The open stack a Delta actually deploys.

Cockpit over the matching engine: a demand↔supply board, per-person and per-team next best steps, milestone at-risk flags, and an LLM-drafted staffing update. Move a lever and the plan re-computes.

Vet one candidate on the three criteria that predict on-the-job performance — reliable, technically competent, resourceful — for a defensible GO / NO-GO. Grounded in Schmidt & Hunter (1998).

Score yourself against the real Anthropic & OpenAI FDE bar (knowledge · skills · mindset · habits · portfolio) → the exact gaps to close. Same rubric an mcp-fde-readiness MCP server exposes to agents — one source of truth, two surfaces.

Try a custom access-control layer: give a subject a role on a project, request an action on a resource, watch the Policy Decision Point allow or deny — deny-by-default, project-scoped. Client-side, parity-checked against the tested rbac-mlrun-demo PDP.

One interactive graph merged from many sources — 5 research notes + retrieved Wikipedia/GitHub definitions — where every node is tagged and colored by its source. Click a node to see its definition and origin; filter by source in the legend.

Score a RAG/agent eval set: precision@k, recall@k, MRR, hit-rate + a grounding/hallucination proxy + citation coverage, with an optional gate.

Score a draft against the TRUE rubric (Transferable, Reusable, Understandable, Evaluated — 0–3 each) and apply the ship gate.

Score any artifact against a list of binary, mechanically-checkable pass/fail criteria → a 0–1 score + a gate.

Turn a sequence of scored versions into a kept winner + a run-log (Round / Change / Score / Verdict) — the self-fixing artifact loop.

Run FDE-style discovery on any messy problem → a structured Site Survey (the political terrain, the data reality, the smallest end-to-end win).

Reconstruct how a “yes” actually happens inside an org — from partial signals, before they can say it. Scores adoption-readiness, infers the decision-workflow archetype + its traps, and hands you the probes to ask next. Because the biggest failures are workflow failures, not model failures.

Turn a Site Survey into a proposed agent architecture — the bridge from discovery to “architect agentic solutions.”

Parse messy enterprise docs — DOCX with merged tables and track-changes, XLSX with shared strings — into one canonical structured representation, then gate the parse itself (coverage · structure · fidelity). The moat a real regulated engagement named: output quality ≤ input representation quality.

A minimal, dependency-free MCP server (stdio, JSON-RPC 2.0) that exposes the toolkit's seven request→result skills — true_score, rag_eval, criteria_score, eval_loop, invisible_workflow_map, jd_compile, doc_gate — as callable tools to any MCP host (Claude Desktop, Claude Code).

Will it get adopted? (invisible-workflow-mapper) AND does it work? (rag-eval-harness) → one GO / NO-GO verdict. Ship only if both pass; an unassessed stage blocks GO. No new scoring — it just sequences the skills.

Compile a job description into structured competency knowledge — which FDE clusters it needs, the specific tools it names, its level — and aggregate many JDs (Google, Reflection, CVS) into a cross-company demand matrix that grows the knowledge base. A JD is the input.

Turn a local, heterogeneous prose research vault (Markdown notes, synthesis docs) into a navigable knowledge spine — deterministic, no network.

Scaffold and lint a Delta Field Kit to convention (Checklist / SOP / MCP-spec / …) so a Field Manual becomes a reusable agent skill.