Elysia Tools
Navigation
Workflow Playbook
Prepare a canonical JSON or JSONL input, compare it with a baseline or environment bundle, investigate quality signals, and deliver a documented normalized result.
Hubs
Treat the incoming payload as evidence, not as a draft to overwrite. Keep a sanitized raw copy, name the baseline or environment bundles, and decide whether the review asks about a configuration difference or a population of records. Formatting makes structure legible; it does not establish intent.
Compare the canonical payload before normalizing it. An added key can be an approved staging override, while a missing key can be a deployment defect. For JSONL or repeated events, parse fields and profile completeness, duplicates, inferred types, and outliers so a visible diff is supported by record-level evidence.
Choose deterministic normalization when field rules are explicit. Reserve assisted normalization for genuinely heterogeneous values, review a representative sample, and list every unresolved exception. A useful handoff includes the preserved input, named baseline, accepted differences, applied rules, and the output that downstream teams should use.
Workflow playbook
Format the sanitized source without changing its values so reviewers can inspect nesting, arrays, nulls, and suspicious keys against the preserved raw copy.
Compare the formatted payload with its approved baseline, then isolate environment-specific missing keys, value changes, and additions before treating any difference as a data defect.
Parse JSONL or structured events into fields and profile completeness, duplicate rows, inferred types, format drift, and numeric outliers to distinguish a real pattern from a one-off diff.
Apply deterministic rules where they are specified; use assisted normalization only for ambiguous messy fields, then compare a representative sample with the preserved source and record unresolved exceptions.
Use the baseline and environment comparison path for missing, changed, or unexpected configuration keys; use log and dataset analysis when the concern is malformed, incomplete, duplicated, or drifting records.
Use rule-based normalization when mappings, casing, dates, and null handling are known. Use the AI-assisted path only for heterogeneous text or formats, then sample-review every proposed transformation.