Loomworks — Knowledge Elevation Pathway Investigation — v0.1

Version. 0.1 Date. 2026-05-08 Status. Investigation. Methodology-conceptual material is well-settled; build is far behind. Filed to preserve the trajectory and the open questions for when the build catches up. Author. Claude.ai (investigation layer). Operator: Marvin Percival. Provenance. Conversation initiated by the Operator's question about how knowledge accumulating across many engagements gets extracted and elevated to domain knowledge layers, and how those layers in time create resources that skills can be built from. Sister investigation to loomworks-closed-loop-engagement-investigation-v0_1.md (filed earlier in the same session) — both touch the cross-time and cross-engagement flow of knowledge that the compounding thesis depends on. Informed by. Methodology v0.20 (Companion identity, Memory expectations, the four-room pipeline). Methodology v0.18 — promotion-requires-operator-approval (the load-bearing rule), comprehensive manifestations, hierarchical domains. Methodology v0.17 — domain hierarchy introduction, skill versus agent distinction. Phase 14 Person Layer Discovery (Domain Expert designation, person-declared / AI-assisted / Operator-recognized domains). Companion Expertise Note v0.1 (the engagement/general boundary). Queued Directions v0.2 §5.1 (Claude Dreaming as a governance-inverted Companion behavior). Playground Reference Design v0.1.2 (the common pool, cross-engagement reference, cross-engagement considerations). Phase 17 (redirect — relocation, not promotion).

1. The framing question

The Operator opened with: As an Operator engages Loomworks, there are many separate and unique Engagements that are going to be created, along with appropriate memory. How do we extract and elevate certain knowledge (memory) to Domain knowledge layers, that in time will create the resources that skills can be created from, providing a comprehensive guide to operational consistency?

The question carries a multi-stage transformation: per-engagement Memory → domain-layer knowledge → resources → skills → operational consistency for new engagements. It also presumes that this transformation is desirable, that it is safe to do, and that there is some pathway through which it actually occurs. The methodology has already settled most of the structural answer; the build has barely begun on it; one terminological distinction needs to be nailed down before the answer is clean.

2. What lands

The pathway exists at the methodology level under the name promotion, and the methodology has been unusually thorough about it. Engagement-scoped Memory accumulates first; the Operator (with the Companion's pattern-recognition assistance) decides which fragments belong at which higher level in a domain hierarchy; promotion is gated by Operator approval and recorded with full provenance; promoted fragments live at their new level under that level's governance; over time, mature levels accumulate enough material to support a comprehensive manifestation — a reusable structural account of a kind of thing — and from there, more specific operational guides (in three distinct senses, §4) can be specified.

The compounding thesis depends on this pathway working. Without it, every engagement starts blank, and the methodology is a single-engagement productivity tool rather than infrastructure for compounding knowledge work.

The build sits at single-engagement Memory with redirect (Phase 17) for cross-engagement relocation. Cross-engagement promotion — the methodology operation — is not yet implemented.

3. Trajectory of the conversation

3.1 The methodology pathway named

The conversation started by laying out what the methodology already commits to: the hierarchical domain structure (domain → sub-domain → sub-sub-domain → engagement layer beneath); the promotion rule (Operator approval required, target level chosen, receiving level governs admission, event recorded with provenance, embargoes and rescindable refusals supported); comprehensive manifestations as reusable references at each level. None of this was a novel contribution from the conversation; it surfaced material the methodology had already settled (v0.13 introduced the hierarchy, v0.17 sharpened it, v0.18 defended the operator-approval rule against the open-source thesis's pull toward auto-flow).

What the conversation did add: the full chain from Memory accumulation to operational guides, including the Companion's role as proposer, and the connection between fragment-level promotion and person-level Domain Expert recognition.

