A knowledge graph for product thinking · v0.1

Never lose the thread between a signal and the work you eventually shipped.

gx structures your product thinking into a traceable graph — the observation that started a debate, the need it implied, the work it became, the decisions made along the way. All linked. All recoverable. Months later, the argument is still legible.

Get startedInstall MCP →
01 / Premise

Most product work happens twice: once in the conversation that surfaced the idea, and again — months later — when someone tries to remember why. The chain breaks; the argument has to be re-fought from worse evidence.

gx keeps the chain intact. Signals (something you observed), needs (what they imply), work (what you decided to do), decisions (the constraints along the way) — all linked into a graph that refuses to let work float without a reason behind it.

That refusal is the point. It's how the argument stays explainable.

02 / Surface
iCapture

Capture

Observations land as signals — terse, dated, attributed. You don't structure them. Your agent does.

iiTriage

Triage

An opinionated read on what to work on next, computed from the graph itself. No kanban. No lanes.

iiiPlan

Plan

Stakeholder-facing rendered subgraph: the signals that fed in, the needs addressed, the work proposed.

03 / Output

A real-shape sample.

This is what gx produces from a ten-minute capture session. No templates filled in by hand — every node was authored by an agent talking to a human.

Initiative

Reduce driver-side trip cancellations on shared rides

activeexperimentCreated 4/15/20263 signals · 2 needs · 3 work

We're seeing higher cancellation rates on shared rides than solo rides, and ops thinks it's hurting matched-rider experience. We want to test whether changing the way drivers see shared-ride pickup details reduces cancellations within 4 weeks.

Hypothesis
If drivers see a shared-ride badge and second-stop info before accepting, post-match cancellation rate on shared rides will drop by 30% within 4 weeks.
Validation criteria
Treatment group cancellation rate is at least 25% lower than control, p < 0.05, with no >5% drop in match acceptance rate.
Assumptions
Drivers cancel because of unexpected length, not because shared rides are fundamentally less profitable. Driver incentives unchanged.
Out of scope
Rider-side UI changes. Pricing changes. Onboarding for existing drivers (only new in this iteration).
iSignals

The evidence behind this initiative, in time order.

  1. 4/15/2026
    Ops weekly review (Apr 8): shared-ride cancellation rate is 2.3× solo

    Reviewed in ops weekly. Driver cancels are concentrated in the 60s after match. Ops believes drivers are seeing the second pickup and pulling out.

  2. 4/15/2026
    Driver feedback in app store reviews mentioning 'two stops'

    37 reviews in March mentioning 'two stops' or 'pool' negatively. Sample: 'didn't know it was a pool until I accepted, lost half my hour'.

  3. 4/15/2026
    Rider chat: 'driver canceled twice in a row'

    Tier 1 support ticket trend — 4× increase in shared-ride re-match requests since the new matching launch.

iiNeeds

What the signals suggest we have to address.

Drivers need shared-ride context before accepting
Today, the second pickup point is hidden until after acceptance. Drivers feel deceived and bail. We need to give them enough info to opt in honestly.
Riders need predictable shared-ride pickup
The cascading re-match effect when a driver cancels is much worse on shared rides. Reducing initial cancellation also reduces compounded delay.
iiiWork

Sorted by priority. Each item traces back to the needs above.

  • Show shared-ride badge + second-stop ETA on match screennow

    Add a badge to the match-acceptance screen indicating 'Shared (2 stops)' and the projected total trip time including the second pickup.

    Constrained by: Don't show absolute trip duration — show relative pickup distance only

  • Run a 4-week A/B test in the SF and Austin marketsnext

    Split drivers 50/50 in two markets. Measure post-match cancellation rate, completion rate, driver hourly earnings.

  • Add 'view trip details before accepting' settinglater

    Opt-in driver setting that surfaces both pickup points before the accept tap. Surface as part of onboarding for new drivers.

ivKey decisions

Constraints and choices shaping this work.

Don't show absolute trip duration — show relative pickup distance only
Legal flagged showing total trip duration during match as a potential commitment. We'll show 'second pickup +5 min away' instead of 'total 35 min'.
Made with gx.
Log inDocs