the odds move before the headlines. i write down why polymarket · kalshi · odds swings · whale bets not advice, just what the money says
Grainger Market, Newcastle
Joined October 2009
- Tweets10.9K
- Following3.2K
- Followers8.7K
- Likes8.6K
POLYMARKET WALLET THUNDEROUS-AUTHORITARIAN-SNARL MADE $340K IN 24 HOURS ON ESPORTS. SIX DAYS AGO IT DIDN'T EXIST.
His first 4 hours of trading: -$1.04M.
The next 5 days: +$1.38M back.
Profile: thunderous-authoritarian-snarl
Every position on his profile is esports: CS2 at IEM Beijing, LoL at Worlds.
His stats:
-> PnL: +$345.8K
-> Volume: $11.1M
-> Markets: 36
-> Trades: 3,055
-> Joined: Oct 30
The wins:
1. Na'Vi vs Falcons, 876K shares at 45.4¢ -> +$477,951
2. B8 vs Legacy, 645K at 39.9¢ -> +$387,375
3. Invictus, Game 4, 745K at 56.5¢ -> +$324,388
The losses:
1. Hanwha Life, Game 1, 1.22M shares at 65.1¢ -> -$795,638
2. Heroic vs Falcons at 28.6¢ -> -$279,319
3. LOS, Game 2 -> -$227,686
One Game 1 bet cost more than his total profit, twice over.
What the profile doesn't show: $68.6K paid in fees. Rebates gave back $35.3K.
And the tracker says $361K for the day. The profile's 1D: $340,053.
Copy a wallet that bets $796K on one game, or fade it?
Check the wallet yourself:
polymarket.com/@thunderous-a…
Want to watch the same esports markets live? Start here ↓
polymarket.com/?r=Vaill
Yeon Vail retweeted
SHE CHECKS 6 RUG FLAGS IN 1.7S AND APES IN 214MS.
most pump.fun setups ape first and read the dev wallet after. SAKURA reads first
the idea: the edge is not the buy. it's the 3 mints you skipped
i built an anime trader in cat-ear headphones who runs every fresh mint through the same checklist, snipes, takes profit and walks before the dev does
▹ what runs under the hood
1. SNIPE RADAR: sweeps every new mint · 3 skipped per target · dev 31% / bundled 14 / top10 64%
2. RUG CHECKS: 6 flags in 1.7s · mint auth / freeze auth / dev hold / bundles / top 10 / snipers
3. SNIPE SPEED: detect 38ms · quote 164ms · sign 12ms · 214ms to entry · lands +1 slot
4. MOCHI: in at $6.1k mc · 0.50 SOL · TP +82% · +0.41 SOL · hold 3.2s
5. UWU: in at $4.8k · TP +54% · +0.27 SOL · the dev dumps 12.4 SOL right after · -82% from top · she was out
6. SESSION: 2 mints · 2 / 2 wins · +0.68 SOL · 2 lucky cats carry mints in and SOL out
RADAR → 6 CHECKS → APE → TRAIL → TP → NEXT MINT
MOCHI +0.41 SOL · UWU +0.27 SOL · dev sold 12.4 SOL → dodged
alone each check is a checkbox. together 6 checks and a 214ms entry are the difference between a bag and a rug
ZACHXBT NAMES FRENCH BANK CRÉDIT MUTUEL ARKÉA AS THE ALLEGED VICTIM OF A $133M CRYPTO THEFT. THE BANK HASN'T CONFIRMED IT.
Nov 1: $106M in USDC and USDT leaves the exchange accounts of an unnamed EU entity.
Then $27M more. Some of it after he went public.
His claim: a bank with $235B in assets, hit over several days, and no proper disclosure. That silence, he says, cost the chance to freeze about $30M.
The money went USDC to ETH, through instant exchanges, partly into Monero, then across chains.
Not Lazarus, he says, "with a high degree of confidence." And in May, Arkéa applied for a MiCA license to sell crypto to retail clients.
A bank lost $133M in crypto and stayed quiet, or ZachXBT named the wrong victim?
POLYMARKET OPENED 24/7 ANTHROPIC PERPETUAL FUTURES WITH 10X LEVERAGE. $5.86M TRADED IN 24 HOURS BEFORE ANT EVEN LISTS.
Anthropic filed its S-1 on Oct 26. IPO pricing is expected mid-November.
-> Now Polymarket Perps lets you trade it at 3 AM on a Sunday.
