Version. v0.12 Date. 2026-07-06 Carries forward. GRANTHA Discovery Record v0.11 (Decision J3, architecture fork closed). This entry adds Part K, Decision J5.
Trigger. STELE_TO_GRANTHA_ANSWER-v0_1.md (2026-07-06), answering GRANTHA_TO_STELE_REQUEST-v0_1.md (2026-07-05), which carried §G5 (drafted 2026-06-20).
§G5 assumed Stele's core already exposed a principal-nature property — the host-owned ActorRef.kind enum (person / companion / agent / contributor) — and asked Stele to neutralize it so GRANTHA wouldn't inherit Loomworks vocabulary.
Stele's answer corrects this: ActorRef.kind was never Stele's. It lives engine-side, in the Loomworks memory substrate. Stele core, grounded twice against the live tree (8246729, re-confirmed 3d9086a), exposes no principal-nature concept at all — no neutral property, no enum, nothing. GRANTHA was not asking Stele to strip a leak out of the substrate. It was asking Stele to build its first-ever principal-nature concept.
This is a correction to record, not discard: §G5's underlying worry — a second host inheriting the first host's role taxonomy — was the right worry. It just named the wrong artifact as the threat.
Settled. GRANTHA does not cite a Stele field, type, or location for the human/autonomous distinction, because none exists yet. Instead, GRANTHA's seed records:
> Stele has committed to exposing a neutral principal-nature distinction (human/autonomous) in stele-core, as part of the Path C architecture (sibling human-Stele / agentic-Stele products sharing a core library). Name, type, and read surface are unfixed as of 2026-07-06. GRANTHA's identity input binds to this property when Stele publishes its contract — not before.
Why not invent a placeholder name. An invented field name breaks GRANTHA's seed the moment Stele fixes the real one, trading a currently-honest gap for a later silent lie. Rejected for that reason.
Sequencing GRANTHA is blocked behind, unchanged by GRANTHA: TS-GROUND-1 (Stele security grounding) → remediation schedule → packaging completion → Miami handover → agentic sibling design opens (nature property shaped here). No date. GRANTHA does not gate this sequence and cannot accelerate it.
Also unbuilt, same gate: the RFC 8693 act-claim on-behalf-of chain for audit. GRANTHA relies on Stele's commitment to it, not on a shipped contract.
Unchanged: the R-B20 human-commit gate remains host-side — GRANTHA implements it for grant operations as the consuming host. Stele owns authentication and provenance; this authorization layer was always GRANTHA's, and Stele's answer confirms the split as intended, not as a gap.
GRANTHA's identity-reading code cannot be written against a real contract yet. The seed carries the dependency as named and dated (2026-07-06) rather than silently assuming a shape. This is a soft gate, not a hard one: GRANTHA's other build items (local decision layer, FORAY anchoring) do not wait on it. Only the literal line of code that reads "is this principal human or autonomous" waits.
Decision J3 (architecture fork closure, local hot path + FORAY-anchored state-changes) and Decision J4 (what did not move) stand untouched. This entry is additive.
DUNIN7 · GRANTHA Discovery Record · Part K, Decision J5 GRANTHA_DISCOVERY_RECORD-v0_12.md — 2026-07-06