Papers & References

Academic Foundations of GESA

GESA synthesizes from several established fields while introducing novel integration. Each field contributes a specific component to the architecture.

Academic Sources

Field	Key Reference	Contribution to GESA
Simulated Annealing	Kirkpatrick, S., Gelatt, C.D., & Vecchi, M.P. (1983). Optimization by Simulated Annealing. Science, 220(4598), 671–680.	Cooling schedule, exploration/exploitation tradeoff, acceptance probability
Episodic Memory	Tulving, E. (1972). Episodic and Semantic Memory. In E. Tulving & W. Donaldson (Eds.), Organization of Memory. Academic Press.	Situated temporal episode structure — the distinction between knowing (semantic) and remembering (episodic)
Case-Based Reasoning	Aamodt, A., & Plaza, E. (1994). Case-Based Reasoning: Foundational Issues, Methodological Variations, and System Approaches. AI Communications, 7(1), 39–59.	Retrieve–Generate–Adapt cycle; using past cases to solve new problems
Reinforcement Learning	Sutton, R.S., & Barto, A.G. (2018). Reinforcement Learning: An Introduction (2nd ed.). MIT Press.	Episode-as-experience; value function as outcome learning; exploration vs exploitation
Cormorant Foraging Framework	cormorantforaging.dev	3D sensing (Chirp/Perch/Wake), DRIFT measurement, Fetch action layer — the framework GESA extends

The Novel Contribution

GESA's contribution is integration, not invention. Each source above provides one component. The novelty is in combining them:

Simulated Annealing  +  Episodic Memory  +  Case-Based Reasoning
         ↓                    ↓                      ↓
  Cooling schedule    Situated episodes        Retrieve-Generate
         ↓                    ↓                      ↓
         └──────────────────────────────────────────┘
                              ↓
                     GESA: Explicit, observable,
                     biomimetically-grounded,
                     temperature-scheduled
                     episodic optimization

What makes this novel:

Explicit temperature — CBR systems don't have annealing schedules. GESA makes the exploration/exploitation tradeoff observable and tunable.
Biomimetic grounding — The cooling schedule is not arbitrary mathematics. It maps to observable cormorant behaviour (young birds explore; experienced birds exploit).
Observable anchoring — Every recommendation traces to specific episodes, a specific temperature, and specific reasoning. No black boxes.

Framework	Relationship to GESA
OODA Loop (Boyd, 1976)	GESA extends OODA with episodic memory and temperature-governed response
PID Control	DRIFT is analogous to error signal; GESA adds integral (historical) and derivative (gap velocity) terms
Cybernetics	GESA implements Ashby's Law of Requisite Variety — variety in episode history enables variety in response
Bayesian Updating	GESA's confidence updates across episodes are structurally similar to posterior updating

Cormorant Universe Publications

Publication	URL
Fetch Framework	fetch.cormorantforaging.dev
DRIFT Framework	drift.cormorantforaging.dev
Cormorant Foraging (main)	cormorantforaging.dev
StratIQX Platform	stratiqx.com

Open Questions

The following research questions remain open in GESA v0.2:

1. Generator Architecture

Rule-based, LLM-prompted, or hybrid? LLM offers richer generation but introduces non-determinism. Rule-based is fully observable but narrower. Hybrid (rule-filtered LLM generation) is likely optimal — but the optimal filter boundary is domain-dependent.

2. Cross-Domain Episode Transfer

Can workplace episodes inform content strategies? Probably yes at the pattern level — "high context switching in both domains responds to load reduction." But the transfer learning scope needs definition. What constitutes sufficient context similarity for cross-domain transfer to be beneficial rather than misleading?

3. GESA Self-Application

Can GESA optimise its own cooling schedule by treating annealing parameter choices as episodes? This is theoretically valid and would enable the Adaptive Cool profile to fully self-tune. Non-trivial to implement without circular reasoning — the meta-episode (choosing α) is a different kind of episode than the object-level episodes (business interventions).

4. Multi-Agent Shared Episode Stores

Multiple agents sharing an episode store with independent temperature schedules. Enables collective learning while preserving individual optimization trajectories. Directly relevant for team-level HEAT deployments. The research question: when should agents share episodes vs maintain isolated stores?

Papers & References ​

Academic Foundations of GESA ​

Academic Sources ​

The Novel Contribution ​

Related Frameworks ​

Cormorant Universe Publications ​

Open Questions ​

1. Generator Architecture ​

2. Cross-Domain Episode Transfer ​

3. GESA Self-Application ​

4. Multi-Agent Shared Episode Stores ​