ANTHROPIC-USD went live on Oct 29 at 16:00 UTC.
Price now: $1,829.6, +3.9% in 24 hours. Day range: $1,700.1 to $1,893.4.
24h volume: $5.86M. Open interest: $454.5K.
Funding: 0.0054% per cycle. Mark is 0.19% above the oracle.
Up to 10x, isolated margin by default, settles in pUSD, no expiry. The reference price comes from Hyperliquid's market.
At 10x long, liquidation sits near $1,733. That's a 5% dip.
What it shows: a private company now has a live price before the exchange gives it one.
Caveat: it's synthetic. You don't own shares, and the price follows an external feed, not the IPO book.
The IPO is no longer the start of price discovery. It's a checkpoint ↓
ANT perp: long into the IPO or short the hype?
ANTHROPIC Q3 2026 BEFORE THE ANT IPO: $17.0B REVENUE, +1,127% YOY, AND $895.4B OF COMPUTE COMMITMENTS IN THE AMENDED S-1.
Revenue +47% in one quarter. Operating income gone: -$567M after +$974M in Q2.
The biggest number isn't on the income statement.
$895.4B in purchase and lease commitments, almost all compute. $69.8B is due within a year.
That's 27x the $33.2B Anthropic made in nine months.
The $104B nine-month net loss looks worse than it is: $94.3B of it is noncash, the convertible notes getting more valuable.
Adjusted operating income: $2.5B, a 15% margin, down from 19%. Anthropic blames $5B of pricier short-term compute and a $2B noncash donation-matching charge.
Growth slowed from 144% to 47% QoQ. Pricing is expected mid-November.
ANT at the IPO: are you buying the 1,127% or the $895B?
10 MONTHS OF X TRENDS. 9 THEMES. ONLY ONE NEVER DROPPED OUT, AND IT’S NOT POLITICS
i mapped every month Jan–Oct by its top themes on US X trends. the chart looks simple. the reasons behind it aren't
▹ month by month
JAN: Maduro captured (Jan 3) · ICE shootings (Jan 7, Jan 24) · winter storm, 85+ dead · Indiana wins the CFP title · Epstein files out Jan 30
FEB: #SuperBowl #1 · 5.25M+ social mentions, halftime 1.87M · Grammys + Bad Bunny · "Epstein" #1 · Nancy Guthrie #2 · US/Israel strike Iran Feb 28
MAR: Iran war · BTS ARIRANG, 18.4M live on Netflix, ~110M Spotify streams day one · Oscars · No Kings protests
APR: Artemis II #1 · WHCD shooting · Justin Fairfax #1 · Gilgo Beach plea #10
MAY: #MetGala #1 · Kentucky Derby · Late Show finale · Knicks playoff run
JUN: Knicks #1, first title since 1973 · SpaceX IPO at $2T+ · World Cup opens Jun 11
JUL: World Cup, 148.2M posts, peak 8.9M in a day · US 250th · Spain beats Argentina in the final
AUG: Luigi Mangione #1 · Lindsay Clancy trial · Keffe D convicted for Tupac, #3
SEP: Charlie Kirk #1 · Clancy mistrial #1 · GPT-6 launch, only #43
OCT: #DWTS + WWE most tweeted · World Series, #WinOurWay · GTA 6 #1 · Binance hack #1 · Keffe D sentenced, only #24
▹ why the chart looks like this
1. SPORTS 10/10: it runs on a calendar. Super Bowl, Olympics, March Madness, NBA Finals, World Cup, World Series. every month has a final, and every game night brings back the same team hashtags
2. POLITICS 8/10: shocks, not schedule. Venezuela, Iran, Epstein files, the WHCD shooting. it only missed Aug and Oct, when sport and crime took the slots
3. TRUE CRIME 4/10: spikes, not seasons. Guthrie, Fairfax, Mangione, Clancy each hit #1 or #2 for a day or two. no case held a whole month, and the sleuthing crowd lives on TikTok
4. TV, MUSIC, TECH 3/10 each: they need a moment. Oscars, Met Gala, a BTS comeback, Artemis. AI barely trends in the US: GPT-6 peaked at #43
5. CRYPTO 2/10: X only cares when something breaks. the Binance hack went #1, regular price action doesn't
6. YEAR-ROUND VOLUME: the most tweeted US hashtags of the year are #WWERAW, #SmackDown, #AEWDynamite. weekly shows beat one-off news on raw count
