Skip to content

Session Quality Signals

The session improvement loop should not rely only on explicit framework gaps.

Some important framework problems show up first as degraded session quality:

  • user frustration
  • trust erosion
  • repeated wrong direction
  • generic or templated output
  • weak recovery after failure
  • unclear or repetitive guidance

These are evidence signals, not automatic truth.

Signal Categories

  • frustration signal
  • trust erosion signal
  • wrong-direction signal
  • repetitive behavior signal
  • generic or AI-generated-feel signal
  • poor failure-handling signal
  • unclear guidance signal
  • over-automation or under-explanation signal
  • setup-theater signal
  • missed-coaching-opportunity signal

Required Signal Fields

Each signal should capture:

  • category
  • observed evidence
  • likely impact
  • likely source area
  • confidence
  • likely cause classification

Cause classification should be one of:

  • framework issue
  • model issue
  • unclear cause

How To Use Signals

Use signals to improve judgment in the session-end review.

Signals should help answer:

  • did the session feel bad in a reusable way?
  • is there enough evidence to justify framework work?
  • is this better treated as a local adaptation instead?
  • should this stay internal, become a draft, or become a real issue candidate?

One weak signal is usually not enough.

Repeated or high-confidence signals tied to the same framework surface are much stronger evidence.

Use templates/session-quality-signal.md.