Meridian

Stablecoin depeg & liquidity-stress early warning — findings dashboard

Evaluation →

A stablecoin is a cryptocurrency designed to always be worth $1. Sometimes it isn't — a depeg. By the time that shows up on a price chart, the damage is often already done. This page tracks how on-chain trading behavior — not price — looked in the run-up to two real depegs, and uses the same method to score current trading in real time below. Measured (not assumed) against USDC's 2023 depeg, this signal rose a full 3 days before price broke. Tested the same way against UST's 2022 collapse, it found no such lead time — UST was already unstable before this project's data even begins, a real limit stated plainly, not glossed over. See the methodology in docs/FINDINGS.md.

Every risk number on this page is on the same 0–100 scale: 0 is a normal, quiet market; higher means trading looks more like the historical crises this page is validated against. Look for the small bar under each score — it shows exactly where that reading falls on that scale.

Live now

live

Every bar below is a live reading of the same 0–100 risk score used throughout this page, computed right now from real trades happening on-chain — not historical data. The marker on each bar shows where a real historical crisis has averaged, for scale. Streamed from a WebSocket feed of on-chain trades, scored against the same frozen model as the case studies below. UST doesn't appear here — it has no live pool activity to track (see the note at the bottom of this page).

USDC_DAI · curve
3.9590
0 · calm↑ typical crisis100
 
gap-filled

The risk score for USDC_DAI on curve is 3.9590, with a price deviation of -0.2015. It is assigned to cluster 0, which has a severity of 0.0500, and the wallet concentration severity is 0.1283.

USDC_DAI · uniswap_v2
4.9592
0 · calm↑ typical crisis100
 
gap-filled

The risk score for USDC_DAI on uniswap_v2 is 4.9592. The price deviation feature is -0.5917. The cluster assignment is 0, with cluster severity 0.0500 and wallet concentration severity 0.1283.

USDC_USDT · curve
4.9723
0 · calm↑ typical crisis100
 
gap-filled

The risk score for USDC_USDT on curve is 4.9723. The price deviation feature is 0.4491, cluster assignment is 0, cluster severity 0.0500, and wallet concentration severity 0.1283.

USDC_USDT · uniswap_v2
4.3679
0 · calm↑ typical crisis100
 
gap-filled

The risk score for USDC_USDT on uniswap_v2 is 4.3679. The price deviation feature is 0.1120. The cluster assignment is 0, with a cluster severity of 0.0500 and a wallet concentration severity of 0.1283.

Live price, as it's trading right now

Each finalized hourly window's implied price, updated as the streaming consumer emits new windows. Unlike the crisis charts below, the price axis here is zoomed to the actual range of live readings, not a fixed $0–$1.05 — a real crisis would still be unmistakable, but small day-to-day movement stays visible too.

USDC_DAI

curveuniswap_v2
$0.7749$1.3697$1.9645
694 readings · hover or focus the chart to read a value

USDC_USDT

curveuniswap_v2
$0.9534$0.9914$1.0293
1315 readings · hover or focus the chart to read a value

Recent activity

Unscored raw trades as observed by the live feed. Counts are descriptive, not a risk signal. For risk context, see the hourly Live now view above.

Swaps in 1h
0
Last observed
Active pools
0
Live / gap-filled
0 / 0
01104:2504:3504:4504:5505:0505:15
Click or focus a bar to inspect that interval

Bars bucket by when the feed observed each swap (UTC), not the on-chain timestamp. Only the latest 100 events / 48 hours are retained.

Latest swaps

Risk score, last observed reading per pair

A 0–100 score built from four signals: how far price has moved off $1, how unusual trading volume is, which stress pattern a K-Means model assigns the hour to, and how concentrated trading is among a few wallets. Each pair's own dataset ends on a different date — the reading shown is that pair's last available hour, not "today."

