Factory Line · System
Quality Factory and Six Sigma KPIs
Quality becomes a named, repeatable workflow step instead of a vague opinion after the fact.
Why this matters
Many AI teams stop at “the model returned a file.” That is not the same as “the result is good enough to ship.” Quality Factory exists so the team can define what good looks like and score against it in a repeatable way.
That gives operators a clearer review step, gives managers a more honest read on readiness, and makes it easier to improve weak workflows over time.
Scorecards and readiness
Typical quality flow: probe the package for facts, score it against the rubric, and route failures clearly. The package passing quality is the real ship signal, not just a technically successful job run.
Readiness scoring in Builder looks at whether a workflow has enough real setup behind it to be treated seriously. Current founding cuts include:
| Rule | Value | Meaning |
|---|---|---|
leaf_gold_min | 800 | Performable leaf prompt_context characters |
leaf_thin_max | 199 | Below this is thin — not Perfect Context |
strategy_prompt_min | 400 | Strategy sandwich is not a slogan |
ship_layer_min | 80 | Layer coverage before ship talk |
ship_strategy_min | 80 | Strategy coverage |
strategy_online_stages | 3 | A real rail, not one echo stage |
ship_require_quality | 1 | Quality bind required to ship |
ship_require_research | 1 | Research completeness required |
ship_require_host | 1 | Host / surface bind required |
green_audit_days | 30 | Audit window |
Incidents and readiness findings are stored so the platform can learn from defects instead of repeating them as folklore.