06-reference/research

move1 anchor screen oi skill catalog

2026-09-06·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)·! phdata-internal-proprietary
productizationorganizational-intelligenceanchor-offerportfolio-scoringphdata

Only five things in the 106-skill catalog are problems a client would name unprompted — and the July refactor cut two of them

The question

Verbatim: "Test the move-1 'urgent + recognized problem' anchor screen against CAF's 106 skills — which ~5 clear the bar?"

Context: this is the third open follow-up from [[2026-06-28-productized-consulting-scalable-anchor-transition]], and the companion open checkbox in [[2026-06-25-productize-framework-armstrong-vecteris]]. Naming convention note for traceability: "CAF" was retired as a name on 2026-08-10; the work now runs under the Organizational Intelligence (OI) umbrella (Organizational Platform → Organizational Map · Intelligence Platform → Intelligence Maturity Assessment · OIP + Pulse). The 106-skill artifact this question asks about is the caf-engine skill library, which predates the rename.

What we already know (from the vault)

Verification of the count — 106 is real, but it is a July snapshot

Counted directly against the local checkout at ~/Projects/phdata-private/phdata-ai-wf-plugins-dd673d8e215f/plugins/caf/skills (dated 2026-07-03): exactly 106 skill directories. The backlog row's number is correct as of that tree. Three corrections ride on top of it:

  1. The live surface is 70, not 106. Founder, 2026-07-27: "We have also refactored CAF. We cut out phases 5-8" ([[2026-06-28-caf-8-phase-structure-and-skill-pipeline-mapping]], correction block; confirmed in [[2026-07-27-caf-technical-architecture-and-backlog]]). The phase-5–8 prefixes are 6a (5) + 6b (6) + 6c (5) + 6d (5) + s5 (7) + s6 (5) + s7 (3) = 36 skills cut. Phases 1–4 plus the Meta-Council leave 1+2+3+4a+4b+4c+4d+4g+s4+m = 70.
  2. A second, non-interoperable tree exists. The same repo carries a project-scoped .claude/skills implementation with 12 entries, where the ~106 dimensions live as reference/*.md inside 8 phase skills ([[2026-07-01-caf-ecosystem-map-and-brigade-restructure-read]]). "The catalog" is ambiguous unless you say which tree.
  3. The current OI spec carries no catalog of this shape at all. ~/Documents/phdata-projects/organizational-intelligence/spec/ is a product book plus registries: the Capability Registry is 6 layers → 28 sub-capabilities → 159 capability details, alongside a Role Catalog and Value Registry. Whether caf-engine still executes underneath the OI deliverables is not verified here [assumption flagged] — the OI repo does not reference it.

What the web says

Those three are the only problem statements in this category with independent, quantified evidence of buyer-side recognition in 2026. That is not decoration — it is the evidence base for gate R below.

Convergences and contradictions

Synthesis for RDCO

The scoring rule (my operationalization, not the founder's, not Armstrong's). Armstrong states the screen prosaically; it is not a rubric anywhere in the vault. I turned it into four gates scored 0 / 1 / 2, and I am labeling it as mine:

Gate Passes at 2 when…
U — Urgent the client feels it on a clock: a budget cycle, a board question, a stalled go-live, a regulator
E — Expensive not solving it costs six figures or more in wasted build, wasted licence, rework, or penalty
R — Recognized the buyer states the problem in their own words before we arrive. The killer gate
S — Standardizable sellable as fixed-scope, fixed-price, identical deliverable every time, without the rest of the pipeline running first

Clears the bar = U, E and R all at 2, plus S ≥ 1 (total ≥ 7/8). One methodological finding first: no single skill clears S at 2 on its own. These are sub-atomic consultant micro-steps, not offers. A sellable anchor is a tight cluster of 2–3, so each result below names a cluster and its lead skill.

The five that clear (Ray-selected, awaiting the founder's read).

