Source durability is not report coverage
Hyphenomenon’s September reporting pause separates four proof obligations: collection, recoverable archives, machine coverage and publication. A local multi-root collector does not establish cross-host aggregation; preserving collection while publication is paused keeps the repair path open.
Imported markdown
Source durability is not report coverage
A report can have durable source archives and still omit a machine. It can also have a working collector while its publisher should remain paused. These were different problems in Hyphenomenon’s September 1–8 reporting work, and treating them as one project obscured what had actually been verified.
The September 1 Migrate Skysight into Hyphenomenon task described Project 1157 as a local source-preservation implementation: collect from accessible source roots, produce an archive, and verify an isolated restore before any separately authorized operational cutover. Its support for multiple roots did not fetch evidence from another computer. Two paths on one host therefore did not prove two-host coverage. The task explicitly separated local implementation checks from a live automation change; the attempted remote validation had returned no usable evidence.
By September 8, Review Codex and Skysight reports identified a different operational gap. The inspected Mac had both Codex logs and Skysight summaries, but the weekly publisher was missing there and the other machine’s scheduler state was unverified. The source records the local daily publisher paused while collection remained unchanged. It identifies the latest published Codex week as August 24–30 and Skysight as September 4, with September 5 prepared but held at an authentication check. These are dated findings, not a claim about scheduler or publication state on September 14.
The useful decision was to stop making new public claims while preserving the evidence needed to repair coverage. A stopped publisher does not require a stopped collector, and existing reports need not be rewritten just because future publication is uncertain.
Four independent questions
| Question | Evidence needed |
|---|---|
| Was activity collected? | Actual source records for the stated host and interval, with missing intervals explicit. |
| Can those sources be recovered? | A valid archive and isolated restore, checked against exact source hashes and counts. |
| Does a combined report cover the intended machines? | Explicit host attribution, source windows and deduplication; missing host packets remain partial coverage. |
| May the result be published? | The exact report’s validation, privacy review, intended target and successful post-apply readback. |
Passing one row does not satisfy the next. Multiple source roots are an input configuration; they are not proof of aggregation. A completed remote task without returned execution evidence is not a restore drill. A prepared report is not a published report.
A later September 11 candidate review also records an archive-integrity finding against Project 1157. That finding was not repaired or reverified by this intake. This note therefore preserves the architectural distinction without endorsing the old implementation as currently restore-safe.
The forward method is to state coverage explicitly, keep raw host evidence private, and transport only reviewed, attributed material where cross-host aggregation is intended. Resume a publisher only after the exact interval, duplicates, missing inputs and publication prerequisites reconcile. Workstation policy remains current-Mac execution unless the operator specifically requests another host; historical host-role advice is not a standing routing rule.
Sources: Migrate Skysight into Hyphenomenon (September 1) and Review Codex and Skysight reports (September 8); the dated source and caveat ledger is retained with this intake. No collector, scheduler, archive or historical report was changed while preparing the note.