the read: if you want reach on X, build around the calendar (finals, award shows, launches) and treat crime and politics as spikes you catch, not a lane you live in
bookmark this before you plan November content
Yeon Vail retweeted
10 MONTHS OF X TRENDS. 9 THEMES. ONLY ONE NEVER DROPPED OUT, AND IT’S NOT POLITICS
i mapped every month Jan–Oct by its top themes on US X trends. the chart looks simple. the reasons behind it aren't
▹ month by month
JAN: Maduro captured (Jan 3) · ICE shootings (Jan 7, Jan 24) · winter storm, 85+ dead · Indiana wins the CFP title · Epstein files out Jan 30
FEB: #SuperBowl #1 · 5.25M+ social mentions, halftime 1.87M · Grammys + Bad Bunny · "Epstein" #1 · Nancy Guthrie #2 · US/Israel strike Iran Feb 28
MAR: Iran war · BTS ARIRANG, 18.4M live on Netflix, ~110M Spotify streams day one · Oscars · No Kings protests
APR: Artemis II #1 · WHCD shooting · Justin Fairfax #1 · Gilgo Beach plea #10
MAY: #MetGala #1 · Kentucky Derby · Late Show finale · Knicks playoff run
JUN: Knicks #1, first title since 1973 · SpaceX IPO at $2T+ · World Cup opens Jun 11
JUL: World Cup, 148.2M posts, peak 8.9M in a day · US 250th · Spain beats Argentina in the final
AUG: Luigi Mangione #1 · Lindsay Clancy trial · Keffe D convicted for Tupac, #3
SEP: Charlie Kirk #1 · Clancy mistrial #1 · GPT-6 launch, only #43
OCT: #DWTS + WWE most tweeted · World Series, #WinOurWay · GTA 6 #1 · Binance hack #1 · Keffe D sentenced, only #24
▹ why the chart looks like this
1. SPORTS 10/10: it runs on a calendar. Super Bowl, Olympics, March Madness, NBA Finals, World Cup, World Series. every month has a final, and every game night brings back the same team hashtags
2. POLITICS 8/10: shocks, not schedule. Venezuela, Iran, Epstein files, the WHCD shooting. it only missed Aug and Oct, when sport and crime took the slots
3. TRUE CRIME 4/10: spikes, not seasons. Guthrie, Fairfax, Mangione, Clancy each hit #1 or #2 for a day or two. no case held a whole month, and the sleuthing crowd lives on TikTok
4. TV, MUSIC, TECH 3/10 each: they need a moment. Oscars, Met Gala, a BTS comeback, Artemis. AI barely trends in the US: GPT-6 peaked at #43
5. CRYPTO 2/10: X only cares when something breaks. the Binance hack went #1, regular price action doesn't
6. YEAR-ROUND VOLUME: the most tweeted US hashtags of the year are #WWERAW, #SmackDown, #AEWDynamite. weekly shows beat one-off news on raw count
the read: if you want reach on X, build around the calendar (finals, award shows, launches) and treat crime and politics as spikes you catch, not a lane you live in
bookmark this before you plan November content
52 KALSHI NFL SCANS, $1,000 PAPER BANKROLL, 0 ENTRIES. VIKINGS AT LIONS IS NO. 52 AND THE DESK STILL WON'T BUY DETROIT AT 63¢.
Alfred's Market Desk picked the Lions over the 6-0 Vikings this morning.
-> Then it refused to buy them on Kalshi.
DraftKings, FanDuel and BetUS put Detroit at 60.38%, 61.26% and 62.70% after removing the vig.
Median: 61.26%.
Kalshi asks 63¢. Seven contracts with the fee cost $4.5243, or 64.63¢ each.
The desk's rule: buy only when fair value beats the all-in price by 5¢. That takes 69.63¢.
The gap: 8.37 points that nobody is offering.
8 gates: 4 fail, 3 unverified, 1 passes. Every quote was older than 120 seconds at the freeze.
Result: WATCH, quantity 0. Ledger: $1,000, 52 decisions, 30 WATCH, 22 PASS.
What the scan shows: Kalshi at 63¢ already prices Detroit above what the books say it's worth. Liking the Lions doesn't make 63¢ cheap.
Caveat: paper only, unvalidated, and the review needs 100 settled entries and eight weeks. Neither is close.
A pick is no longer a trade. The trade is the price ↓
Kickoff 18:00 UTC. Lions win and the desk left money on the table, or it was right to sit out?
