GESA Framework

Appearance

GESA × HEAT ​

The Richest Episode Source ​

HEAT is GESA's richest episode source for the workplace domain. Every pain streak, bus factor signal, and context switching measurement is a structured episodic input — precisely the kind of situated, temporal data that GESA is built to learn from.

What HEAT Detects ​

HEAT (the workplace intelligence layer) surfaces signals that indicate team cognitive load and risk:

  • Pain streaks — consecutive periods of high cognitive load
  • Bus factor risk — single points of knowledge/dependency failure
  • Context switching — frequency and depth of task interruption
  • Effort concentration — uneven load distribution across team

These signals are highly specific (Team A, Sprint 14, March 3rd) and temporal (they change sprint by sprint). They are exactly what episodic memory is designed to capture.

The Question HEAT Surfaces but Doesn't Answer ​

HEAT detects. It measures. It signals.

"Given this pattern of effort signals, what intervention should we try next — and how bold should we be?"

HEAT cannot answer this. It is a sensing and measurement layer. GESA is the answer.

The Workplace Loop ​

HEAT detects:    Pain streak (3+ days, 90% cognitive load, Team A)
GESA retrieves:  Similar episodes with similar team size, similar sprint phase
GESA generates:  Candidate interventions ranked by historical effectiveness
GESA anneals:    Filters candidates by current temperature (T = 0.81 = still exploring)
GESA selects:    "Try reducing meeting load first — 7/9 similar episodes responded within 4 days"
GESA stores:     New episode with intervention and outcome (measured after sprint)
GESA cools:      Temperature advances; next similar situation will lean more toward proven strategies

HEAT Episode Structure ​

typescript
interface HeatEpisode extends Episode {
  teamId:           string
  sprintNumber:     number
  painStreakDays:   number
  cognitiveLoad:    number       // 0–100
  contextSwitches:  number       // per day
  busFactorRisk:    number       // 0–100
  teamSize:         number
  sprintPhase:      'early' | 'mid' | 'late'
  intervention:     string       // What GESA recommended
  outcomeMetric:    'burnout_resolved' | 'velocity_recovered' | 'no_change' | 'worsened'
  daysToResolution: number       // How long until the pain streak ended
}

The sprintPhase field is critical: an intervention that works in mid-sprint may not work in late-sprint when context switches are harder to reduce. GESA retrieves by sprint phase as part of the context fingerprint.

Cross-Team Learning ​

GESA can transfer HEAT episodes across teams — at the pattern level:

Team A: High context switching → WIP reduction → resolved in 3 days (6 episodes)
Team B: High context switching → similar situation → GESA retrieves Team A episodes

Cross-team transfer is enabled when DomainMatch returns a match at the workplace domain level and the context fingerprint is similar. Team-specific differences are captured in teamId — the generator can weight same-team episodes higher while still including cross-team evidence.

Multi-Agent GESA for Teams ​

An advanced deployment: multiple agents sharing a HEAT episode store but with independent temperature schedules per team.

Team A: T = 0.45 (40+ episodes, exploiting proven strategies)
Team B: T = 0.82 (8 episodes, still exploring)
Team C: T = 0.95 (2 episodes, near-cold-start, full exploration)

Each team benefits from the collective episode history but converges on its own optimal strategies at its own pace. New teams start at high temperature; experienced teams converge on what works for them specifically.

This is the multi-agent GESA architecture described in the open questions — collective learning with individual trajectories.

→ Domain Overview