3.2 The "skills" terminology — a distinction worth nailing now

The Operator's question used "skills" without specifying which sense. Three meanings are in active circulation:

• Methodology-skill (v0.13+). A bounded operation performing a structural transformation against well-defined inputs — translation, transcription, hashing, metadata extraction, content selection. Distinct from agents, which are goal-pursuing over multiple steps with bounded autonomy. The methodology's economic claim about skills is that they are commodity (cheap, replaceable), while agents are not (load-bearing for whether the methodology's loops converge).
• SKILL.md-style instructional content. The Anthropic pattern — a directory with a SKILL.md that gives operational guidance for a kind of work. The forge-spec skill, the candidate-seed skill, the foray skill in this project's /mnt/skills/ are examples. These are instructions for an executor, written by someone who has done the work enough times to articulate the pattern.
• Operational consistency guides. Domain-level material that prescribes how similar engagements should run — close in spirit to SKILL.md, but expressed as engagement-domain content for a community rather than as instructions for an LLM executor.

The Operator's phrase "providing a comprehensive guide to operational consistency" reads most strongly as the third sense, possibly with elements of the second. The methodology supports all three as downstream products of elevated knowledge:

• Bounded-operation skills get specified once a domain's transformations stabilize and are recurrent enough to commodify.
• SKILL.md-style instructional guides get specified once the how-to of a kind of work has been articulated enough to instruct an executor.
• Comprehensive manifestations get specified once the structural account of a kind of thing has matured at a domain level.

These are not the same artifact. They serve different roles. A domain that has all three is well-developed; a domain that has only one or two is still maturing.

3.3 The full elevation chain

The chain, end to end:

1. Engagement-scoped Memory accumulates through ordinary contribution (held → committed, with refinement loop, with provenance).
2. The Companion notices patterns — assertions that recur in form across engagements, decisions that look transferable, structural commitments that are not engagement-specific. The pattern-detection capability is what queued directions v0.2 §5.1 named under the Claude Dreaming family: governance-inverted memory consolidation, where the Companion proposes and the Operator approves. Autonomous distillation is rejected for the same reason promotion is gated — most engagements contain confidentiality, professional privilege, regulatory constraints, or unconfirmed observations.
3. The Operator approves promotion of specific fragments to specific levels in the hierarchy. Promotion event lands with provenance: which fragment, from which engagement, by which Operator, to which level, with what rationale, under what embargo (if any). Provenance cannot be stripped.
4. The receiving level's governance accepts, qualifies, or rejects the promotion. Domain-level governance is the level's own seed; what governance looks like is domain-specific (clinical research has its conventions, software has different ones, legal practice has different ones again — methodology defers).
5. Over time, fragments at a domain level cohere into a comprehensive manifestation — the structural account of what a clinical trial protocol contains, what a puppy breeding program covers, what a software architecture review addresses, what an ad-management engagement requires.
6. From a mature comprehensive manifestation, the three downstream artifact types can be specified: bounded-operation skills, SKILL.md-style instructional guides, and structural templates that new engagements draw on as founding Memory.
7. New engagements in the same domain encounter the comprehensive manifestation as founding Memory and the operational guides as available skills. They start informed rather than blank. The compounding loop is closed.

The Companion is the proposer at step 2 and likely the contributor for step 6 (drafting the operational guide once a level matures, subject to operator approval). Steps 3, 4, 5 require human governance — proposing is cheap, deciding is not.

3.4 The personal-identity counterpart — Domain Expert recognition

Fragment-level promotion is one half of the compounding loop. Person-level recognition is the other. The Phase 14 Person Layer Discovery settled that domains are person-declared, AI-assisted, Operator-recognized — a person declares an area of expertise; the Companion surveys the existing domain graph and surfaces topology (overlaps, sub-domain relations, near-duplicates); the person reviews and adopts/extends/distinguishes; recognition lands per-engagement when an Operator confirms the expertise within an engagement.

The Domain Expert designation is persistent — the badge stays earned. The activity dot indicates currency, not authority. This is the asymmetry the UI design protects: recognition compounds (the badge), participation is variable (the dot).

