DSS

DSS 2026 / Day 2 / Amber Hall

Follow the Collateral: A Mental Model for Auditing Polymarket V2

Paul Clemson

6:59 · Recording

Watch on YouTube ↗Open recording viewerOriginal programme entry
Download plain textDownload timestamped JSON

Source: YouTube automatic caption track · Exported 2026-11-16. Timestamps link to the source video. Cue end times were not supplied.

Transcript

0:00Thank you.

0:01Hello, everyone.

0:03Uh, who in the room is a security

0:05researcher?

0:07That's a good number of hands. How many

0:10of you have looked at Polymarket V2?

0:14No hands. No, one hand.

0:15Okay. So cool. I think as someone at

0:20Quantstamp who one of a few companies

0:23that have been looking at Polymarket a

0:26lot over the past year,

0:28I think it's really cool and I think

0:30it's a really under explored area of

0:34crypto, D DeFi kind of.

0:37And I think more people should take a

0:38look. So what I want to do in the next

0:41seven minutes apparently is give you

0:44guys a bit of a a run through of a

0:47mental model that you can go away with

0:50and hunt that huge Polymarket bug

0:52bounty. So let's go.

0:57Oh, it worked. Okay, that's the first

0:59slide. Okay, so the base of Polymarket

1:03is a simple yes and no market.

1:06[clears throat] So for example, will my

1:08talk actually finish in five minutes?

1:11They've given me seven, which makes me

1:14think there is insider trading at play.

1:17So I think they might have manipulated

1:19this market. But yeah, essentially

1:22every Polymarket

1:24condition should always add up to the

1:27sum of the collateral. So if you put $1

1:30in, you get 100 yes and 100 no.

1:33And this always has to hold. This is the

1:36key thing I want you guys to take away.

1:39Oh.

1:41Did Donald Trump moment, sorry.

1:42[laughter]

1:43What I want you to take away is this has

1:46to always hold.

1:47Whatever they add, every new module,

1:49every bridge, every upgrade, if there's

1:52any way that you can get more than $1

1:55for a $1 worth of both yes and no,

1:58then there's trouble.

2:01Okay, so yeah, there is three very core,

2:03just kind of speed on this, three very

2:05cool actions you can do. Split is $1 in,

2:09100 yes, 100 no.

2:11Merge is the opposite of that. I've got

2:13100 yes and 100 no. I can get $1 back.

2:16And then redeem is obviously after the

2:19Oracle

2:21confirms Pulse Talk did indeed last

2:23longer than 5 minutes.

2:25If you're holding no, you can redeem for

2:27$1. If you're holding yes, then you

2:30lost, so

2:32GG.

2:35And then

2:36to go a little bit further, there's also

2:38neg risk markets. [snorts] Neg risk

2:40markets are essentially

2:44a group of markets where only one of

2:46them can be true.

2:47So for example, what color will my

2:49t-shirt be?

2:52Unless I'm wearing two t-shirts or I do

2:55like a Lady Gaga costume change, there's

2:58going to be one answer to this question.

3:00And today the answer is black.

3:02About 15 minutes ago it was white, but

3:04we won't get into that right now.

3:07But yeah, so this is another layer of

3:09complexity

3:11where

3:12the invariant changes a little bit and

3:14instead of yes plus no equals $1, the

3:18sum of all yes, blue, black, red, I know

3:23colors, well done me,

3:25all must equal $1.

3:27And then as of Polymarket V2, there's

3:30also an other position and this is

3:33essentially if

3:35I didn't turn up to the talk,

3:38kind of evacuate emergency,

3:41the event didn't happen, so cancelled

3:43essentially.

3:47And then

3:49the Oracle, another big source, the

3:51Oracle always has to report

3:54a legitimate result that sums to a

3:56dollar. So if they change Uma or Uma

3:59updates how they handle the results

4:03always has to stay with this core

4:06invariant of like

4:08one.

4:14And then as a V2

4:16you can there are markets that are not

4:19going to resolve anytime in the near

4:21future that exist on the Polymarket V1.

4:25For example, the 2028

4:27American presidential election, this

4:29market already exists. If they want to

4:32take this market over to Polymarket V2

4:35to get all the benefits of the cool new

4:37stuff they've added

4:39they need to map the existing market

4:41over to V2

4:43or have a split of the liquidity which

4:46is not cool. So yeah, and when they

4:50suddenly have presidential market on V1,

4:54presidential market on V2, you can

4:56imagine here

4:58even more ways that

5:01yes plus no

5:03if somehow yes

5:06person A

5:08uh could be true on the legacy V1 and

5:12person B is true on the new Polymarket

5:16V2

5:17suddenly yes plus no or yes plus yes

5:21is more than $1 and you can quite

5:24quickly

5:26drain lots of money and we don't want

5:28that to happen.

5:30So yeah, just as a

5:32kind of sum it up, if you guys want to

5:34take a look, these are a few of the key

5:36questions I think you should keep in

5:38your mind when you're planning things.

5:41When you are minting tokens or whether

5:44you're redeeming tokens, is there any

5:46way this can happen twice? For example,

5:48in the legacy

5:49to uh V2 migration I spoke about, is

5:53there any way I can claim on both? That

5:55would be terrible.

5:57And another one would be

5:59can I have

6:02the market of what color t-shirt Paul is

6:05wearing, black.

6:06Can the condition idea of that somehow

6:09collide with

6:12who wins

6:14the football tomorrow, whatever game?

6:16If that's the case, then

6:18again,

6:20how do you resolve the same question

6:22twice to the same condition

6:24problems?

6:25And then finally,

6:27similar to the migration question,

6:30is there any way that yeah, the Oracle

6:34on V1 and the Oracle on V2 can come to a

6:37different conclusion?

6:39Cuz if that happens, we got problems. So

6:41yeah, if you want to take a look at

6:43Polymarket, I would definitely

6:44recommend. It's very cool, and they're

6:46doing some fun stuff. And I've finished

6:49under 7 minutes,

6:51but not under 5. So

6:54I hope you guys lost a lot of money.

6:56Okay, thank you.

Source and accuracy details

Source export SHA256: 6681743e30e9d103fc7a8448094235c717c12837d6a73b1532be187e6d2aecb7