Version. 0.1
Date. 2026-08-06
Author. Claude.ai (drafting session). Operator: Marvin Percival.
Filed beside. scoping-notes/loomworks-c5-scoping-note-domain-reference-model-v0_1; verticals/venture-diligence/venture-diligence-candidate-seed-v0_2; verticals/venture-diligence/loomworks-vertical-venture-capital-diligence-v0_2; investigations/loomworks-foray-operational-visibility-investigation-v0_1.
Companion to. investigations/loomworks-reach-enforcement-grantha-correction-note-v0_1 (2026-07-29) — the first pass, which this one completes.
Occasion. A record-wide sweep run after seed v0.13 corrected the foundation document, to establish what had been written against the superseded model and never marked.
Status. Correction note. Nothing it names is edited — the record's discipline is that a superseded position stands where it was written and the correction sits alongside.
Architecture Decision J1 (June 2026) placed GRANTHA in the authorization seat. The grant is its sole primitive. There is no access-control list and no group object. The holder-set is never readable. OVA remains, beneath GRANTHA, as the blind-verification layer — it answers whether a grant holds without learning who holds it.
This is not a rename. The prior model's restricting act was establishing a list. The current model has no list to establish. A list can be read, copied and leaked; a grant model has no list to read.
Seed v0.12 predated J1 and committed to the superseded model in five places. Seed v0.13 (2026-08-06) corrected all five.
The 2026-07-29 correction note named four documents — the reach-enforcement scoping handoff and scoping note, the OVA Slice 1 change-request drafting handoff, and CR-2026-157 — and recorded them as correctable at their next version bump. That was right, and those four remain properly marked.
It swept the arc it was written inside. It did not sweep the record.
Three documents written the same day, in parallel work, carry the same staleness and were never named. They were not in the arc; nobody looked outside it.
> The durable finding, and it generalises past this instance. A correction note written from inside an arc naturally covers that arc. What it cannot see is what was written in parallel from the same stale premise — and a superseded architectural commitment propagates by parallel adoption at least as much as by sequence. A correction note should sweep the record, not the arc. (Recorded as a methodology candidate.)
scoping-notes/loomworks-c5-scoping-note-domain-reference-model-v0_1 (2026-06-05). States an ACL established through OVA as the restricting mechanism, live, underpinning the C5 domain-scoping work. The domain layer is not built, so nothing has been built on this — but C5 scoping is the next thing that would consume it, and it would consume the wrong model.
verticals/venture-diligence/venture-diligence-candidate-seed-v0_2 and loomworks-vertical-venture-capital-diligence-v0_2 (both 2026-07-28). The first states: Verified state (engine inspection, 2026-07-28): reach restriction is a committed seed posture — open by default, restricted by access-control list through OVA. The second states it differently but says the same thing: The seed's Memory section carries two independent access axes — reach, open by default and OVA-restricted when an access-control list is established.
> **These are the most serious of the three, and the reason is the phrase verified state. The claim was checked against the seed and the seed said it — so the verification was performed correctly and returned a true answer about a document that was itself stale. A reader has no way to distinguish that from a claim verified against a current source. A stale claim carrying a verification stamp is worse than a stale claim**, because the stamp is exactly what a careful reader looks for. > > And these were written the same day as the arc the first pass swept, which is how the gap at §2 arose in practice rather than in principle.
investigations/loomworks-foray-operational-visibility-investigation-v0_1 (2026-07-28). Describes the target-hosted Model 2 design as an access-control list naming the fund. That design is unbuilt and parked, so nothing rests on it — but the design as described cannot be built, because the mechanism it names does not exist. Whoever picks it up would find the model missing rather than find it wrong.
Recorded so the sweep's negative results are visible, not only its positive ones.
Already correct, naming GRANTHA in the seat: the stack architecture and dependencies document; the reach-enforcement correction note itself; the Stele agentic-identity investigation; and the current-status manifests v0.76 through v0.78, whose Entries 119 and 120 carry the supersession explicitly.
Correctly stale and already marked — named by the first pass, standing as history: the reach-enforcement scoping handoff and scoping note, the OVA Slice 1 drafting handoff, CR-2026-157, and the Cognisee positioning note.
Not a problem, by genre. candidate-seeds/ova/ova-candidate-seed-v0_1 uses ACL to describe OVA's own internal egg-condition primitive — a different and still-accurate claim, not a Loomworks-level authorization commitment. And discovery/loomworks-discovery-note-domain-reference-model-v0_1 is a Discovery note, whose genre is deliberate trajectory preservation; its ACL language is expected history.
Handled separately. The confidentiality-posture standing note also carried the superseded model, and is receiving a version bump rather than a correction beside it — a standing note is read as current authority, so a correction filed elsewhere does not reach its reader. (See loomworks-confidentiality-posture-v0_3-amendment-v0_1.)
DUNIN7 — Done In Seven LLC — Miami, Florida Loomworks — GRANTHA correction note, second pass — v0.1 — 2026-08-06 The first pass swept its own arc. Three documents written the same day, in parallel, carried the same premise and nobody looked.