Yeon Vail retweeted
THIS FREE CLAUDE OPUS 5.5 SKILL TURNS ONE ANIMAL PHOTO INTO A WORKING 3D ROBOT YOU CAN TAKE APART
a dev just gave it away for nothing.
i doubt it stays free for long, so here everything i pulled out of it before someone notices ↓
▹ HOW THE CLAUDE SKILL BUILDS THE ROBOT
↳ you drop in one reference image and name the animal
↳ claude builds the body in code, renders one frame next to your image and closes the gaps over 3 rounds
↳ only then does it touch the film
▹ WHAT'S UNDER THE SHELL
↳ a few hundred parts generated in loops
↳ pistons, hoses, boards, screws
▹ ONE CLOCK FOR EVERYTHING
↳ the whole tour is a single function of time
↳ every beat lives in one timeline file
↳ camera, panels and sound all read the same clock
▹ HOW IT CHECKS ITS OWN WORK
↳ it can't watch video
↳ so it reads contact sheets of every beat
↳ fixes what's wrong, re-renders, repeats until a full pass comes back clean
▹ VIDEO + SOUND
↳ 60 fps capture on your GPU
↳ soundtrack synthesized in python from the same beats
i'll keep pushing it until it breaks and post every result here.
POLYMARKET WHALE ALARMING-CIRCADIAN PUT $349K ON "YES" AT 77¢: DEMOCRATS WIN THE MICHIGAN SENATE RACE IN 2026.
Happy Halloween. Three days before the vote, one wallet bought 455,715 Yes at 76.6¢.
Two hours later he went back for more.
+$273K on "No" for Republicans in the same race. Same bet, second door.
$622.6K on Michigan in one morning. If El-Sayed wins, it pays $803K.
The backdrop: Mitchell poll, Oct 26-27. El-Sayed 51%, Rogers 44%. Early October it was 45.6% to 41.4%. 28% had already voted.
Democrats Yes: +9 pts this week, 79.5¢ now.
Not a partisan: the same morning he bought Republicans in Nebraska for $146K. Best past hit: Mamdani at 89¢, +$126.8K.
Good Saturday so far: +$20.7K on paper.
El-Sayed holds the lead, or Rogers pulls a Halloween surprise on Tuesday?
The whale's wallet:
polymarket.com/profile/0x9c9…
Want to take the other side, or ride with him? Start here ↓
polymarket.com/?r=Vaill
(ref link, not financial advice)
Yeon Vail retweeted
Every influencer is posting the same two things. The Polymarket "hack" (it's a joke, nothing got hacked) and the $3M Binance drain. Nobody is talking about the fact that half the servers out there are exposed right now. Unless you run your own hardware and patch it yourself.
CVE-2026-98432, a bug in the Linux kernel TCP/IPv4 stack. One CPU frees memory while another is still reading it to answer a SYN packet. Use-after-free. Remote, no auth, an open port is enough.
CVSS 9.8 Critical on v3.1, 8.3 High on v4.0. In practice the most likely outcome is a crashed kernel, which means a dead server.
Ubuntu 16.04, 18.04 and 20.04 are vulnerable. On 22.04, 24.04, 26.04 and 26.10 the fix is still in progress. The AWS, Azure, GCP and Oracle cloud kernels are on the list too. So basically every VPS running your bot, node or site.
What to do:
sudo apt update && sudo apt upgrade
uname -r to check your kernel version
and reboot after the update, the new kernel doesn't load otherwise
Older releases can use Ubuntu Pro (ESM, Livepatch). Upstream Linux already has the fix in 7.3-rc7, now it's on Ubuntu to ship the packages.
While everyone argues about someone else's $3M, go check your own server. It takes 2 minutes.
1 REDDIT TRADER PUT $10M ON RDDT RIGHT BEFORE EARNINGS. HE'S ALREADY DOWN $949K TODAY.
RDDT is down more than 50% from its highs and ~18% in seven sessions.
-> Now he posts his whole account and says he can't hide from the results.
It's not one bet. It's four.
72,000 shares at $122.77. Still +$413,830.
620 March 2027 $100 calls. Paid $3.09M, worth $2.29M. -$805,498.
347 short puts across 12 strikes, $120 to $210, out to Jan 2028. If they all get assigned, that's 34,700 more shares.
120 of those puts expire Friday. Four of the five strikes are already in the money with RDDT at $128.52.
And 400 short $140 calls, also Friday. If RDDT pops past $140, the upside on 40,000 shares is capped.
