: Action → Result → Adjust action Double Loop: Action → Result → Question assumptions → Reframe

Mapping to Cormorant Foraging

Loop Type	CF Implementation
Single Loop	Fetch → Outcome → Adjust Fetch
Double Loop	Fetch → Outcome → Re-evaluate DRIFT → Re-evaluate 3D weights

CF Supports Both

• Single loop: Keep same formula, adjust inputs
• Double loop: Question whether Chirp/Perch/Wake weights are correct

---

9. Bayesian Inference (Statistics)

Origin: Thomas Bayes, 18th century Domain: Probability, statistics, epistemology

The Bayesian Update

P(H|E) = P(E|H) × P(H) / P(E)

Prior × Likelihood → Posterior

Mapping to Cormorant Foraging

Bayesian	CF Component	Function
Prior	Wake	What we believed before
Evidence	Chirp + Perch	New observations
Likelihood	DRIFT change	How much evidence shifts belief
Posterior	Updated Wake	New belief state

The Parallel

Both frameworks update beliefs based on evidence:

• Bayesian: Mathematical probability update
• CF: Wake (memory) updates based on outcomes

---

Dimensional Comparison

How Other Frameworks Decompose "State"

Framework	State Decomposition
OODA	Implicit (not decomposed)
PID	Single error signal
RL	Abstract state vector s
Cybernetics	Input signal
CF	Three orthogonal dimensions

CF's Unique Decomposition

State = Sound × Space × Time
      = Chirp × Perch × Wake
      = Signal × Structure × Memory

Why this matters:

• Each dimension can be measured independently
• Bottleneck identification (which dimension is weak?)
• Targeted improvement (strengthen weak dimension)

---

Formula Comparison

Gap/Error Measurement

Framework	Gap Formula
PID	e(t) = setpoint − current
RL	δ = r + γV(s') − V(s)
CF	DRIFT = Methodology − Performance

Action Calculation

Framework	Action Formula
PID	u = Kp·e + Ki·∫e + Kd·de/dt
RL	a = argmax Q(s,a) or π(s)
CF	Fetch = Chirp × |DRIFT| × min(Perch,Wake)/100

CF's Unique Properties

1. Multiplicative gating: Any zero blocks action
2. Confidence floor: min(Perch, Wake) creates conservative bound
3. Threshold decisions: Not continuous output, but discrete states

---

Uniqueness Analysis

What CF Borrows

Element	Source
Feedback loop structure	Control theory, cybernetics
Gap measurement concept	PID error, RL value
Learning through iteration	Scientific method, RL
Dimensional thinking	Linear algebra, factor analysis

What CF Contributes

Element	Uniqueness
Biomimetic grounding	All components map to cormorant behavior
Semantic dimensions	Chirp/Perch/Wake vs x₁/x₂/x₃
Observable anchoring	Explicit rejection of speculation
Layered derivation	Level 0 → 1 → 2 dependency chain
Multiplicative confidence gate	Action requires alignment across dimensions
Threshold-based decisions	Execute/Confirm/Queue/Wait states
Emerged, not designed	Framework discovered through use

---

Completeness Proof via Cross-Reference

A complete decision framework must handle:

Requirement	Standard Solution	CF Solution	✓
Sensing	Input/Observation	Chirp + Perch + Wake	✅
State representation	State vector	Three orthogonal dimensions	✅
Goal comparison	Error/Value function	DRIFT	✅
Decision logic	Policy/Controller	Fetch formula	✅
Action gating	—	Confidence threshold	✅
Feedback	Loop closure	Re-measurement	✅
Learning	Update/Revision	Loop iteration	✅

CF addresses all requirements that established frameworks address.

---

Entry Points by Background

If You Come From...

Background	Start Here	Familiar Parallel
Military/Strategy	OODA mapping	Observe = Sense, Orient = DRIFT
Engineering	PID mapping	Error = DRIFT, Output = Fetch
Data Science	RL mapping	State = 3D, Policy = Fetch formula
Psychology	System 1/2 mapping	Chirp = fast, DRIFT = slow
Quality/Ops	PDCA mapping	Check = DRIFT, Act = Fetch
Science	Scientific method	Hypothesis = DRIFT, Experiment = Fetch

---

Summary Table

Dimension	Control Theory	Cybernetics	RL	CF
Input	Sensor reading	Input signal	State s	Chirp + Perch + Wake
Goal	Setpoint	Reference	Reward	DRIFT = 0
Gap	Error e(t)	Deviation	Value V(s)	DRIFT
Decision	Controller	Processor	Policy π	Fetch
Output	Actuator	Effector	Action a	Action
Feedback	Sensor loop	Feedback loop	State update	Re-sense

---

Conclusion

The Position

Cormorant Foraging is not a replacement for established frameworks. It is a biomimetic implementation of universal decision-making principles with specific additions:

1. Grounding: Abstract math becomes observable behavior
2. Decomposition: Single state becomes three orthogonal dimensions
3. Gating: Action requires confidence alignment
4. Emergence: Discovered through practice, not designed in theory

The Claim

"A biomimetic decision framework that implements established control theory principles through three orthogonal sensing dimensions, layered derivation, explicit confidence gating, and observable anchoring—grounded in cormorant foraging behavior."

The Invitation

Practitioners from any field can enter CF through their familiar framework:

• The loop is the same
• The structure is parallel
• The grounding is new
• The bird makes it memorable

---

References

Established Frameworks

Framework	Key Reference
OODA Loop	Boyd, J. "Patterns of Conflict" (1986)
PID Control	Åström & Murray, "Feedback Systems" (2008)
Cybernetics	Wiener, N. "Cybernetics" (1948)
Reinforcement Learning	Sutton & Barto, "RL: An Introduction" (2018)
PDCA	Deming, W.E. "Out of the Crisis" (1986)
System 1/2	Kahneman, D. "Thinking, Fast and Slow" (2011)
Double-Loop Learning	Argyris, C. "On Organizational Learning" (1999)

Cormorant Foraging Framework

Resource	URL
Main Framework	cormorantforaging.dev
DRIFT	drift.cormorantforaging.dev
Fetch	fetch.cormorantforaging.dev
Research	semanticintent.dev/papers
DOI	10.5281/zenodo.17114972

---

"The loop is borrowed. The bird is yours." 🦅