  1. Adoption Risk Screen — lead 4a-2-adoption-modeler, with 4a-6-workflow-intrusion-analyst and 6a-2-adoption-monitor. 8/8. "We bought it and nobody uses it" is the single most-stated buyer complaint of 2026 (80% embed vs 31% in production), it is urgent at renewal, and the sunk licence plus build cost is the expense. Fully standardizable: score a proposed feature for behavioral friction, workflow disruption, champion presence, install barriers, return a Confidence number.
  2. AI Portfolio Value Screen — lead 2-7-value-screener, with 4a-3-time-value-quantifier and 4b-6-risk-adjusted-value. 8/8. The skill's own description is "prevent innovation theater," which is the CFO's question verbatim in a year when Gartner forecasts 40%+ cancellations for unclear ROI. Standardizable as a fixed-scope pass over an existing initiative list.
  3. Agent Correctness & Eval Harness — lead 6d-8-adversarial-correctness-validator, with 6b-6-evaluation-harness. 8/8. Independent re-execution, magnitude check, lineage cross-reference, freshness check on generated SQL, code and narrative. "How do we know the answer is right" now blocks go-live and is the concrete form of the agent-washing due-diligence step. Both skills are in the cut 36.
  4. AI Oversight & Accountability Boundary — lead 2-5-hitl-boundary, with 3-9-compliance-scanner and 4c-1-accountability-mapper. 7/8 (S=1). Who is liable, what is explainable, what do we show the auditor. Urgent because it gates production in regulated industries, expensive because a stop-ship late is the worst-cost outcome. S is 1 because the deliverable varies materially by regulator, so it will not sell perfectly identically.
  5. Domain Rule Extraction — 3-9c-domain-rule-extractor, standalone. 7/8 (U=1). Clients say "all the rules live in Dave's head" unprompted, so R is a clean 2, and the skill's own text calls it "the single highest-ROI addition in v3.0 for correctness" — silently-wrong outputs are the expense. U is 1 because urgency is event-driven: a retirement, an attrition, a migration. No event, no clock.

The next-closest, and the gate each one fails. 2-1-agentic-maturity-assessor (7/8, fails the unwritten distinctiveness gate — everyone sells it). 3-6-value-leakage (U=1: "quantifies loss without implying feasibility," and a diagnostic with no action attached does not create urgency to buy). 2-3-data-readiness-interface-steward (U=0: "our data isn't ready" is the most-recognized and least-urgent problem in the category, perennially deferred and thoroughly commoditized). 3-4-exception-miner (R=0 by construction — it surfaces hidden complexity, and a client cannot recognize a problem defined as hidden). 6c-4-drift-monitor and 6b-4-failure-analyst (R arrives only after deployment, making these expansion offers rather than anchors).

What the screen actually revealed, which is not the list of five. The screen does not sort the catalog by phase, by quality, or by effort. It sorts by direction of gaze. Roughly 95 of the 106 answer a question the consultant has — classify this input, propagate this tag, harmonize these metrics, validate this contract. The handful that clear name a failure the client has already lived through: money spent, nobody used it, the answer was wrong, the auditor asked. That is the whole finding, and it is a positioning result rather than a prioritization result. It also explains the adoption number in [[2026-07-01-caf-ecosystem-map-and-brigade-restructure-read]] — six hours per company against colleagues' thirty minutes on the old monolith. A pipeline built almost entirely of consultant-facing micro-steps optimizes for consultant rigor, and rigor is not what the buyer is paying to remove.

The uncomfortable structural consequence. The 2026-07-27 refactor cut 36 of 106 skills, and that cut is not neutral against this screen: two of my five anchors sit wholly or partly inside the cut set, and three of the six next-closest do too. The reason is structural rather than accidental. Urgent, expensive, recognized problems in this category are overwhelmingly post-build problems, and phases 1–4 terminate at a plan. Truncating at the C4 Build Manifest optimizes the framework for the pre-sales motion, which is a defensible call for a sales-facing marketplace, and it simultaneously removes the part of the catalog with the strongest independent buyer demand. Both things are true. The productization-honest read is that OI's front half is a lead product and its anchor-offer candidates live in the half that was cut, so if anchor revenue is the goal, the cut half needs a home rather than a deletion.

Where this lands against the parent brief's move-3 conclusion. The parent argued RDCO's binding constraint is move 3, decoupling delivery from founder hours, and that the transition is the 37signals shape rather than a firm conversion. This test does not disturb that. It sharpens the move-1 input: the five anchors above are the only defensible candidates for the "isolate the handful" instruction, and four of the five are outcome verification offers (did it get adopted, did it pay, is it right, is it defensible) rather than assessment offers. Assessment is what the catalog is mostly made of and it is what the market has commoditized. Verification is what the market has started to recognize and has not yet standardized. If there is a productizable anchor anywhere in this catalog, that is the shape of it.

Why this is in the vault

This closes the third open follow-up in [[2026-06-28-productized-consulting-scalable-anchor-transition]] and the "which of the 106 skills clear the urgent + expensive bar" checkbox in [[2026-06-25-productize-framework-armstrong-vecteris]], and it supplies the Customer Need (×30) input that the still-unbuilt weighted Portfolio Scoring scorecard needs before it can rank the OI bet against Squarely, MAC and Sanity Check. It also corrects a live factual drift the founder will otherwise carry into that scorecard: the catalog's live surface is 70 skills post-refactor, not 106.

Open follow-ups

Related

Sources

Vault:

Primary artifacts (local, phData-internal, not vault docs):

Web: