PK26 · browser-local payload reduction

Large Payload Reduction Kit

Reduce explicitly selected JSON, JSONL, text, or log content into a bounded whole-unit artifact with exact omission evidence.

Choose format explicitly and edit the structured selector/range, dedupe, literal neighborhood, whole-unit truncation, artifact-byte, and token-estimate fields.

Reduction envelope

Not generated. Reduce a new selected payload.
Download current envelope
Build-time verified sample
{"ok":true,"schema_version":"utilito.product-kit.v1","kit":{"id":"large-payload-reduction-kit","version":"1.0.0"},"result":{"artifact":{"format":"log","mime_type":"text/plain","text":"startup complete\nERROR request failed\nrequest_id=42"},"retained":{"selection_ids":["startup"],"unit_ordinals":[1,2,3],"context_anchors":[{"marker_ordinal":1,"line_ordinal":2}]},"omissions":{"by_reason":{"unselected_line":1,"unselected_record":0,"unselected_node":0,"unselected_field":0,"exact_duplicate":0,"artifact_budget":0,"anchor_evidence_cap":0},"totals":{"source_primary_units":4,"retained_primary_units":3,"omitted_primary_units":1,"unselected_nodes":0,"unselected_fields":0,"duplicate_units":0,"artifact_budget_units":0,"anchor_evidence_items":0},"unselected_ranges":[{"start":4,"end":4}],"duplicate_ordinals":[],"truncated_ordinals":[],"missing_ids":[]},"diagnostics":{"duplicate_json_keys":0,"duplicate_json_key_locations":[],"secret_heuristics":[],"context_anchor_omission":{"reason":"anchor_evidence_cap","count":0}},"estimates":{"method":"ceil_unicode_code_points_divided_by_chars_per_token","chars_per_token":4,"input_code_points":67,"artifact_code_points":51,"input_tokens":17,"artifact_tokens":13},"context_pack":{"schema_version":"utilito.large-payload-context-pack.v1","artifact_format":"log","retained_units":3,"omitted_units":1,"raw_content_outside_artifact":false},"report":{"format":"json","contains_raw_content":false,"status":"review_required","summary_codes":["EXPLICIT_SELECTION","WHOLE_UNIT_OUTPUT","REVIEW_BEFORE_SHARING"]},"safety":{"model_used":false,"semantic_summary":false,"semantic_ranking":false,"inference_used":false,"network_used":false,"storage_used":false,"submitted_content_executed":false}},"receipt":{"input_bytes":406,"output_bytes":2278,"complete_bytes":2278,"artifact_bytes":51,"operations":3,"omitted_items":1,"warnings":[],"input_lines":4,"retained_lines":3,"input_records":0,"retained_records":0,"input_nodes":0,"retained_nodes":0,"omitted_nodes":0,"omitted_fields":0,"omitted_lines":1,"omitted_records":0,"omitted_evidence":0,"duplicate_omissions":0,"truncation_omissions":0},"provenance":{"deterministic":true,"model_used":false,"network_used":false,"storage_used":false,"core_versions":{"large_payload_reduction_kit":"1.0.0"}},"error":null}

Explicit limits

  • Format is always explicit; CSV and format inference are not supported. JSON selection uses own-data-property structured segments only. Wildcards, recursive descent, filters, slices, negative indices, dangerous keys, duplicate paths, and ancestor/descendant overlaps reject.
  • JSONL and text/log ranges are one-based, inclusive, sorted, and non-overlapping. Cross-item range ordering/non-overlap, unique selector IDs, and JSON path duplicate/ancestor overlap rules are explicit semantic constraints published in x-utilito-semantic-constraints; violations return typed INVALID_SELECTION because standard JSON Schema cannot express arbitrary pair relationships. Error neighborhoods use bounded caller-authored literal substring markers only; there is no regex, severity inference, semantic ranking, summarization, or model. At most 5,000 anchors are exported and additional matches are counted exactly as anchor_evidence_cap without raw reflection. Parsed JSON/JSONL is limited to 20 levels and 20,000 aggregate nodes before projection.
  • Dedupe is exact and source ordered. Truncation never splits UTF-8 code points, JSON tokens, lines, or records. A mandatory unit that cannot fit fails; optional later units are omitted only under whole_unit_omit with exact reason evidence.
  • Payload may intentionally include credential material. Selected raw content appears only in result.artifact.text and must be reviewed before sharing; all metadata, diagnostics, reports, failures, and omission evidence remain non-reflective.
  • Token counts are labeled estimates using ceil(Unicode code points / submitted chars_per_token), not a named tokenizer. No network, storage, execution, inference, telemetry, pipeline, or semantic processing is used.