Originals
Artifacts with an identity
Netlists, models and raw simulation results receive a hash of their content. The verification program itself is also preserved as an artifact.
Concept project · Engineering & AI
Engineering Dataset: Technical data. With the evidence behind it.
A proposed circuit tells only part of the story. Engineering Dataset connects it to the task, the verification run and the conditions under which the result holds. An isolated result becomes a traceable example for the next design.
A fictional project concept by elsyQ, inspired by the idea of an evidence registry for Circuit Forge.
Task
Halve a voltage
- Input
- 10 V
- Target
- 5 V
- Candidate
- Two equal resistors
- Verification
- Simulation and voltage-divider formula
- Verdict
- Target met in this example
Ideal model · no load
Inputs, verifier and original result stay connected to the verdict.
The path to evidence
Every result has a history.
The concept preserves the whole path. A model can propose a solution. A deterministic verifier decides whether it meets the requirements within explicitly stated conditions.
-
Task
Record the target and its acceptable limits.
-
Candidate
Store the circuit and parameters as a proposal.
-
Execution
Run a known version of the verifier.
-
Measurement
Extract the required quantities from the result.
-
Verdict
Check the values against the requirements.
-
Evidence
Connect the verdict to data and validity conditions.
-
Provenance
Trace sources and tool versions back to their origins.
The registry
Preserve the originals. Build knowledge from them.
Immutable files and an ongoing history form the foundation. Search views and datasets can be rebuilt from them when needed. A later check adds a verdict instead of silently overwriting the previous one.
Connections
Claims and their evidence
A claim stays separate from its evidence. Each piece of evidence names its source and conditions, such as temperature, device model or allowed tolerance. Contradictions remain visible.
Use
Reproducible datasets
A versioned manifest defines which records belong to a dataset. The search index, graph view and exports can all be rebuilt from the same registry.
Learning from failure
The wrong proposal belongs in the dataset, too.
A passing example shows a solution. A documented failed attempt also explains why another approach fails. The concept therefore keeps the failure class connected to the later correction.
First proposal
R1 = 10 kΩ · R2 = 20 kΩ
6.67 V
Target missed
The resistor ratio is wrong. The failed attempt is kept together with its verification result.
Correction
R1 = 10 kΩ · R2 = 10 kΩ
5 V
Target met
The corrected proposal gets a new verification run. A reference connects it to the first attempt.
A model proposal starts out unverified. Even a successful simulation becomes evidence only when it checks a specific requirement under known conditions.
Intended uses
A foundation for the next engineering decision.
Assistance with sources
An assistant could retrieve relevant examples and trace them back to the original verification. The validity conditions travel with the answer.
Meaningful benchmarks
Proposed solutions are checked against defined tasks. Training and evaluation data stay separate so familiar circuits do not distort the result.
Focused AI training
Successful attempts and documented corrections provide training examples. Only records whose sources permit the intended use are included in an export.
Connected to Circuit Forge
The simulator computes. The registry preserves the evidence.
Circuit Forge inspires the first engineering domain: circuits. The proposed connection uses versioned results. Other verifiers or measurement instruments could later contribute the same kind of evidence.
Explore Circuit ForgeFrom concept to system
What stands behind your technical data?
If results from simulation, verification or measurement need to remain traceable, let’s talk about the data behind them.