From a decoder log
to a traceable report.
Inspect output agreement, syndrome consistency and configuration identity. Every finding points back to a record in the input.
This version runs deterministic checks in your browser. It does not call Claude. The examples are synthetic, not measurements from a customer or quantum device.
01 / Input
Paste JSONL: one JSON object per line. Required fields are shown in the example. Use non-confidential data.
Each row: record_id, expected_config, observed_config, reference_correction, fpga_correction, syndrome_expected, syndrome_actual. Bitstrings contain 0 and 1.
What does the independent checker check?
It compares configuration identifiers and bitstrings explicitly provided in each record. It checks syntax, uniqueness of record IDs, configuration agreement, output agreement and syndrome agreement. The supplied syndrome values are not recomputed from a parity-check matrix. Passing these checks is not proof of decoding optimality, logical success, timing, power, or a complete noise model.
How will Claude fit into this workflow?
A planned server integration will give Claude the log and the independent checker findings. Claude will summarize evidence, propose hypotheses and suggest follow-up tests. A citation check will reject references to nonexistent records. Any hypothesis will remain distinct from the checker's factual comparisons.
02 / Report
Awaiting inputEvidence first.
Run the checker to see record-level findings, possible explanations and the tests needed to distinguish them.
Findings and evidence
Possible explanations
Next tests
Generated by deterministic rules, not Claude. “Possible explanations” are hypotheses, not proven root causes. This report compares fields supplied by the user.