# Process-Device-Yield Integration Teaching Model Card

## Identity

- Experiment: `SA-INT-PROCESS-DEVICE-YIELD-042`
- Version: `2026-08-30-process-device-yield-v1`
- Mode: deterministic normalized teaching replay
- Intended use: evidence-boundary literacy, role routing, replacement-evidence planning, and public-atlas preparation

## Function

The model enumerates a complete categorical state space, computes twenty normalized readiness or burden proxies, classifies eight review regimes, assigns four role gates, produces five upstream and one public handoff readiness band, and closes three exact integer bookkeeping ledgers.

## Non-function

It does not simulate or estimate process physics, equipment behavior, device I-V or C-V, contact resistance, compact-model parameters, circuit delay or margin, reliability lifetime, defect density, test outcome, yield, uncertainty in physical units, qualification, signoff, publication fitness, or production fitness. No source number, equation, coefficient, geometry, process condition, measurement, model parameter, result, figure, table, dataset, code, or named-tool output enters the calculation.

## Interpretation

Readiness and burden proxies compare only states inside this locked teaching replay. They are not probabilities, physical margins, risk scores, acceptance limits, or rankings of actual technologies. The two evidence postures are physically identical; `replacement_slot` changes ownership, not a score.

## Roles

Process integration owns module boundaries, variation-source hypotheses, equipment lineage, and process-to-device evidence. Device modeling owns characterization, extraction, calibration and validation separation, model corners, applicability, and transfer uncertainty. Circuit reliability owns topology, PVT and parasitic context, delay, stability, margin, reliability mechanism, stress, use condition, and signoff boundaries. Yield learning owns test, defect, genealogy, loss taxonomy, denominator, confidence, uncertainty, and causal-learning boundaries.

Role gates are review routing only and confer no operational authority.

## Validation

The generator requires complete Cartesian coverage, all eight regimes, every declared categorical label, exact ledger closure, 112 numeric-zero prohibited fields, and byte-stable serialization. The independent verifier reimplements all equations without importing generator code, compares every row and CSV cell, checks evidence-pair identity, and runs two isolated unchanged-input regenerations.

## Limitations

- Categorical axes compress real cross-domain diversity.
- Coefficients and thresholds are source-independent teaching choices, not calibrated values.
- No causal inference is performed.
- No uncertainty is represented in physical units.
- Upstream packet labels are not imported as row data.
- The public bridge remains blocked pending owned cross-domain replacement evidence and human publication review.

## Required replacement

Before operational use, replace every teaching assumption with owned and revisioned measurement, model, lineage, uncertainty, applicability, acceptance, rollback, qualification, publication, and accountable human approval evidence.