Skip to the story
SubtextA map of attention
00Session replay, built for agents.

Capture everything. Read what matters.

One captured session. 950,931 tokens at full detail. Your agent shouldn’t have to read it all.

Scroll to follow the evidence
One captured session · actual debugger totals
950,931

tokens at full detail.

DetailAll signal kinds + errors
950,931
StandardAll signal kinds + errors
261,963
DigestNavigation, interactions + errors
14,045

Same 0–13,182-second range in every row. These are signal-context totals from one session, not video-token estimates or a benchmark.

01Choose the detail

Keep the time. Change the detail.

Same time range. Different kinds and grains of detail. Try the controls and see how much less the agent needs to read.

Same full time range0–13,182s
Context returned
tokens
Choose a recorded configuration
navigationinteractionnetworkconsoleerror
Standard merges repeats. Detail includes payloads. The time range and kinds stay fixed.

Try a configuration. Actual debugger totals from one session.

02Now narrow the window

Three clicks. Still stuck.

An error tag in the checkout journey gives the agent a place to look. The agent reviews the surrounding 47 seconds.

Checkout journey
47seconds
14:12 14:59
A 15-minute visit
00:0015:00
AddressValidationErrorerror tag · 3 Continue clicks nearby
THE NEXT TOOL CALLOpen the window around the error.
03Follow the signals

The requests pass. The shopper is stuck.

Delivery options loaded. Payment was never requested. The failure sits between those two facts.

47 seconds · selected signals/checkout

GET /delivery-options200

Continue to paymentclick

AddressValidationErrorerror

Continue to paymentretry

Continue to paymentretry

POST /paymentnot sent
WHAT THE AGENT KNOWS SO FARWhy didn’t the shopper correct the field?
04Pixels and context

The error exists. It’s hidden.

Screenshots show what the shopper saw. The Sightmap-enriched accessibility tree names the components and exposes the error hidden beneath the checkout bar.

One moment. Two views.
What the shopper sees
Mobile checkout: the fixed Continue bar covers the ZIP validation message.
Fixed checkout bar
The bar covers the error.
Accessibility tree Sightmap enriched
[ShippingForm]
  [ZipCodeInput] value="9410"
  [AddressValidationError]
    role: alert · tag: error
    Enter a 5-digit ZIP code
    to continue.

[StickyCheckoutBar]
  [ContinueButton]
Error boundsy 541–593
Bar boundsy 518–660
Fixed bar
Hidden message

Present in the tree. Covered on screen.

05Evidence in the pull request

A fix reviewers can see.

Your agent carries Subtext screenshots into the pull request, showing what went wrong and how the change makes the error visible.

fieldwork / storefrontExample pull request

Keep ZIP validation above the checkout bar

OpenYour agent proposes a fix to main
Your agentbot

The fixed bar covered the ZIP error. This change reserves space below the form and scrolls the invalid field into view.

Screenshot evidence
BeforeError covered
Mobile checkout: the fixed Continue bar covers the ZIP validation message.
AfterError visible
Corrected checkout: the ZIP validation message is visible above the Continue bar.
Error visible at 390 × 660Try the fix
06Zoom out to other sessions

Who else hit this page?

review-search finds other sessions that reached /checkout. Review the matches to separate the same bug from different outcomes.

Turn your agent into a 24/7 analyst
01 / 07The cost of everything
950,931tokens · full recorded range
Explore
Camera controls
1.0× offset

Drag the map to rotate. The next chapter resumes the guided view.

From one diagnosis to a daily practice

Wake up to a story
you can act on.

Give your agent a journey worth reviewing. It finds the sessions, investigates the friction, and brings back the evidence.

01Find the journeys
02Review the evidence
03Report the finding
04Repeat tomorrow
Start free Put it to work

Token totals come from a separate captured session. The checkout journey, Sightmap names, and pull request are illustrative; checkout screenshots come from a working example. Daily runs require a schedule in your agent environment. The terrain illustrates information density. Inspired by Attention Is a Zoom and A Map, Not a Movie.