docs: correct Kautz classification to Type 3; record Type 4 as the target

[hyperpolymath]

The label contradicted its own description

NEUROSYM.a2ml read:

kautz-type = 1
description = "Neural | Symbolic — strict separation with defined interface"

The | pipe notation is Kautz Type 3's — a neural and a symbolic engine cooperating as co-routines across a defined interface. Type 1 (symbolic Neuro symbolic) is a plain neural net where the neural net does the work.

Type 3 is what is actually built here:

Neural role	"Natural language understanding and generation ONLY"
Symbolic role	"ALL mathematical and statistical computation"
The gate	src/tools/executor.jl::execute_tool()
MOLLOCK enforcement	validate_numeric_provenance() in src/tools/guardrail.jl, tested in test/guardrail_test.jl

Under Type 1 the LLM would itself be computing — precisely the failure this project exists to prevent. The old label didn't just misfile the project; it described the thing the MOLLOCK rule forbids.

Corrects all six artefacts (README.adoc ×2, EXPLAINME.adoc, guix.scm, 0-AI-MANIFEST.a2ml, ECOSYSTEM.a2ml, NEUROSYM.a2ml prose + kautz-type), and writes the taxonomy reasoning into NEUROSYM.a2ml so a future reader doesn't silently revert it.

A target end-state, which had never been recorded

docs/decisions/ held only 0000-template.md and 0001-adopt-rsr-standard.md; no target appeared in RSR_OUTLINE.adoc, STATE.a2ml, or NEUROSYM.a2ml. The convention exists elsewhere in the estate — idaptik-ums has docs/adr/0001-ai-edit-kautz6-nesy.adoc (Kautz 6) — statistikles just hadn't used it.

Adds ADR 0002 and kautz-target = 4.

Type 4 (Neuro:Symbolic → Neuro) — symbolic knowledge compiled back into the neural component. The lever already exists in [verification-pipeline], where exactly one of six stages is implemented:

Stage	Engine	Status
compute	Julia symbolic functions	implemented
explain	echidna GraphQL	planned
prove	echidna formal verification	planned
demonstrate	R/Julia walkthrough	planned
annotate	code + maths annotation	planned
verify	adversarial neurosymbolic SLM	planned

That verify stage is the Type 4 transition.

The consequence worth stating plainly

Reaching Type 3 is not a work item. It is already built — this is a paperwork correction. The engineering programme is 3 → 4, and the five planned stages are its backlog.

Corollary recorded in the ADR: those stages must stay honestly marked planned until they actually run. Marking an unbuilt stage implemented would be the documentation equivalent of a MOLLOCK, and the same objection applies.

Verification

• Zero Kautz Type 1 / kautz-type = 1 occurrences remain (grep clean)
• All 8 sites now read Type 3
• Commit signed (ED25519, %G? = G)
• Docs-only — no change to src/, test/, or CI

Note for reviewers

This is a classification judgement, not a code change. If you read the executor/guardrail boundary as Type 2 (Symbolic[Neuro] — symbolic in charge, calling the neural as a subroutine) rather than Type 3, that's a reasonable alternative reading and worth arguing here before merge. I chose Type 3 because the LLM drives the tool-calling loop rather than being invoked by the Julia pipeline.

🤖 Generated with Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud