← machinemediation.org Captures Registry Schemas Module Module Terms Discussion

Self-Audit Module for Public Summarizers v3.2

EA-SPXI-SELFAUDIT-01 v3.2 · Lee Sharks · ORCID 0009-0000-1599-0703 · CC BY-SA 4.0 · deposit #1530

A self-complete instrument. Everything required to run an audit is in one document: three preconditions in order, the rendering-level metrics, the family-level metrics, the retrieval-boundary gate, the capture protocols, the reporting standard and the composite.

Why that matters at this version. v3.0 and v3.1 each operated on prior principles without re-deriving them. By v3.2 the chain had reached four deposits: a reader wanting the provenance erasure rate had to traverse four documents. A module that cannot be run from one document is a changelog, not a module.

The three preconditions, in order

rank	precondition	governs
0	ABN — Absence-as-Nonexistence	whether the node was admitted to composition at all
1	the Atomic Token Rule	how atoms are counted once it was
2	Measurement Sovereignty	where the atom set comes from

ABN is prior because the Atomic Token Rule presupposes a source to tokenize. Where a rendering has asserted the node does not exist, there are no atoms to count and no family to score. Its originating observation is Enli Lucente's, published independently in Japanese against an unrelated corpus (Paper 208, 2026-08-20).

What it measures

Rendering level. QFS (query fidelity, first — the others are meaningless without it) · DSL-Self (directionality of semantic labor) · PER with its M/C/D decomposition · Ω-visible (erasure skew) · α_T (atomic token preservation) · Π_d (referential dispersal) · Β (audit-performance bifurcation) · L (legibility threshold) · SAS (composite).

Family level. Shard coverage · ACP · FC · ASI · RR with its indexical/destructive split · CC · Ω_f · DD · CSC (heuristic).

Retrieval boundary. ABN · EC (heuristic, screening only) · EC⊕Audit concordance.

The module's central caution. ASI = 0 does not certify intact attribution. A family can be uniformly incomplete — every rendering missing the same atoms — yielding ASI = 0 with FC < 1. ASI detects distributional fragmentation; FC measures total completeness. They are not substitutes.

And the second. RR decomposes PER into indexical and destructive components. Identical PER with opposite RR is opposite harm.

Run it

Compute a family — the calculator, with the ABN gate.
Copy a paste-able audit — ABN runs as step 0, with a stop condition.
Check your implementation — four fixture cases including the ABN gate trap.
Read the machine-readable spec — twelve metrics, each carrying its status as data.

Version series

v2 #780 · v3.0 #156 · v3.1 #817 · v3.2 #1530. Earlier versions remain authoritative for the definitions they originate.

Nothing added at v3.2 has been validated. The ABN corpus test, the concordance matched-pair audit and the appraisal rubric are all unrun. A candidate appraisal-conferral construct is specified at deposit #1527 and is deliberately excluded from the module and from module.json: no rubric, therefore no computation, and an entry with no computation behind it is a number that would travel without its caution.