Why "why did our infra costs jump in Q2?" doesn't fit a graph

Описание и применение

В статье описывается, как фреймворк ctxloom заменяет фиксированные графы на реактивные артефакты, позволяя динамически обрабатывать запросы о расходах инфраструктуры без скрытых ветвей и обеспечивая воспроизводимость и аудит.

Задача

Answering questions like "Why did our infra costs jump in Q2?" requires multiple heterogeneous data sources and computations, but typical graph frameworks force pre‑defined paths or hidden branching, making the workflow brittle and opaque.

Результат

The ctxloom approach eliminates the need for a fixed execution graph, enabling dynamic, reproducible, and auditable workflows that separate deterministic calculations from LLM reasoning.

Рабочий процесс

• Receive Question
• search_docs produces SourceRef
• search_table produces Spreadsheet
• resolver produces TypedDoc
• evidence_builder produces Evidence
• table_resolver produces Spreadsheet
• calculator produces Calculation
• verifier produces Claim
• answer_builder consumes Claim, Calculation, Evidence to produce Answer

Доказательства

ctxloom has no execution graph — it has artifacts (typed, versioned objects) and agents that react to their appearance.

Источник

The same question goes through text search, table search, a deterministic calculation, and verification — in a single run.

Источник