Article: ThreadWeaver v3 launched its Causal Work Graph, a cross-tool lineage engine that lets AI-driven queries return not just a document but a provable chain of evidence linking Slack chats, Jira tickets, GitHub commits and other artifacts. Teams that adopt it can answer “Why was this built?” with a visible subgraph instead of a speculative paragraph.

The context: scattered data, missing connections

Engineering groups today live in a patchwork of platforms. A customer complaint lives in a ticketing system, the ensuing discussion lives in a chat app, the design decision lives in a project-management board, the code lives in a repository, and the release notes sit in a documentation tool. The raw information is there, but the causal links between them are invisible. When a product manager asks why a feature was released, the answer hides in a web of messages, issues and commits. Traditional search tools can surface items that contain similar keywords, but they cannot tell which item actually triggered the next.

Why it matters: provenance versus hallucination

Most large language models (LLMs) answer by matching semantic similarity. A Jira ticket that mentions a Slack channel may appear related, yet the model cannot prove that the chat caused the ticket. The result is “hallucination” – an answer that sounds plausible but lacks a verifiable source. In regulated environments, or whenever accountability matters, that gap is costly. The Causal Work Graph replaces guesswork with a graph whose edges are backed by concrete proof: timestamps, actor identifiers, relationship types and confidence scores.

How the Causal Work Graph works

  • Event-centric modeling – each node represents an event (e.g., a Slack message, a Jira issue creation) rather than a static document.
  • Explicit relationships – edges encode the specific causal claim (“Slack discussion informs PM decision”) together with the supporting evidence.
  • Provenance metadata – every edge stores the source, target, timestamp, actor, confidence level and a pointer to the original artifact that substantiates the claim.
  • Uncertainty handling – if the system cannot locate a linking event, it returns “Unknown” instead of fabricating a connection.
  • Permission-aware exposure – users only see edges whose underlying artifacts they are authorized to view; a missing Slack message simply hides the corresponding edge.
  • LLM as interpreter, not repository – the language model translates the graph into natural-language explanations, while the graph itself remains the authoritative source of truth.

When a user asks, “What led to the release?” the engine assembles a subgraph that might look like:

  1. Customer complaint → Slack discussion (timestamp, user)
  2. Slack discussion → PM decision (Jira ticket)
  3. PM decision → GitHub commit (code change)
  4. GitHub commit → Release (artifact)

The response includes links to the exact Slack message and Jira comment, letting the asker verify each step.

Takeaway

ThreadWeaver v3’s Causal Work Graph turns scattered engineering artifacts into a single, auditable chain of cause and effect. By demanding evidence for every link, it sidesteps the hallucinations that plague pure-LLM answers and gives teams a concrete way to trace the “why” behind every release.