Version. 0.2 Date. 2026-08-16 Status. Working draft. Investigation note; records a position, not a build commitment. v0.2 adds §8 (FORAY and regulated systems of record — the anchored-reference pattern), folded here per Operator direction of 2026-08-16 against critique §1.4; open question 6 added; sections 8–9 of v0.1 renumbered 9–10. v0.1 stands in the record as trajectory. Consumer. The Operator; future scoping sessions for the ingestion pathway; the critique document v0.2, which points here. Provenance. Operator direction, 2026-08-16: against the critique's viability verdict ("not viable: horizontal knowledge management against connect-native incumbents"), the Operator proposed that Loomworks should cooperate with, and ingest from, those knowledge systems rather than merely conceding the territory. This note works the proposal out. Prior position engaged. Critique v0.1 §3 P1: "never promise connectors to live systems as a core motion, because that is Glean's game on Glean's field." Revised here; the prior position is preserved in §6. Canonical home. loomworks-record. HTML primary, this Markdown as source.
Loomworks should ingest from the systems where knowledge already accidentally lives — Slack, Drive, Confluence, data rooms — but as a curation act, never as a sync. Selected material is pulled on the Operator's commission, extracted into held contributions each carrying a thread back to its source, and enters the permanent record only when someone with commit authority admits it. This is not competing with connect-native search; it is distilling a governed record from the mess those tools search, and it converts them from competitors into upstream suppliers. The admission gate that already protects Memory from outside contributors is exactly what makes ingestion safe. Three refusals keep the position from drifting: no continuous mirroring, no connector-count marketing, no entry into horizontal knowledge management. One cost is recorded honestly: bulk ingestion pulls personal and privileged material in at volume, so the retention-and-erasure investigation now gates the ingestion pathway at scale, not just the first European sale. A v0.2 addition extends the same reference-not-copy mechanic to regulated systems of record: FORAY-anchored references let Loomworks hold the deliberation beside a certified system without ever holding a second copy of the regulated fact — anchoring makes the beside-the-record position evidentiary, while certification itself remains the certified system's alone.
The critique v0.1 ruled horizontal knowledge management non-viable for Loomworks: connect-native incumbents index where knowledge already lives and are useful in week one against years of existing material, while a contribute-native engagement starts empty. The Operator's counter: rather than conceding that territory outright, provide a way for Loomworks to ingest from those systems — cooperation instead of competition. Is there a shape of ingestion that attacks the contribution wall without dragging Loomworks onto the incumbents' field?
There is, and the distinction is a single word: admission.
Connect-native tools index in place. Knowledge stays in the source systems; the tool retrieves over the mess; answer quality decays with the sources, which is the market's own named ceiling. Loomworks ingesting is a different act. Selected material is extracted into held contributions — the same held state every contribution passes through — each carrying an outward provenance thread to its origin: which system, which document or thread or message, which original author, which date. The material enters the permanent record only when someone with commit authority admits it. The mess stays where it lives; what crosses into Memory is curated, attributed, and gated.
One sentence for external use: search tools search the mess; Loomworks distills the record from the mess and can prove where every piece came from.
The machinery that makes this safe already exists and was built for another purpose. The held-until-admitted gate — built for credentialed outside contributors — is precisely what prevents ingestion from poisoning Memory with noise. Without the gate, connectors recreate the source systems' decay inside the record. With it, ingestion is curation by construction.
Three existing commitments compose into the ingestion pathway with nothing invented:
Critique v0.1's P1 refused connectors because a hundred-system connector layer is an arms race a one-Operator company cannot staff, and because connector-count is the axis on which the incumbents win. Both remain true. What the refusal missed is that the connector layer is commoditizing industry-wide: open connector standards — the Model Context Protocol ecosystem chief among them — are turning system connectivity into shared infrastructure rather than proprietary moat. Loomworks does not build a connector business; it consumes standard connectors and adds the two things they do not carry: the governed admission gate and the provenance thread. The incumbents' connector moat erodes under the same standards. Loomworks's differentiation does not, because its differentiation was never the pipe.
Cooperation runs the other direction too, and should be said out loud when positioning: Loomworks's rendered artifacts and the public commons are good material for the incumbents' indexes. An organization running Glean over its Drive can point Glean at Loomworks renders. Coexistence, both directions, each tool doing the thing it is for.
Ingestion is a commissioning act, never a sync. The Operator — or the Companion under an explicit delegation, surfacing an approval card — points at a bounded source: a Slack thread or channel range, a Drive folder, a Confluence space, a data-room export. The commission runs once: extraction produces held contributions, each carrying its source thread (system, locator, original author where the source exposes one, original date) alongside the ingesting actor. The Operator or a member with commit authority reviews the held set and admits what belongs in the record. Re-running a commission is a new deliberate act, not a schedule.
Attribution is dual and honest. The contribution's actor is the person who commissioned and admitted it; the source thread carries the original author. The record never pretends the admitting Operator authored a colleague's Slack message, and never pretends a Slack message is more than what it was — the trust axis stays tie-to-source, exactly as the provenance-threads framing commits.
The commercial effect targets P1 directly. An engagement's Memory can be seeded from five years of existing material in an afternoon of commissioned distillation instead of starting empty — the contribution wall's cold-start half falls. The recording-is-already-the-job test from the critique still governs which domains sustain ongoing contribution; ingestion fixes the start, not the habit, and the note says so to keep the claim honest.
Refused: continuous mirroring. A live sync recreates the source systems' mess inside Memory and destroys the admission gate's meaning at volume — a reviewer facing ten thousand auto-arriving held items admits by exhaustion, not judgment. Rejected.
