Snapshot from · monthly cadence
ENGINEERING VIEW
5 PRs in flight — none tied to a top customer ask.
And $1.1M of top demand has nothing being built. The drift runs both ways.
Building what no one asked for
5 PRs in flight map to no top demand — speculative or internal work.
- spike: on-device model inference prototype Engineer-initiated spike; no demand item or sales signal requested local inference.
- chore(obs): migrate logs to the OTLP exporter Observability plumbing — not traceable to any customer ask.
- feat(infra): experimental vector cache for embeddings Speculative perf work; no demand theme above the weight threshold mentions caching or latency.
- refactor(queue): migrate the grading worker to the new job runner Internal refactor — no customer demand item references the queue runtime.
- feat(billing): proration edge-case handling for mid-cycle plan swaps Billing correctness fix; not surfaced by any customer demand evidence this snapshot.
Top demand with nothing being built
3 top asks — including the #1 — have zero build in flight. Click any for the customer signal.
Slack thread summarization for the support inbox
-
“Our support team re-reads 40-message threads every morning. A per-thread summary would save us an hour a day.”
+3 more from customers
Check your own PRs →Add Notion as a first-party connector
-
“Half our product specs live in Notion. We keep pasting links by hand — a first-party Notion connector would close the loop.”
+3 more from customers
Check your own PRs →Audit log retention policy controls
-
“Enterprise security review flagged that audit logs can't be set to a 1-year retention window.”
+2 more from customers
Check your own PRs →Re-prioritization suggestion
◆ SUGGESTED SWAP · synthetic model output
Stop the experimental vector cache spike [Assumption]
Start Slack thread summarization for the support inbox [Observation]
Because it's the single highest-revenue customer ask ($456K of ARR) with zero builds in flight, while the vector cache maps to no customer demand this snapshot [Hypothesis]
Every claim is tagged so you can tell hypothesis from observation. Synthetic output, hand-written for this demo — the live loop generates it weekly.