Runtime Title Specialization Handoff

What Changed

The runtime payload builder now has an ordered deterministic resolver in src/aina_data_engine/runtime_payloads.py. It uses title text to assign sidecar runtime functions when upstream function data is missing, generic, or plainly wrong, while preserving the original source_function and a function_resolution record for every change.

The resolver now covers sales, finance, healthcare, technology, retail operations, hospitality/travel, supply chain, manufacturing, quality/safety, facilities, real estate, community services, personal services, engineering/hardware, legal/compliance, public safety, research/science, leadership strategy, people/HR, customer success, marketing, product, data/analytics, design/creative, education, operations, administration, and strategy consulting.

The generated learner and tutor payloads also gained workflow language for the new functions, so titles like Store Manager, System Administrator, Primary Care Physician, Teller, Executive Chef, Police Officer, and Postdoctoral Researcher no longer receive generic business exercises.

Before And After

Current Function Coverage

Semantic Sanity Check

I inspected 50 changed runtime rows with these columns: title, source function, resolved function, deterministic reason, and first module goal. The sample looked semantically coherent from learner, tutor, platform, and evaluator perspectives.

The remaining 12 general_business serviceable rows are deliberately broad: Entry Level Professionals, Associate, Intern, Community Manager, Team Lead, Referee, Summer Intern, Immediate Entry Level Opportunity No Experience Needed, Experience Management-Senior Manager, Cognitive Performance Specialist, Manager, and Attendant.

Artifact Inventory

Generated runtime artifacts live under /srv/aina/aina-data-engine-room/artifacts/validation/. The artifacts/ directory is ignored by default, so this checkpoint explicitly preserves the runtime v1 outputs with git add -f.

Validation

.venv/bin/python -m ruff check src tests
.venv/bin/python -m pytest -q

200 passed in 92.88s
runtime evals: 1,000 rows · 965 local serviceable · 35 blocked · 0 failing

What This Means In Reality

As of this checkpoint, the data engine can generate local synthetic runtime payloads for a much broader and more realistic set of ICP titles. It can serve role-native first-module goals and evaluator cases for frontline retail, banking, healthcare, technology, food/hospitality, manufacturing, safety, real estate, leadership, research, and other domains instead of collapsing them into generic business work.

It still cannot truthfully claim production readiness or real-user personalization. All outputs remain local-only, source-preserving, and blocked from external writes, real-user runtime, and production claims. The next trust move is not more broad deterministic mapping; it is semantic adjudication of the 526 follow-up rows, source/BLS enrichment for the 12 residual broad rows, and packet-quality hardening for the 295 packet candidates.

Next Best Slice

Start with the semantic follow-up queue and prioritize high-volume, high-impact domains. Review the 484 deterministic function changes with multi-LLM adjudication, resolve the 12 residual broad titles using source context, attach BLS/SOC context where it helps, build packet-quality fixtures for the 295 packet candidates, and keep the 670 caveat-service rows available for local testing with explicit caveats.