KB-7763

Semantic Text-as-Code Deferred

3 min read Revision 1
<!-- DOC_STATUS: ACTIVE_NON_AUTHORITY -->

Limitation — Semantic Text-as-Code is DEFERRED

Authority: NON_AUTHORITY / NOT_PROMOTED · Date: 2026-06-11

TKT Base is a structural-evidence and governance layer. It is explicitly not the semantic Text-as-Code / IU Smart-Brick layer, and it must never claim to be.

What "structural" can and cannot see

TKT Base verifies bytes, hashes, reconstruction, fail-closed behavior, and governance. It treats a file's content as an opaque byte string: it can prove the bytes are exactly the committed bytes, but it cannot judge whether those bytes mean the right thing.

TKT Base CAN prove (L0–L3) TKT Base CANNOT prove (L4–L6, deferred)
the file exists and hashes to X the file's claims are semantically true
the packet reconstructs + reruns the content is a coherent Information Unit
bad input fails closed a Text-as-Code block is semantically valid
IDs/lanes/firewall/no-vector are consistent an IU traces correctly through its relations
the artifact is release-ready / bundle-complete

The deferred levels

  • LEVEL 4 — IU TRACEABILITY PASS — requires an IU graph: each unit's id, metadata, and relations, plus a traversal checker that confirms references resolve and no unit is orphaned at the semantic layer.
  • LEVEL 5 — SEMANTIC TEXT-AS-CODE PASS — requires a semantic linter over the Text-as-Code surface: that declared types/relations/constraints are mutually consistent and that the prose-as-code compiles to a coherent model.
  • LEVEL 6 — RELEASE/BUNDLE PASS — requires L0–L5 plus a release policy and a bundle manifest binding the released set.

None of these can be emitted by a base run. See IU_INPUT_REQUIREMENTS_FOR_LEVEL_4_5_6.md for the exact inputs each needs.

Forbidden overclaims (restated)

A TKT Base run MUST NOT emit IU_TRACEABILITY_PASS, SEMANTIC_TEXT_AS_CODE_PASS, or RELEASE_BUNDLE_PASS, and MUST NOT claim "the Tool is complete for semantic Text-as-Code." Any attempt to assert these from a base run is a fail-closed condition (probe P10): the harness rejects it with a nonzero exit and emits no PASS. The overclaim guard is part of L2/L3, not an afterthought.

Why deferral is the honest choice now

The IU metadata / relation / semantic-lint inputs do not yet exist in a form a checker can consume. Building a semantic PASS on absent inputs would be exactly the fail-open the base layer was built to prevent. The base layer ships the floor that is real today and names the ceiling that needs future inputs. When those inputs arrive, a TKT Semantic Extension adds L4–L6 on top of a passing L0–L3 base — it does not replace it.

Back to Knowledge Hub knowledge/dev/laws/tool-kiem-thu/base/tkt-base-structural-evidence-governance-pack-2026-06-11/limitations/TEXT_AS_CODE_SEMANTIC_DEFERRED.md