Why this is the counterpart to fragment-level promotion: a person whose expertise gets recognized across many engagements is the human-side mirror of a fragment whose substance gets promoted across many levels. Both are how the compounding works. Both feed the same goal — that the next engagement starts informed rather than blank.

This connects directly to the operational-consistency question. A skill-equivalent guide is most credible when its provenance includes the named Domain Experts whose engagements contributed to its development. Recognition makes the guide trusted, not just available. Without Domain Expert provenance, a comprehensive manifestation is just a document; with it, it carries the weight of a community's recognized expertise.

4. The build state versus the methodology state

4.1 Methodology — settled

• Hierarchical domain structure (domain → sub-domain → sub-sub-domain → engagement).
• Promotion as a discrete event with operator approval, target-level choice, receiving-level governance, recorded provenance, embargo and rescission support.
• Comprehensive manifestations as reusable references at each level.
• Skill versus agent distinction (skills bounded, agents goal-pursuing).
• Domain Expert designation (Phase 14) as personal-identity recognition that compounds across engagements.

4.2 Methodology — partially settled

• Domain-level governance specifics. The methodology defers — clinical research has its conventions, software has different ones, etc. For Loomworks specifically, who governs at each level is still open. A natural shape (filed §6.1): the first Operator to receive a promotion at a level becomes its founding governor, with succession governed by the level's own seed.
• When a comprehensive manifestation crystallizes versus a pile of fragments. "Matures over time" is the methodology's phrase; in product terms there is a real question about the threshold and the act of declaration.

4.3 Build — present

• Single-engagement Memory with full contribution lifecycle (held, committed, retracted, redirected).
• Phase 17 redirect: cross-engagement relocation of a single held fragment. This is not promotion — it moves a fragment, it does not lift it to a higher level.
• Phase 39 cross-engagement aggregation: a dashboard surfacing what is happening across engagements (active jobs, items needing attention, recent artifacts). This is operational visibility, not content elevation.
• Phase 14 Person Layer: Domain Expert designation conceptually settled; UI badge designed; recognition flow not yet wired.
• Phase 41–43: Companion identity and personal memory contribution. The proposer-role substrate exists for personal facts; the cross-engagement-pattern proposer is not yet built.

4.4 Build — not yet

• Domain-layer Memory store. There is no substrate-level place where a promoted fragment lives beyond an engagement. Each engagement's Memory belongs to its own repo (the established discipline); the domain layer would be a new tier above that.
• Promotion pathway. No event kind, no endpoint, no lifecycle for "promote this fragment from engagement A to sub-domain X."
• Hierarchical domain registry in substrate. The hierarchy is a methodology concept and a UI design element; substrate has no domain table beyond what was specified in the Phase 14 person layer.
• Comprehensive manifestation as a multi-engagement artifact. Today's manifestation is a per-engagement organized reading of Memory; the multi-engagement / domain-level version is a different artifact.
• Domain-level governance. No mechanism yet for a level to admit, qualify, or reject a promotion.
• Companion as cross-engagement pattern detector. Phase 43 gave the Companion personal memory; cross-engagement (across-Operator) pattern detection is in queued directions and not built.

5. The Companion's role at each step of the chain

The earlier closed-loop investigation crystallized the Companion as attribution channel for non-human reporters. The elevation pathway extends that role to attribution channel for cross-time, cross-engagement pattern detection — with one important asymmetry.

For fragment promotion within one Operator's engagement portfolio, the Companion is a natural proposer. It already knows the Operator's engagements; it already has read access to engagement Memory; the Operator's authority chain extends naturally. Pattern: Companion notices that an assertion in Engagement A and an assertion in Engagement B share form; proposes a candidate fragment for sub-domain X; Operator reviews the candidate and the rationale; approves or refines; promotion event lands.

