Public site reader map

A routing guide for the public site: what to read first, what is audit evidence, and what is historical.

This page explains the site structure. The public site includes both a reader-facing narrative and the audit trail behind it. Do not start with the ledgers unless you are auditing provenance.

Start here

  • Fast scan: Project at a glance gives the fastest read on what was built, what worked, and what failed.
  • Results-first technical read: Results gives the direct-detection and OOD-generalization findings before the methodology details.
  • Full methodology read (pick a register): WRITEUP_PAPER is the academic IMRAD version (~20–25 min); WRITEUP_NARRATIVE is the narrative-arc version (~15–20 min); WRITEUP is a 1-page router that points at the two guides.
  • Repository scan: README is the GitHub-front-door version of the same story.

Evidence and audit trail

  • Evidence ledger: EVIDENCE tracks external evidence used in claims.
  • Spec lock sheet: SPEC_SHEET is the historical Phase-0 fill-in sheet. It is useful for auditing how decisions were locked, not for a first read.
  • Submission audit ledger: SUBMISSION_AUDIT is generated from ADR frontmatter. It proves that claims have decision records.
  • Assumptions registry: assumptions records load-bearing assumptions and their current interpretation.
  • Carryforward log: NEXT_STEPS separates completed carryforward work from actual future directions.

Historical decisions

The decisions section contains immutable ADRs. Read the index first. Individual ADRs are historical records: later patches clarify the current state around them instead of rewriting their original decision content.

Static analysis appendices

The notebook pages are frozen, rendered HTML appendices. They are there for inspection and reproducibility context; they are not interactive notebooks on the live site.