The context graph, made legible and kept current

TL;DR

The agent's answers are only as trustworthy as the graph behind them, and that graph was invisible. I designed a dashboard around the four questions a skeptic asks before delegating:

• What does it know?
• How does it stay current?
• What has it been doing?
• Is it reliable?

Plus a real-time ingestion feed so you can watch the knowledge stay fresh.

The problem

A promise that the graph exists is not trust. Users could not see what the agent knew, whether it was current, or whether ingestion was even working.

Ingestion is a live pipeline: cloud resources change, and every git push should re-ingest code into the graph. If that pipeline stalls silently, the agent quietly answers from stale reality, which is the most dangerous failure of all.

The insight

A dashboard earns trust by answering questions, not by displaying metrics, and freshness has to be a first-class, observable fact rather than an assumption.

Ingestion is a stream of jobs, so I designed it to be watched like one.

The solution

• What it knows: the IaC trinity (Code, State, Runtime) as a scannable grid, big number plus breakdown, so the structure mirrors how infra engineers already think.

terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 planstate + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { }vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 planstate + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { }vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9What it knows: the semantic context dashboard

• How it stays current: a live WebSocket sync view (context sources sync) plus a sidebar ingestion indicator and feed popover. Jobs stream their status and timing in real time, per source (cloud, code, k8s), with running jobs surfaced and a last new data X ago freshness signal. When you push code, you can watch it re-ingest into the graph.

terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 planstate + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { }vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 planstate + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { }vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 planstate + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { }vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9How it stays current: live ingestion across sources

• Is it reliable: reliability stated in plain English (for example half of replies are faster) instead of p50/p95 jargon.

terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 planstate + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { }vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 planstate + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9 vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { }vpc deny state + ~ terraform module ingest lock 0 1 apply drift graph { } 9fa3 plan iam helm => b4f aws_s3 allow k8s == 2f9What it has been doing: the activity view

Cross-functional reality

I designed the ingestion UX end to end; our frontend lead implemented the feed, and it plugs into the backend team's ingestion-observability API and git-push ingest stream.

Designing it meant defining the job and event model with the backend so the UI had something honest to render.

Impact

Turned the agent's grounding from a claim into something you can audit at a glance, and turned a silent backend pipeline into a visible, trustworthy one.

Reflection

The four-question framing was the unlock: it turned an admin dashboard into the reason to trust the agent.

Watching ingestion happen live is what makes people believe the graph is real.