For cross-Operator promotion at a domain level — many Operators' engagements feeding a sub-domain — the Companion is not the right proposer. A Companion is bound to one person; cross-Operator work is a steward role, not a Companion role. This is genuinely open: who or what proposes promotions when the Operators of multiple contributing engagements have to coordinate? Possibilities:

• A domain steward — a designated Domain Expert at the sub-domain who watches for promotion candidates from member engagements, with their own Companion serving the steward role for that work.
• A domain-level agent — distinct from the per-person Companion, scoped to the domain rather than to a person.
• Self-nomination — Operators of engagements contributing similar fragments nominate them for promotion themselves, with the receiving level's governance evaluating.

The methodology does not yet pick. Worth filing.

6. Filed for consideration

6.1 Domain-level governance for Loomworks

The methodology defers. The build will need an answer when the domain layer ships. A natural-shape proposal: the first Operator to receive a promotion at a level becomes its founding governor; the level acquires its own seed and its own membership rules; succession is governed by that seed. This makes domain-level governance recursive — the methodology applied to the methodology's own structure.

Alternative considered: a global domain registry maintained by the Loomworks platform itself. Rejected — top-down taxonomy was already considered and rejected in the Phase 14 person layer Discovery (Decision 5). The same logic applies at the fragment-promotion level: the system surfaces topology, the human governs.

6.2 The crystallization-of-a-comprehensive-manifestation moment

Methodology says "matures over time"; in product terms there is a moment where accumulated promoted fragments cohere enough to warrant being called a comprehensive manifestation rather than a pile. Possibilities for what triggers the moment:

• Operator declaration by a domain steward — "this is now mature enough to call a comprehensive manifestation; here is its scope, here is its membership of fragments, here is its first version."
• Companion proposal — the Companion (or a domain-level agent) proposes "the fragments at this level have reached a coherent shape; want to declare a manifestation?"
• Ad-hoc emergence — there is no declaration moment; the manifestation just is the current organized reading of fragments at the level, and "comprehensive" is a quality judgment rather than a state transition.

The methodology leans toward emergence. Product surfaces will likely need a more discrete moment for the Operator to point at and say "the manifestation now exists." Probably both — emergent in fact, declared in surface.

6.3 The three-skill-meanings clarification in methodology vocabulary

The methodology already names skill (bounded operation) versus agent (goal-pursuing). The SKILL.md-style instructional content is not yet named in methodology terms — it is an Anthropic pattern that has been adopted ambient in the build (forge-spec, candidate-seed, foray, sae-ops). When elevation pathways start producing operational guides, the vocabulary will need a clean term that distinguishes the instructional artifact from the bounded operation. Worth filing as a vocabulary correction for a future methodology revision.

Candidate names for the instructional artifact: playbook, operational guide, executor brief, practitioner skill. None obviously dominant. The conversation deferred picking.

6.4 Cross-Operator pattern detection — the steward role

The cross-Operator promotion-proposer question (§5) is genuinely open. The asymmetry between Companion-as-personal-proposer and steward-as-cross-Operator-proposer is structural, not a gap to be papered over. When the build gets to multi-Operator domains, this needs a clean answer. The agent-fabric pattern from the closed-loop investigation suggests one shape: domain-level task agents that watch for patterns and report to domain stewards; stewards (or their Companions) translate to promotion proposals. Filed for when that build phase arrives.

7. Where the discipline holds

• Operator gating on promotion. Auto-elevation is the failure mode that makes the methodology unusable in serious work. The Companion proposes; the Operator approves; the level governs. The structure is load-bearing.
• Provenance at every step. Promotion events carry full lineage; the originating Operator's role stays in provenance permanently even after the fragment lives at its new level under different governance.
• Embargo and rescission. Promotion can be timed, blocked, or refused. Operator authority over timing is preserved.
• Engagement independence preserved. Promotion does not reach back into the originating engagement; the engagement's Memory continues to belong to its own repo. The promoted fragment lives at the new level, with provenance pointing back, but the engagement's own Memory is unchanged.
• Recognition compounds. The Domain Expert designation persists; the badge does not expire. Activity is current; recognition is accumulated. The asymmetry is what makes the compounding visible.

