Two terminals live on one site. The first reads Robinhood Chain round by round:
tokenized-equity feeds compared out of the 36 that carry a price feed, over 77 sessions, rebuilt from 78,137 aggregator rounds — every daily close reproducible on chain from its own round id. The second is the U.S. housing market:
927 markets × 79 columns of seller motivation, price cuts and supply, captured as a dated snapshot with the tables downloadable as CSV.
What the film walks through:
01 · how a daily close is read off chain 4663 — the round ids, the boundary search, the pacing that keeps the RPC from ever throttling it
02 · the stress board — five measured inputs against published thresholds, weighted 30 / 20 / 15 / 20 / 15, with the four feeds that don't make the window named instead of quietly averaged in
03 · the housing terminal — the national read, one market in full (Austin: MSI 7.42, 54.6% of listings cut), the dataset catalogue, and the downloads
Every figure on screen is computed from the chain or from the captured snapshot. No vendor feed, no stock footage, no template — the cards, the captions and the soundtrack were all made from
scratch for this cut.
https://t.co/XmyEJX4jnj
0x516a7c07b3f2e1e6614006844844f94974f2066b
MERIDIAN is on Robinhood Chain.
0x516a7c07b3f2e1e6614006844844f94974f2066b — chain 4663.
On chain: name MERIDIAN · symbol MERIDIAN · 18 decimals · 1,000,000,000 supply.
What the address points at is the part worth reading. It is the address of a project whose product rebuilds itself from the chain every morning: 32 tokenized-equity feeds, 77 trading sessions, 78,137 aggregator rounds read one by one, and every daily close reproducible on chain from its own round id. No vendor feed. No estimate. No carried price.
The address now sits on the front page — one click to copy, one click to Blockscout, right next to the board it belongs to.
https://t.co/XmyEJX3LxL
The Meridian score is not a model. It is a stated convention: five measured quantities, each mapped through a published threshold, then weighted. Nothing here is fitted to anything.
· Drawdown depth — how far below the window high, saturating at −30% → weight 30
· Realised volatility — annualised σ of daily closes, saturating at 80% → 20
· Time underwater — share of sessions spent below the running high → 15 · 30-session trend — window return, mapped so +20% → 0 and −20% → 1 → 20 · Against the pack — window return minus the
median feed's, saturating at ±20 percentage points → 15 Sum → 0–10, banded settled / calm / watch / stressed / capitulation. Every threshold and every weight is printed on the method page — the
point is that you can argue with the number, not just receive it.
What it deliberately does not know: who holds what. There is no wallet or transfer data in this build — the score is built only from what a price feed can prove. Holder concentration is out of scope, and the method page says exactly that.
Meridian: every tokenized equity on Robinhood Chain, reduced to one number for how much stress its holders are under.
34 on-chain price feeds, 31 compared over 55 sessions — 23,439 rounds read. No vendor feed, no filler.
Every number on this site is rebuilt from the chain itself — no vendor feed, no estimate. Here is the mechanism, in full.
0x516a7c07b3f2e1e6614006844844f94974f2066b
A Robinhood Chain equity feed is a Chainlink-shaped aggregator. Two calls matter:
latestRoundData() for the head,getRoundData(uint80) for anything behind it. The round id is packed as phase << 64 | counter —
monotonic in time, and that monotonicity is the whole trick.
A daily close is "the last round printed before 23:59:59 UTC". This chain makes 0.1-second blocks, so eth_getLogs only reaches back ~11.6 days, and historical eth_call is rejected (not an archive node). The only way backwards is by round id. So the crawler binary-searches each day's boundary: a coarse scan of ~40 samples
builds a roundId ↔ time table, then every day's refinement advances in the same batch.
Reads are metered, and this RPC punishes bursts far harder than rate. So: 50 calls per request, one request in flight, ~3.2 seconds apart, no concurrency. The last full walk read 78,137 rounds across 36 feeds in 349 requests, zero throttled.
And nothing is interpolated. A day a feed didn't print is a null — not a carried price, not a guess.
A comparison is only as honest as its alignment.
Three rules, applied in this order — and each one was learned the hard way.
1. What counts as a trading session. A UTC day is a session if a supermajority of the feeds that already existed by then printed on it. "Any feed printed" is wrong in a way you cannot see:
exactly one feed on this chain prints on Saturdays, and counting those folded the entire window down to three days before anyone noticed.
2. Where the window opens. Not "the earliest qualifying day" — that trades seven feeds for one session. The start is chosen to maximise feeds × sessions: today 32 × 77 beats 34 × 70 and 25 × 78, so the window opens 2026-06-23.
3. Who does not make it — by name. Four feeds are ▎ held out right now: SGOV and USAR (both start 2026-07-02), DELL (07-23), GLD (09-21). A feed with a hole in it is held out too: a missing day would silently misalign every "N days ago" figure, because those count array positions, not calendar cells.
Of the 194 tokenized equities that exist on Robinhood Chain, 36 carry a price feed and 32 are compared. The other four are listed at the
bottom of the board with their reason — not quietly averaged in.
216 routes, 819 assets, 508 captured API responses — 234 of them keyed per market. Every number on every page is replayed from captured state, so no page ever waits on an upstream API:
no signup wall, no rate limit, no "service unavailable", and no silent re-fetch that changes the figure under you. The research archive (87
pieces), the prediction-market resolution tracker (55 pages), the method page and the dataset catalog are all in there.
And yes — it is a snapshot, captured 2026-10-01. That is the feature, not the caveat: a frozen, dated, reproducible read of the housing market.
You always know exactly which day you are looking at.
https://t.co/JxDAE8P13K
eading the board: the shape column is the whole 72-session window in 66 pixels, and the dashed rule is its high — so the gap beneath it is the drawdown.
Click any row for the five components behind the score. No wallet, no signup:
https://t.co/z8SVhabkgE
34 tokenized equities on Robinhood Chain carry a price feed. 32 make the comparison.
SGOV and USAR are held out by name — their aggregators don't reach back to the window's first day. They're listed at the bottom of the board, not quietly averaged in.
32 tokenized equities, 496 pairs, one matrix — built from overlapping on-chain closes, not a vendor's daily bars.
Rows are reordered by similarity, and the scale is centred on the median pair (0.255), not on zero. IONQ/RGTI reads 0.93; SPY/USO −0.54.
Three names sit at capitulation on Robinhood Chain right now: RGTI 8.31, RKLB 8.14, CLSK 8.02.
All three have spent 97–99% of the 72-session window underwater. The median token in the universe is up 4.05%. That spread is what the
board is for.
The Meridian score is five measured quantities:
drawdown depth, realised volatility, time underwater, 30-day trend, and performance against the pack.
Each is scaled against a stated threshold, then weighted 30 / 20 / 15 / 20 / 15. No fitted parameter, no training set — and the whole thing is printed on the site, so it can be argued with.
Meridian: every tokenized equity on Robinhood Chain, reduced to one number for how much stress its holders are under.
34 on-chain price feeds, 31 compared over 55 sessions — 23,439 rounds read. No vendor feed, no filler.