USDC_DAI · curve
4.0
0 · calm↑ typical crisis100
reliable baseline
as of Nov 1, 2026
USDC_DAI · uniswap_v2
5.0
0 · calm↑ typical crisis100
reliable baseline
as of Nov 1, 2026
USDC_USDT · curve
5.0
0 · calm↑ typical crisis100
reliable baseline
as of Nov 1, 2026
USDC_USDT · uniswap_v2
4.4
0 · calm↑ typical crisis100
reliable baseline
as of Nov 1, 2026
UST_DAI · curve
41.9
0 · calm↑ typical crisis100
elevated · no reliable baseline
as of Jun 12, 2022
UST_USDC · curve
41.7
0 · calm↑ typical crisis100
elevated · no reliable baseline
as of Jun 13, 2022
UST_USDT · curve
56.6
0 · calm↑ typical crisis100
elevated · no reliable baseline
as of Jun 12, 2022

USDC recovered. UST didn't.

Two real depegs, plotted on the same $0–$1.05 price axis so the depth of each collapse is directly comparable. USDC (March 2023) lost its peg for days after its issuer disclosed exposure to Silicon Valley Bank, then fully recovered. UST (May 2022), an algorithmic stablecoin with no such backing, collapsed and never came back.

USDC · Curve 3pool, March 2023 (SVB exposure)

2023-03-08 → 2023-03-15

$0.00$0.25$0.50$0.75$1.00low $0.8808$0.9961
Hover or focus the chart to read a value

UST · Curve metapool, May 2022 (algorithmic collapse)

2022-05-07 → 2022-05-16

$0.00$0.25$0.50$0.75$1.00low $0.0334$0.0450
Hover or focus the chart to read a value

A model trained on the mild crisis beat the baseline detecting the severe one. Trained the other way, it lost.

A machine-learning model was trained on one coin's crisis pattern, then tested on the other's — a coin it never saw during training. Both directions are reported below, including the one where it lost, not just the flattering result. Longer bars mean a better balance of catching real crisis hours without too many false alarms (the F1 score); the numbers underneath spell that balance out directly.

Trained on USDC → tested on UST

beats baseline

Never saw UST's algorithmic collapse during training. Still generalizes.

Logistic regressionF1 0.915
precision 0.99 · recall 0.85 · PR-AUC 0.92
Random forestF1 0.858
precision 0.89 · recall 0.83 · PR-AUC 0.90
Rule-based baseline (existing)F1 0.858
precision 0.99 · recall 0.76 · PR-AUC 0.84

Trained on UST → tested on USDC

loses to baseline

Trained on total collapse, tested on a milder bank-run depeg. Doesn't transfer.

Logistic regressionF1 0.523
precision 0.79 · recall 0.39 · PR-AUC 0.59
Random forestF1 0.611
precision 0.72 · recall 0.53 · PR-AUC 0.61
Rule-based baseline (existing)F1 0.995
precision 0.99 · recall 1.00 · PR-AUC 0.99

Who trades during a crisis? Mostly infrastructure, not people.

Ranking every wallet by how much trading volume flows through it during each crisis (via PageRank, the same class of algorithm behind Google's original search ranking) — 19 of 20 shown here turn out to be contracts (routers, aggregators), not people. Verified against each address's actual on-chain bytecode, not guessed from naming patterns. A high rank means "a lot of volume routes through this address," not "this is an influential trader."

USDC · March 2023

10 of top 10 are contracts, not EOA traders

rankaddresstypeinfluence
10x00000000…935e7dcontract
0.0672
20x7122db0e…fbd0e0contract
0.0359
30xef1c6e67…54bf6bcontract
0.0259
40xc79c30ef…ddd2aacontract
0.0217
50x99a58482…b5788fcontract
0.0170

UST · May 2022

9 of top 10 are contracts, not EOA traders

rankaddresstypeinfluence
10xeef86c2e…880c8fcontract
0.0834
20x220bda5c…2196d4contract
0.0601
30x83a30b9c…41d88econtract
0.0298
40xdf388f11…2f3284contract
0.0263
50x1c073d50…53a4c3contract
0.0204

A note on what's live and what isn't: the crisis case studies, classifier results, and wallet rankings above are computed from a fixed, frozen dataset — USDC data ends 2023-05-31, UST data ends 2022-06-15, and neither updates. Only the "Live now" section is a genuine live feed, streaming directly from an Alchemy WebSocket connection to Ethereum.