8. Open questions to watch as this matures

• Q1. Who governs at each level when the level first comes into existence? §6.1 proposes founding-Operator-governance with recursive seeding; alternatives exist.
• Q2. How does a Domain Expert designation move from per-engagement recognition to multi-engagement recognition? Phase 14 settled the per-engagement case; the cross-engagement aggregation that produces the persistent badge is not yet wired.
• Q3. Is there a "negative promotion" — fragments that should explicitly not be elevated? Confidentiality, regulatory constraint, professional privilege all give reasons. Methodology supports embargo and refusal but does not yet have a "do-not-promote" marker on fragments themselves. Worth thinking about for engagements that start under known constraint.
• Q4. When a promoted fragment is later corrected or retracted at the engagement level, does the correction propagate to the level it was promoted to? Methodology's non-erasure discipline suggests both versions should be visible, with the lineage recorded; the propagation mechanics are not specified.
• Q5. How does this interact with FORAY? Promotions are themselves transactions — fragment X moved from engagement A's substrate to sub-domain X under operator Y's authority at time Z. The audit case is real. Whether each promotion is its own FORAY transaction, or whether promotions batch under domain-level audit, is open.
• Q6. How does the elevation pathway interact with the closed-loop engagement pattern (sister investigation, filed earlier this session)? A closed-loop engagement that has run many cycles is a particularly rich source of promotable knowledge — the loop produces its own evidence base. The elevation pathway should accept closed-loop residue as one source among many; whether it warrants special handling is open.

9. What this investigation does not produce

• A change request. No phase falls out of this directly. The build is multiple arcs away from cross-engagement promotion. Closing on the present arc (Companion identity, personal memory, proactive behavior, delegation, Operator Layer frontend) takes priority.
• A methodology revision. The methodology already settles most of what this investigation surfaces. A future revision (v0.21+) might absorb the three-skill-meanings clarification (§6.3) and the cross-Operator steward question (§6.4) when the answers mature.
• A vocabulary settlement on the SKILL.md-style artifact. Filed §6.3; deferred.
• An implementation plan for the domain layer. When the build approaches it, a fresh investigation will be needed — the present one is conceptual scaffolding.

10. Connection to the closed-loop investigation

This investigation and loomworks-closed-loop-engagement-investigation-v0_1.md are sister documents. Both touch the cross-time, cross-engagement flow that the compounding thesis depends on. The closed-loop investigation is about how outcomes flow back into the same engagement's Memory — through the Companion as attribution channel, with held + committed lifecycle. The present investigation is about how fragments flow up from one engagement's Memory to higher levels in a domain hierarchy — through Operator-gated promotion, with full provenance, with the receiving level's governance as the second gate.

Together they describe the two directions in which Memory moves once an engagement is running: backward in time (outcomes informing future renders within the engagement) and upward in scope (knowledge informing future engagements at the domain level). Both pathways converge on the same discipline: the Companion as proposer, the human as approver, the substrate as recorder.

11. Provenance notes

This investigation was produced through a single conversation turn responding to the Operator's framing question. The methodology layer did most of the heavy lifting — v0.13 through v0.18 had already settled the structural answer; the conversation's contribution was assembling the full chain end to end, surfacing the skill-vocabulary distinction, connecting fragment-level promotion to person-level Domain Expert recognition, and filing the open questions.

The chain assembly itself (steps 1 through 7 in §3.3) is the most synthesizing piece — the methodology names each component but does not run them as a sequence. Naming the sequence makes the build state's distance from the methodology state legible: the methodology has six steps named; the build has zero of them implemented at the cross-engagement level. That gap is not a critique of the build — it is a recognition that the present arcs (Companion, personal memory, proactive behavior, delegation) are necessary precursors to the cross-engagement work and have to land first.

DUNIN7 — Done In Seven LLC — Miami, Florida Loomworks — Knowledge Elevation Pathway Investigation — v0.1 — 2026-05-08