Skip to content

Latest commit

 

History

History
40 lines (31 loc) · 1.83 KB

File metadata and controls

40 lines (31 loc) · 1.83 KB

Publishing cadence — event-driven, not clocked

This log is updated when there is new content, not on a fixed interval.

Trigger (when to append)

Append a new entry, or an update to an existing entry, when any of:

  • A pre-registered experiment concludes (verdict reached: confirm / falsify / malformed-question).
  • A diagnosis lands that changes how a result should be read (e.g. a bug that invalidates prior comparisons, an under-training artifact, an instrument- validation failure).
  • A stage of a multi-stage arc completes with a self-contained, reportable finding.

Do not append on a timer, and do not batch results for a "release." The unit is one settled intermediate result.

Per-append procedure

  1. Draft the entry from the internal result, keeping only scientific content (design, metric, pre-registration, numbers, verdict, limitations).
  2. Redact — strip every owner / third-party / infrastructure detail. The firewall is content, not location: being "in a public repo" does not make a line safe. Private detail hitchhikes via commit messages, example data, and correlatable prose. Scrub all of it.
  3. Gate — run a mechanical content review over the entry, scanning for any private third-party / owner / infrastructure detail. It must come back clean before the file is staged.
  4. Stage explicitlygit add the exact reviewed path(s). Never git add -A.
  5. Index — add a row to the README entries table.
  6. (At publication) commit + push; write a receipt.

What never ships

Real third-party names (collaborators, family, relationships), financial figures, private infrastructure (hostnames, IPs, tokens, internal memo IDs), host-specific paths, and any owner-private operational doctrine — in any artifact, including commit messages and examples.