A full, plain-language tour of the whole system as it stands today — every layer, what's live, what's built-but-not-yet-visible — and a detailed map of what's left to do.
The Brick Drop is a live, installable web app that grades 5,483 LEGO sets — each through four buyer lenses, colored by its own photo, with a brand-voice take. The engine, the catalog, personalization, and the share loop are all built and shipped. The current frontier is trust and activation: a just-shipped value upgrade that isn't visible yet, an accuracy fix waiting to be committed, and the work to prove the grade is right.
Four layers, each feeding the next — from raw reviews all the way to a card you can post to Instagram.
The engine is the heart, and it's the most mature part of the project — built and hardened over weeks before the app existed.
It answers one question — how good is this set, really? — by distilling what actual reviewers said. It's deliberately honest: it detects sarcasm ("oh great, another grey spaceship"), it weights community consensus over lone hot-takes, and when there isn't enough evidence it says so ("Early read") rather than faking confidence.
Most "best LEGO" lists are one person's taste. The Brick Drop's grade is distilled from many reviews, adjusted for hype and sarcasm, and shown with its receipts. That's the whole credibility of the product — which is why the roadmap leans so hard on proving it (see §08, "Trust the grade").
A grading tool is only as good as its coverage. In a few days the catalog went from a demo to nearly complete.
28 sets. Enough to prove the app worked, not enough to be useful. Browse was tiny.
5,483 sets graded and stored — essentially the entire buildable catalog from 2017 to 2026. Browse is a real, grade-sorted shelf.
A harvest of Brick Insights added 5,883 review references, and confirmed a hunch: there's more written coverage than video.
| Source | References | What it adds |
|---|---|---|
| Blog / written | 3,916 | The biggest well — detailed, considered reviews the engine distills via article text |
| YouTube | 1,711 | Transcripts + top comments (community consensus, like-weighted) |
| Forum | 256 | Enthusiast back-and-forth, useful for niche/older sets |
The sweep also logged ~271 sets it couldn't finish — almost all transient rate-limit blips from the review source, not real gaps. They're re-gradeable on a retry pass.
Everything above is invisible plumbing. This is the part a fan actually touches — and it's polished.
The newest engine change — and a perfect example of the gap between "shipped" and "live."
The problem it fixes: the "does it hold value?" cluster was a frozen 5.0 for every set. Its two ingredients were both dead — one needed retirement/exclusive flags we never populated, the other needed resale prices we don't have. So the investor grade was uniformly mediocre and meant nothing.
The fix (PR #20): rebuild those two ingredients from data we actually have —
How much of your money is backed by the raw bricks — the set's part-out value ÷ its price. We have part prices for 98% of sets.
How likely it is to grow — how far into its life it is (age) × how collectible its theme is (a curated, tunable list of themes).
The engine now computes real hold-value, but the 5,483 grades in the live database were written ~30 minutes before this merged — under the old frozen-5.0 engine. So on the site right now, investor grades still don't reflect the upgrade. The feature is shipped in code but not yet activated in the data — the catalog needs a refresh pass to bring it to life. (Because these two ingredients are computed from stored facts, that refresh can likely be a fast offline recompute, not a full re-mine.)
An honest ledger of what a visitor sees today versus what exists in the code but hasn't reached them yet.
| Capability | In code | Live on site |
|---|---|---|
| Grades for ~5,483 sets | Yes | Yes |
| Four buyer lenses + switcher | Yes | Yes |
| Blog-review depth in grades | Yes | Yes |
| Investor hold-value (new) | Yes | Not yet — needs re-grade |
| Smart Price accuracy fix | Written, uncommitted | Not yet |
| Per-lens takes (words match lens) | Spec'd only | No |
Before new features, close the gap between what's built and what's live, and back up loose work.
The grade's credibility is the whole product. This is the most important track.
| In flight | Smart Price fix — commit & ship the pricing fail-soft (see §07). |
| Next | Validate the sentiment engine — hand-label a sample of reviews and measure the AI's read against them, so we can retire the "unvalidated" caveat. This is the single biggest un-started credibility item. |
| Next | Tune theme desirability — the investor "Appreciation Outlook" leans on a curated list of which themes hold value. Seth's eye on that list makes investor grades sharper. |
| Later | Prediction model — an honest, labeled "Early read" grade for sets before they release. |
| Done | Full sweep — 5,483 of ~5,754 buildable sets graded. |
| In flight | Coverage cleanup — retry the ~271 rate-limited stragglers. |
| Next | Living grades — a routine that re-grades sets as fresh reviews land, so grades never drift and new releases sharpen over time. |
| Next | Full-catalog search — find (and grade-on-demand) any set by name, not just the ones already graded. |
| Later | New-release ingestion — auto-pull and grade sets as LEGO announces them; more review sources beyond today's three. |
| Done | Per-lens grades + switcher — live. |
| Next | Per-lens takes (Slice 17, spec'd) — the words match the lens, not just the number: the investor view talks resale and cites the investor grade. Small, safe, high polish-per-effort. |
| Later | Per-set tweak — a light "I care more about play value here" adjustment without a full profile. |
| Later | Profile & a "For You" home — a saved identity (lens, history, watchlist) and curated picks that match it. |
| Done | Share loop — branded link previews + downloadable 4:5 card, both lens-aware. |
| Next | Crop-perfect share cards — bring the Crop Curator framing into the generated 4:5 card so wide sets sit right (today it uses a generic center-crop). |
| Next | True Value, expanded — grow "Is the price fair?" from one line into a real price/parts breakdown. |
| Later | Onboarding polish & discoverability (SEO + a share-driven growth loop). |
Activate what's paid for, secure loose work, prove the grade, then build depth.
The build is done and live — engine, catalog, personalization, share loop. The immediate job isn't to build more, it's to activate what's already built (investor-value, the Smart Price fix) and then prove the grade (validate the sentiment engine). None of it is a rewrite. It's finishing well — and switching on the value that's already sitting in the code.
The Brick Drop · built with receipts, not vibes