# Optimize Support (60 -> 100)

Use this track only after the lead freezes findings and dimension scores. The
range in the title describes a user journey, not a score band. This track uses
Session outcomes to improve an already operating Harness; it is not a reward
for a high score.

## Select This Track

Select Optimize only when the authorized Session evidence is sufficiently
complete for the decision and at least two distinct comparable Task Episodes
support the repeated goal or friction being optimized. A repeated workflow also
requires repeated ordered stages plus a verification or stopping boundary.

Read [Session Evidence](session-evidence.md) and, for workflow candidates,
[Repeated Workflow Discovery](session-repeated-workflows.md). If comparable
Episodes, linked traces, or outcome evidence are unavailable, select
Operationalize or keep the track undetermined instead of extrapolating from a
high score, request count, repeated turn, or one successful session.

## Shape the Recommendation

For every supported optimization candidate:

1. state the repeated goal, stable input shape, observed friction, and affected
   outcome without exposing prompts, Session ids, or private paths;
2. distinguish a goal-level lead from a repeated procedure;
3. compare an observed capability, host built-in, configured owner, extension,
   and new owner in that order; and
4. define a later comparable outcome that could confirm or reject the change.

Join repeated demand with existing coverage through
[Asset Demand Reconciliation](asset-demand-reconciliation.md). Route a stable
uncovered procedure through
[Loop Discovery](../../../references/loop-engineering/loop-discovery.md) before
recommending a new Skill, Agent, Hook, automation, or Memory owner.

Calibrate against the existing
[Frontend Bug Reproduction Skill](../../../case-studies/agent-customize/bug-diagnosis-skills/reproduce-frontend-bug/SKILL.md)
and
[Backend Bug Diagnosis Skill](../../../case-studies/agent-customize/bug-diagnosis-skills/diagnose-backend-bug/SKILL.md).
When one already covers the repeated demand, prefer discover, select, invoke,
exercise, or narrowly extend it over creating a duplicate asset.

## Preserve Boundaries

Do not promote correlation, retries, copied history, or user corrections into a
stable workflow. Do not claim Loop Effectiveness from the current intervention;
require a later comparable Task Episode with an improved observable outcome.
Optimization guidance cannot expand Session, Memory, user-home, provider, or
mutation authority.