Refused: connector-count marketing. The moment the pitch is "we connect to everything," buyers compare on the incumbents' axis and Loomworks loses on it. The pitch is the admission gate and the provenance thread. Rejected.
Refused: entry into horizontal knowledge management. The critique's §4 verdict stands. This pathway is interoperation from the evidence-infrastructure position, not a change of market.
Prior position, preserved and revised. Critique v0.1 §3 P1 closed with: "never promise connectors to live systems as a core motion, because that is Glean's game on Glean's field." The refusal was aimed at the right danger and drawn one line too wide: it conflated the connector arms race and sync-style indexing (still refused, above) with connector-consuming, Operator-commissioned, admission-gated ingestion (adopted here). Critique v0.2 carries the revision with this note as its pointer.
Bulk ingestion pulls in personal data, privileged communications, and third-party material at volume — Slack threads name people; data rooms carry the counterparty's confidences; Drive folders hold drafts counsel never meant to persist. The critique's P3 (memory that never forgets versus law that requires forgetting) recommended a retention-and-erasure investigation before the first European or litigation-adjacent sale. This note tightens the gate: the retention-and-erasure posture must be committed before the ingestion pathway ships at scale, because ingestion is the mechanism by which erasable-by-right material enters an architecture that resists erasure. A second legal edge rides alongside for that same investigation: source-system terms of service and licensing on programmatic export, per connector class.
Provenance. Operator question, 2026-08-16, against critique §1.4 ("several entries collide with regulated systems of record"): does FORAY help here? Folded into this note per Operator direction because it is the same reference-not-copy mechanic as §5 and §9.2, extended from source systems to certified systems.
The answer: substantially, without dissolving the problem. FORAY changes how Loomworks stands beside a certified system; it does not let Loomworks be one.
Where FORAY genuinely helps — three ways.
First, it converts duplication into attested reference. The §1.4 hazard is two records that can disagree. Content-hashing gives Loomworks a third option between holding a copy and holding nothing: hold a reference — locator, content hash, read date — to the artifact as it stands in the certified system. The deviation report, the maintenance sign-off, the dossier section live once, in the validated system; Loomworks holds the hash-anchored pointer plus the deliberation around it. There is no second version of the regulated fact to disagree with the first. If the source artifact later changes, the hash mismatch is detectable and datable — divergence stops being a latent hazard an inspector discovers and becomes a checkable event. This is the reference-not-containment principle the two-vertical decision established, made evidentiary by anchoring.
Second, it makes the deliberation layer itself trustworthy. The regulator's real fear about shadow records is after-the-fact curation — a file tidied once the inspection was announced. FORAY anchoring gives the deliberation layer independent, tamper-evident timestamps: this reasoning existed on this date, in this form, provably. That is the property that turns a second record from liability into asset — the beside-the-record layer can demonstrate it was contemporaneous and unaltered.
Third, it inverts the pitch. §4 converted connect-native tools into upstream suppliers; the same move one level down converts certified systems into anchoring counterparts. The positioning stops being "our record too" and becomes: we do not compete with your validated system — we anchor to it, and we hold the judgment it was never built to hold. This also strengthens the FORAY-as-condition motion already commissioned in the record (verticals Appendix A): the same shape, a party outside the system relying on the anchor, in a different domain.
Where FORAY does not help — stated so the affected entries are not resold on a half-truth.
Anchoring is not certification. The regulated regimes — electronic-records rules in life sciences, continuing-airworthiness record requirements, trial-master-file standards — demand validated systems, defined audit trails, access controls, and signature compliance per their own texts. Tamper-evidence is a strong ingredient toward some of those requirements and discharges none of the certification burden by itself. Loomworks anchored is still Loomworks uncertified.
It sharpens the discipline rather than relaxing it. A hash-anchored side record containing regulated facts absent from the primary record is worse for the client than an unanchored one — it is credible, dated evidence of the gap. The rule in regulated domains is therefore strict: the deliberation layer holds reasoning and references, never facts the certified system lacks. Anchoring cannot enforce that rule; it must be designed and stated (open question 6).
The organizational half remains. Which system staff write into, and which one the standard operating procedure names as the record, are process questions no anchoring answers.
Net effect on critique §1.4. The verdict softens from "reframe or demote entries 12, 14, and 22" to "reframe them onto the anchored-reference pattern, where they become strong." The catalog's aircraft entry named the overreach exactly: "the tamper-evident maintenance narrative" claims the record; "tamper-evident deliberation anchored to the certified record" is the honest version and the more sellable one. Critique v0.3 carries the pointer.
v0.1 produced by Claude under Operator direction, 2026-08-16, in the same session as the critique v0.1 it revises. The position originated as the Operator's proposal against the critique's viability verdict; the distillation-versus-indexing distinction, the admission-gate reuse, and the refused shapes emerged in the working-out. This note records a position and its leans; it commissions nothing. If the Operator elects to proceed, the path is: seed rider (origin kind plus source threads) riding the queued provenance-threads amendment; a Stage 1 posture note for file-export ingestion; a scoping arc for the first connector-consuming slice, sequenced per §9.5.
v0.2, same date: adds §8 on the Operator's FORAY question against critique §1.4, folds per Operator direction ("fold"), adds open question 6, renumbers v0.1's §8–§9 to §9–§10. Internal reference updated accordingly; no v0.1 position is revised — §8 extends the note's own mechanic to a new counterpart class.
DUNIN7 — Done In Seven LLC — Miami, Florida Loomworks — Curated-Ingestion Cooperation Investigation — v0.2 — 2026-08-16