Skip to content
Back to Projects
Active
Web Apps
Jul 2026

Verso

A diary for artworks — Letterboxd where the unit is an individual work, not a visit. Offline-first capture over a real catalogue of 10,000 works currently on view at The Met.

View Source
10k works
Catalogue
The work
Unit of record
Offline-first
Capture
Measured, not assumed
Thesis
Verso project visual

The constraint

Every previous attempt at a Letterboxd for art has stalled at one city, and the usual diagnosis is execution. The more likely one is frequency: a film enthusiast logs 50–150 titles a year, while a dedicated gallery-goer sees maybe 15 exhibitions. Fifteen events a year is too few to form a habit, and a feed with fifteen posts a year is not a feed.

The approach

Log works, not visits. One gallery visit contains fifteen things worth logging, which turns ~15 events a year into plausibly 150+. Every other product decision exists to protect that: the capture screen never navigates away, rating is always deferrable to an evening queue, and nothing waits for the network because gallery basements have no signal.

Process
  1. 1
    Write the PRD first

    Started from docs/PRD.md so the frequency bet was stated explicitly and could be falsified rather than quietly assumed.

  2. 2
    Ingest a real catalogue

    Built ingestion for The Met (and Art Institute of Chicago) with a reconciliation pass, then committed the seed so nothing about the demo depends on an upstream API being up.

  3. 3
    Protect the capture loop

    Made logging re-shortlist for the next work instead of bouncing to a confirmation page — that bounce is exactly how work-logging reverts to visit-logging.

  4. 4
    Instrument the bet

    Added metric gates and an evaluation harness for catalogue reconciliation so the question 'is the frequency thesis holding' has a number attached to it.

Outcomes

  • Committed seed catalogue of 10,000 works currently on view at The Met — the demo touches no network
  • Offline-first capture writing to the device before sync
  • Evening queue that returns unrated sightings one thumb at a time
  • Year in Art recap built to be screenshotted
What I'd do next
  • Expand ingestion beyond two museums; the reconciliation arm is the hard part, not the fetching.
  • Take the capture screen into an actual gallery basement and measure the offline path under real conditions.
  • Test whether the evening queue survives contact with people who are not me.
Details

Designed and built an artwork-logging app around a single product bet — that logging individual works instead of visits raises logging frequency by an order of magnitude — and instrumented the app to measure whether that bet holds.

Shipped offline-first capture, a deferred-rating evening queue, a social layer, and a committed 10,000-work seed catalogue so the whole app runs with no network access.

Related projects