Coverage idea — evidence and derived-data manifest

Storage and purpose

Evidence is shared under knowledge/references/; this idea's data/ holds human-readable derived findings and the session inventory. No datasets, model weights, large arrays, credentials or private raw flight logs were imported. All dates refer to 2026-10-09 America/Los_Angeles unless a UTC timestamp is explicitly given.

Shared evidence packages

Package	Canonical entry	Captured assets	Integrity/provenance
main	source	24 recorded assets; 196,372 bytes	See source metadata for each SHA-256, import time, path and completeness
worktrees	source	26 recorded assets; 207,709 bytes	See source metadata for each SHA-256, import time, path and completeness
sessions	source	12 recorded assets; 19,390 bytes	See source metadata for each SHA-256, import time, path and completeness

The main and worktree packages contain byte-for-byte selected local documents/code/JSON, with complete-file capture but partial repository selection. The session package contains selected exact decoded message text, not original JSONL bytes; its provenance JSON records full source-file hashes/sizes without importing conversations. Its generated provenance inventory is derived metadata, not a source-authored original. See curated session inventory for exclusions and selected lines.

Source version context: main 623ec672…; learned-ray 528dadf7…; raster 0490fb3b…; SpectrumNet 56d7b5ff…. Full commits/absolute paths and dirty context are retained in the source records; all four HEADs were rechecked unchanged after authorization. Main PoC result files and raster plan are explicitly untracked in their origins. Source files remain untouched.

Derived metadata and regeneration

Asset	Purpose/schema/units	Derivation and limits
Main provenance	File origin, HEAD, tracked/dirty status, bytes, SHA-256	Generated during bounded local byte-copy capture; not an experiment result
Worktree inventory	Per-file captures/Git context	Generated local-byte-copy metadata
Evidence map	Source logical key → exact capture resource	Derived navigation; content claims use actual captures
Arithmetic checks	Saved-JSON RMSE dB, coverage percent/pp, time minutes, step counts	Recomputed ring maximum differences, seed spread, best-validation steps and all-dead baseline from captured JSON; no model run
Capture recheck	50 selected source-file checks	Coordinator rehashed all retained main/worktree originals and current source files; 50/50 match. No experiments executed
Session provenance	Source hashes, bytes, timestamp bounds, record counts and selected lines/UUIDs	JSONL parser selected text blocks only; excluded tools/notifications/summaries/images

Temporary capture helpers were /private/tmp/archive_worktree_evidence.py, /private/tmp/curate_radio_sessions.py and the main worker's local capture script. These are processing aids, not durable runnable experiment code. For reproducibility, hashes/paths and selection rules are retained; future imports create new captures and never overwrite originals. Arithmetic is reproducible by reading saved seed test/log JSON, subtracting predicted/reference ring coverage, taking max absolute differences, and subtracting min from max seed coverage. No claim of complete source-project environment reproduction is made.

Capture gaps

Pre-gate candidate-source discovery produced three failed records: ITU catalog, ITU PDF, and ASSURE report. HTTP archive failed on network resolution; the web reader rejected the ASSURE file as too large. They hold no complete local source content and are not used as supporting evidence. Original URLs and errors remain in those records. No download retry was made under the authorized start exclusions.

No paper/license, checkpoint payload, source runtime or field-data audit is claimed. Reused landscape/profile sources remain owned by their existing chats. Do not update their captures in parallel.

Integrity and later assets

Each preserved capture has SHA-256/byte count in its source metadata. Supplemental provenance files are hashed in integrity inventory, generated from retained bytes for this assessment; it is a derived checksum inventory, not scientific verification. Any future dataset/weights require availability, exact source version, units/encoding, rights/access and storage decision before import. Any derived experiment output requires its input manifest and generating code/configuration.

Recorded checks

Archive counts/byte totals above derive from source capture metadata, not dataset sizes.¹²³ Final coordinator check: all 67 integrity entries match retained bytes; all 50 selected main/worktree source files still match archived hashes.⁴⁵ The reviewer independently checked the 66 entries present before the additional source recheck record. Review does not establish complete repository/session coverage or experimental performance.

type: Data Manifest
title: Coverage idea — evidence and derived-data manifest
description: Local originals, selective session text, provenance files and failed
  online captures; no datasets or weights imported.
status: draft
generated:
  by: codex/gpt-6.1-sol
  at: '2026-10-10T05:40:26.586488Z'
sources:
- id: archives
  resource: /references/SRC-local-radio-main/source.md
  title: Main source/capture metadata
- id: work-archive
  resource: /references/SRC-local-radio-worktrees/source.md
  title: Worktree source/capture metadata
- id: sessions
  resource: /references/SRC-local-radio-sessions/source.md
  title: Selective session capture metadata
- id: integrity
  resource: /docs/project-information/ideas/IDEA-002-ai-signal-coverage/data/evidence-integrity.json
  title: Retained-data hashes
- id: recheck
  resource: /docs/project-information/ideas/IDEA-002-ai-signal-coverage/data/capture-check.json
  title: Current source bytes versus archive check