{"id":"crg-verification-loop","level":"system","name":"Verification closes the realization loop","shortName":"Verification loop","role":"Connects simulation, formal checks, silicon measurement, system qualification and workload evaluation.","description":"A realization graph becomes engineering practice when every edge has a check. Functional simulation and formal methods test logical behavior; timing, power and physical signoff test implementation models; wafer probe, package test and system commissioning test manufactured behavior; workload evaluation tests useful output. These checks have different scopes and failure modes. The loop is intentionally iterative: a measured thermal bottleneck can change placement, package choice, runtime scheduling or even the model’s numerical strategy. The atlas uses this loop to distinguish a plausible design story from an accepted physical system.","specs":[{"label":"Evidence ladder","value":"Simulation → signoff → silicon → system → workload","context":"Each rung narrows uncertainty at a different boundary."},{"label":"Feedback","value":"Correction, redesign or bounded deployment","context":"A failed check is information, not an automatic AGI conclusion."}],"interfaces":[{"name":"Predicted behavior","direction":"in","signal":"Models, simulation traces and signoff results"},{"name":"Observed behavior","direction":"out","signal":"Measured margins, errors, useful work and corrections"}],"tradeoff":"More validation reduces uncertainty but adds time, test structures, instrumentation and cost; the boundary should match consequence.","whyAI":"A model’s capability is only meaningful when computation is correct, capacity is available, failures are recoverable and outcomes meet the declared evaluation task.","related":["crg-55-year-resolution","evaluation","known-good-die-test","checkpoint","astra"],"sources":[{"title":"Semi AI Foundry — Communication Theory of Computation Realization","url":"https://semiaifoundry.com/updates/communication-theory-of-computation-realization/"},{"title":"Google — Elastic training and goodput","url":"https://cloud.google.com/blog/products/ai-machine-learning/elastic-training-and-optimized-checkpointing-improve-ml-goodput"}],"scale":"Design evidence → operating evidence","kind":"system","mechanism":["Declare the behavior, metric, unit and acceptance boundary.","Run the least expensive check that can falsify the current assumption.","Move to silicon, package, system and workload measurements as uncertainty narrows.","Track failures to the graph edge and update the affected representation or budget.","Record the qualified scope so evidence is not generalized beyond its task."],"misconception":"Passing a benchmark or simulation at one boundary does not qualify every physical or cognitive boundary of the system.","check":{"question":"What is the purpose of the verification loop?","options":["To turn an animation into a measurement","To connect predictions and observations across design, manufacturing, system and workload boundaries","To eliminate the need for engineering judgment"],"answer":1,"explanation":"Different checks expose different failure modes; the loop preserves scope and feeds corrections back into design."},"science":[{"title":"Observed margin","equation":"m = x_{observed}-x_{required}","explanation":"A margin is meaningful only when x_observed and x_required share a metric, unit, configuration and time boundary. A negative margin identifies a failed declared condition.","assumptions":"The scalar margin is a teaching abstraction; multi-dimensional timing, thermal and quality constraints need vector or worst-case treatment."}],"history":[],"prerequisites":["crg-package-system-closure","evaluation"],"learningObjective":"Design a scoped verification loop that connects a prediction to an observation and a corrective action.","workstream":"CRG","visualFamily":"model-workload","engineeringExample":"If a package thermal measurement is 82 °C against an 85 °C limit, the margin is +3 °C for that configuration and workload; it is not a fleet-wide guarantee or a model-quality score.","evidenceNotes":"The loop is a systems-engineering teaching pattern. Provider-specific goodput sources illustrate one recovery mechanism, not a universal reliability law.","labId":"library","scope":"Use the manufacturing investigation to trace compatible inputs, assembly and operating constraints; no current standalone calculator models this lesson’s full mechanism.","investigationId":"manufacturing","labScope":"Use the manufacturing investigation to trace compatible inputs, assembly and operating constraints; no current standalone calculator models this lesson’s full mechanism."}