A Polymarket market can flip from untradable to efficient in hours—spread compresses, depth refills, and whales suddenly matter. This post builds a practical tradability score from spread, depth, and cancellation pressure, with a reproducible way to compute it using public order book endpoints.
Explore structured research pillars and internal link paths.
Visit Research SeriesMost “bad trades” on Polymarket aren’t bad because you misread the event. They’re bad because you traded a market that wasn’t a market yet.
The same contract can behave like:
That flip is a liquidity regime change.
If you can detect it early, you get two edges:
Internal links:
“Tradable” is not a vibe. It’s a set of conditions:
In other words: execution is stable.
We’ll build a 0–100 score from three components:
Polymarket’s CLOB API exposes a public spread endpoint (token-level).
External source (CLOB endpoints list): https://docs.polymarket.com/quickstart/reference/endpoints
Let spread be best ask minus best bid (in probability units).
Define a normalized spread score:
spread_score = clamp(1 - spread / spread_target, 0, 1)
Where:
spread_target can be 0.005 (0.5¢) for liquid markets, or 0.01 (1¢) for general use.Pull the order book and compute how far the book moves if you trade a fixed notional (e.g., $2k).
If you can buy $2k with <0.5¢ impact, depth is good. If the first $200 moves you 2¢, depth is not a market—yet.
You won’t get perfect cancel-rate data from one snapshot. But you can estimate stability with repeated snapshots:
Proxy:
stability = 1 - (mean_abs_change(top_depth) / mean(top_depth))
Then clamp to [0, 1].
You need token IDs, then the book.
Polymarket documents that:
Fetch a market by slug (Gamma):
curl "https://gamma-api.polymarket.com/markets?slug=YOUR_MARKET_SLUG"
Then pull book/spread (CLOB):
curl "https://clob.polymarket.com/book?token_id=YOUR_TOKEN_ID"
curl "https://clob.polymarket.com/spread?token_id=YOUR_TOKEN_ID"
curl "https://clob.polymarket.com/midpoint?token_id=YOUR_TOKEN_ID"
This script:
import json
import time
import urllib.request
from statistics import mean
TOKEN_ID = "YOUR_TOKEN_ID"
NOTIONAL_USD = 2000
SAMPLES = 60 # 10 minutes at 10s interval
INTERVAL_SECONDS = 10
def get_json(url: str):
with urllib.request.urlopen(url) as r:
return json.loads(r.read().decode("utf-8"))
def snapshot():
book = get_json(f"https://clob.polymarket.com/book?token_id={TOKEN_ID}")
spread = get_json(f"https://clob.polymarket.com/spread?token_id={TOKEN_ID}")
mid = get_json(f"https://clob.polymarket.com/midpoint?token_id={TOKEN_ID}")
return book, float(spread["spread"]), float(mid["midpoint"])
def vwap_for_notional(asks, notional_usd):
remaining = notional_usd
paid = 0.0
shares = 0.0
for level in asks:
p = float(level["price"])
s = float(level["size"])
level_notional = p * s
take = min(remaining, level_notional)
paid += take
shares += take / p
remaining -= take
if remaining <= 1e-9:
break
if remaining > 0:
return None
return paid / shares
spreads = []
top_depth = []
impacts = []
for _ in range(SAMPLES):
book, spread, mid = snapshot()
asks = book["asks"]
best_ask = float(asks[0]["price"]) if asks else None
vwap = vwap_for_notional(asks, NOTIONAL_USD) if asks else None
if best_ask is not None and vwap is not None and mid > 0:
impacts.append((vwap - mid)) # in probability units
spreads.append(spread)
depth = 0.0
for level in asks[:5]:
depth += float(level["price"]) * float(level["size"])
top_depth.append(depth)
time.sleep(INTERVAL_SECONDS)
spread_target = 0.01
spread_score = max(0.0, min(1.0, 1 - mean(spreads) / spread_target))
impact_target = 0.005
avg_impact = mean([x for x in impacts if x is not None]) if impacts else 1.0
depth_score = max(0.0, min(1.0, 1 - (avg_impact / impact_target)))
avg_depth = mean(top_depth) if top_depth else 0.0
abs_changes = [abs(top_depth[i] - top_depth[i-1]) for i in range(1, len(top_depth))]
stability_proxy = 0.0
if avg_depth > 0 and abs_changes:
stability_proxy = 1 - (mean(abs_changes) / avg_depth)
stability_score = max(0.0, min(1.0, stability_proxy))
score = 100 * (0.45 * spread_score + 0.35 * depth_score + 0.20 * stability_score)
print({
"spread": mean(spreads),
"avg_impact": avg_impact,
"top5_notional": avg_depth,
"spread_score": spread_score,
"depth_score": depth_score,
"stability_score": stability_score,
"tradability_score": score
})
This is intentionally plain: no SDK, no auth, no magic.
Liquidity regimes often follow a predictable rhythm:
If your alerts ignore this, you get a false impression:
“Whales stopped being smart.”
No—your market stopped being tradable.
Two concrete policies:
For example:
In thin regimes, require:
This is exactly where backtesting matters:
When we say “copyable,” we mean:
Use:
Research Series
Follow related research articles or jump to the full pillar library.