GESA × DRIFT

From Snapshot to Trajectory

DRIFT is a snapshot. GESA adds trajectory.

---

The Difference

	DRIFT	GESA
Question	What is the gap right now?	How is the gap moving across episodes?
Output	Single gap value	Gap trajectory + optimisation path
Temporal	Present	Past → Present → Projected

DRIFT tells you where you are. GESA tells you whether you're moving toward where you want to be — and how fast.

---

Gap Velocity

The core GESA × DRIFT concept is gap velocity: the rate at which the DRIFT score is changing across episodes.

GapVelocity = (DRIFT_recent - DRIFT_baseline) / episodeCount

Gap Velocity	Interpretation	GESA Response
Negative (gap closing)	Progress is happening	Safe to cool faster; exploit what's working
Zero (gap stuck)	Current strategy isn't moving the needle	Hold temperature; try variations
Positive (gap widening)	Situation is deteriorating	Raise temperature; explore aggressively

Gap velocity directly influences two parts of the GESA loop:

1. GENERATE — The generator adjusts risk tolerance based on velocity
2. SELECT — The scoring formula applies a gap velocity multiplier

---

DRIFT Episodes

Every DRIFT score is captured as a GESA episode context:

{
  context: {
    driftScore: 42,
    driftSign: 'positive',
    // ... other context fields
  },
  driftBefore: 42,     // At time of action
  driftAfter: null,    // Populated after outcome observed
}

Over time, GESA builds a history of DRIFT scores across episodes. This history reveals:

• Which contexts tend to have worsening gaps
• Which interventions close the gap fastest
• What gap magnitude is typical for this domain

---

DRIFT Trajectory in Practice

Example: Content Strategy

Episode 1:  DRIFT = +35   (methodology above performance)
Episode 2:  DRIFT = +28   (intervention: improved hooks)
Episode 3:  DRIFT = +19   (gap closing — velocity = negative)
Episode 4:  DRIFT = +22   (slight uptick — velocity shifts)
Episode 5:  DRIFT = +31   (gap widening — velocity = positive)

At episode 5, GESA detects positive gap velocity. It:

• Raises the effective temperature for this domain
• Retrieves episodes where the gap previously widened (episodes 4–5 pattern)
• Generates more exploratory candidates rather than continuing the previous strategy

Without gap velocity, the system would continue with the same strategy that worked in episodes 2–3, missing the inflection point at episode 4.

---

Measurement Alignment

DRIFT uses a standard formula for its gap:

DRIFT = Methodology Score − Performance Score

GESA does not modify this formula. It reads the output. The independence is deliberate: DRIFT remains a pure measurement layer; GESA remains a pure learning layer.

What GESA adds is memory and trajectory — two things DRIFT is not designed to provide.

---

→ GESA × Fetch