Curriculum Quality: QA Maturity Levels and Human Trials

Generated artifact. Do not edit manually.

Generated by: scripts/build_compendia.py Regenerate with: python3 scripts/build_compendia.py --compendium qa-ci --web-and-tex Source of truth: docs/qa-ci/source/

Language / Sprache: English (EN) · Deutsch (DE) | PDF: English PDF · Deutsches PDF

Compendium: §1 Maturity & Trials · §2 Scopes & CQR Rules · §3 Graph Validation · §4 Semantic Atomicity · §5 Champion Guide · §6 Four-Level Model

---

1. Curriculum Quality: QA Maturity Levels and Human Trials

Status and companion specifications

Binding overall specification for Issue #49. Technical dashboard derivation is specified in 2 and operational Curriculum Champion workflows in 5. Implementation builds directly on existing quality reports, review ledgers, Champion assignments, and completion records without introducing a parallel QA pipeline or general certification engine.

1.1. Objective, Guiding Principles, and QA Maturity Levels M0–M7

SkillPilot communicates the quality of a skill landscape through two orthogonal dimensions:

1. QA maturity M0–M7: Which mandatory quality gates are completely satisfied for the current content snapshot?
2. Human trial: Does an assigned Champion's learning progress show an active trial, has the Champion explicitly confirmed starting a trial, or has one Champion worked through and confirmed the entire advertised scope?

Maturity describes the verified inspection stage. Champion status describes practical human trial. (Example: Physics · M7 · Human-tested — never a blanket guarantee of zero defects.)

Maturity levels form a cumulative ladder derived strictly from current central verification evidence (never copied from historical milestones or manual badges); human trials do not create a milestone M8:

Level	Domain statement
M0	No reliable baseline QA state established yet.
M1	Sources and extracted syllabus goals are ingested and traceable to official URLs.
M2	Source coverage, state/jurisdiction views, and course-level mappings are consistent.
M3	Required learning routes are closed and cleanly modeled at the atomic goal layer.
M4	Required learning routes including their terminal practice/exam endpoints are verified.
M5	School-ready core QA stage: all mandatory core checks pass.
M6	M5 plus reviewed memory-learning decisions, deck traces, and view visibility.
M7	M6 plus strictly completed deep curriculum QA across every current curricularAtomic goal of the subject.

1.2. Normative Definition of M7 and Image Governance

M7 requires every current curricularAtomic goal of the subject to satisfy the strict intersection of all five gates simultaneously (never summing or averaging gate counters or accepting rounded percentages; partial progress renders e.g. as M6 · Deep QA: 472 of 478 learning goals completed):

Gate	Required evidence
D – Descriptions	Two independent description reviews with resolved synthesis bound to current content and context.
P – Understanding evidence	Current, fully validated positive-understanding-evidence-v2 profile with truthful approval status.
A – Atomicity	Current domain decision confirming semantic atomicity and scope (4).
M – Memory	Current justified decision on memorization necessity plus card and view-visibility traces.
V – Visualizations	Current domain AI image inspection bound to the exact active asset hash and content bindings (technical deferrals never count as exceptions).

M7 achieved =
    M6 still satisfied
    AND current goal denominator uniquely determined and non-empty
    AND every current curricularAtomic goal passes all 5 gates (D/P/A/M/V)
    AND required closeout checks pass
    AND zero open M7-relevant blockers

• Separation from human approvals: M7 is a machine closeout gate and never mints human approvals (needs_human_review stays untouched; neither humanApproved nor humanIssueIdentified satisfies or blocks Gate V/M7, while human findings remain separately visible).
• Image principles & separate release gate: (1) KEEP by default for sound existing visuals regardless of generator/age, (2) domain accuracy over origin, (3) generation is not approval, (4) genuine content inspection rather than mechanical hash refreshes, (5) human findings remain independently effective. Separate CI check checkGoalVisualizationQaApprovalCoverage.ts acts as an image release gate (accepting an existing human approval or current AI approval while still blocking open human defect findings even at machine M7).

1.3. Practical Human Trials by Curriculum Champions

Regular school-facing Champion trials require at least core QA stage M5 (and may run in parallel with M7 work) across three states:

State	Precondition	Visible badge
Not started	Neither in-scope progress by an active Champion nor an explicit trial start	Baseline status from QA maturity
In progress	Active, unpaused Champion role with in-scope progress or confirmed start	Human QA in progress
Completed	One Champion's full practical run across the advertised scope + final confirmation	Human-tested

• Start signal (Product Owner decision, 20 September 2026): Existing learning progress (>= 1 completed goal inside the active Champion role's assigned scope once >= M5) automatically counts as Human QA in progress — regardless of registration date or when progress occurred. With zero progress, "Begin trial" remains available; registration alone, out-of-scope progress, or paused/ended roles never trigger the status. Existing mastery is strictly a start signal, never completion proof (100 % imported or manually set mastery alone never produces "Human-tested").
• Four conditions for "Human-tested": (1) one complete pass across the advertised scope by at least one individual Champion (partial passes across multiple people are never pooled), (2) genuine content-bound practical completion evidence, (3) zero known unresolved blocking defects (completing a goal does not auto-close a reported bug), and (4) a single explicit final confirmation ("All goals completed. Complete trial?").
• Scope binding, curriculum edits & pause/resignation: An explicit trial action binds its scope (e.g., Physics, Hesse, Upper Secondary never certifies all states or stages). When curriculum content changes, M7 reverts to the proven lower milestone if current goals become open, while human trials require re-testing only new or materially changed stations. Paused or unfinished trials drop the active "Human QA in progress" badge; a validly confirmed completion remains preserved even if the Champion later resigns.

1.4. Visual Presentation, Gray-to-Green Scale, Privacy, and Acceptance

Every UI surface consumes the same central derivation, labels, and ordered gray-to-green scale from M0 (neutral light gray) through M1–M4 (progressively stronger gray-green/green) and M5 (medium green) to M6 (vivid green) and M7 (deep saturated green) — always paired accessibly with milestone code and text in both light and dark modes. Beside it, the trial status badge independently shows Experimental (< M5, muted warning), Machine QA (>= M5 with no active/completed human trial, orange — combinable with a green M7 badge), Human QA in progress (muted green), or Human-tested (vivid green with check icon). Collection rows such as Gymnasium (DE) never inherit a single subject's top milestone or trial badge, instead summarizing factually (e.g., Human QA in 2 subjects). Public QA artifacts exclude all private learner, session, and chat data.

Acceptance area	Normative verification requirement
M7 award & counters	Granted solely on valid M6 plus strict 5-gate intersection across all current goals; zero averaging, rounding, or deferred-work exemptions.
Approval authority	AI records never masquerade as human approvals; explicit human defect findings remain active.
Trial start & closeout	In-scope progress (date-independent) or explicit start triggers "in progress"; completion requires one full individual pass, zero blockers, and final confirmation.
Scope, UI & collections	Partial scopes stay bounded, edits invalidate only affected stations, uniform gray-to-green palette everywhere, zero blanket collection promotion.