His thesis: 61% revenue growth, 91% gross margins, EPS +178%, ~21x 2026 earnings, $2.8B cash, no debt.
Day G/L: -$949,072. Total: -$733,281.
Earnings after the close tonight. Street expects ~$1.30 EPS.
A drop hits the shares, the calls and the puts at once. A pop gets capped at $140.
Genius or premium loss p0rn? ↓
POLYMARKET: MONKEYMASHINGKEYBOARD +$350K IN 24H. TOP POSITION: GUANGZHOU OPEN, MARIA TIMOFEEVA VS XINRAN SUN. IT WASN'T TENNIS.
Xinran Sun: $605 in at 36¢, +$1,074.
The real money came from one NBA night.
Profile: monkeymashingkeyboard
The pattern: underdog moneylines at 28-36¢ and totals at 50¢, six-figure share sizes.
Last night:
1. Mavericks over Thunder, 144,123 shares at 33.5¢ -> +$95,837
2. Lakers over Clippers, 150,000 at 50¢ -> +$75,007
3. Cavaliers -9.5 -> +$53,402
4. Magic vs Raptors Over 223.5 -> +$49,525
5. Thunder vs Mavericks Under, two lines -> +$63,158
From 4 to 7 AM Kyiv the chart went $634K -> $860K.
His stats:
-> PnL: +$861.1K
-> Volume: $30.3M
-> Markets: 1,147
-> Trades: 31,822
-> Joined: Jul 2026
What the profile doesn't show:
Oct 20: +$306.9K. Oct 24: -$116.7K.
He lost $423.6K in 4 days, then made $977.8K in 5.
Fees paid: $20,274. Maker rebates + rewards: $35,418. The platform pays him to trade.
And the leaderboard says +$350K for the day. The profile chart over the same 24 hours: +$179K.
Copy the underdog sizing, or wait for the next -$423K week?
Check the wallet yourself, profile vs leaderboard:
polymarket.com/@monkeymashin…
Want to watch tonight's NBA markets live? Start here ↓
polymarket.com/?r=Vaill
POLYMARKET WHALE ILLUSTRIOUS-SPECTACLES BET $129K ON "NO" AT 70¢: GADI EIZENKOT WON'T BE ISRAEL NEXT PRIME MINISTER.
Election night, Eizenkot "Yes" hit 65%.
By morning it was 22.5%.
Then a wallet with +$2.45M all-time bought 183,232 No at 70.41¢.
Why the Eizenkot market flipped:
Netanyahu bloc 53 seats. Eizenkot bloc 48. Arab parties 19.
Nobody has 61. Eizenkot says he won't serve under Netanyahu.
If Eizenkot isn't the next PM, the bet pays $183,232.
If he is, $129K goes to zero.
No trades at 74¢ now. The same wallet also bought No on Netanyahu at 54¢ and No on a government formed by Feb 28 at 52¢.
Put together, that's not a bet on a winner. It's a bet on deadlock.
Eizenkot finds 61 seats, or Israel votes again?
THIS IS F*CKING UNREAL, NELSON PELTZ'S TRIAN OFFERS TO BUY WENDY'S FOR $1.4 BILLION AFTER A 70% DROP
HERE IS THE PROMPT I'D USE TO READ THE FILING.
$7.35 a share in cash. a ~20% premium. the stock jumped 15% to $7.10
but one number on wendy's books is bigger than the bid →
the number: wendy's had about $2.4 billion of net debt as of june 28. the offer is for the equity
the case: trian and related parties already own 16.3% of wendy's. peltz has circled the chain for almost two decades, mulled a take-private in 2022 and walked away in 2023. now he's back with partners
who's in the group:
→ trian fund management, leading
→ flynn group, 340+ wendy's restaurants, one of the largest US franchisees
→ bluefive capital
→ sixth street partners
how october 28 played out (ET):
→ 8:00 am: the wall street journal breaks the offer
→ premarket: WEN up as much as 16%
→ 8:29 am: reuters confirms through the regulatory filing
→ 9:00 am: cnbc runs it, six straight quarters of same-store sales declines in the story
trian's pitch: wendy's is "at a critical juncture" and needs "patient, long-duration capital" away from the public market. TD Cowen agrees a take-private makes sense because the turnaround will likely be long
the backdrop: shares down ~70% in five years, ~26% this year. guidance pulled in august. CEO bob wright, in since may: "we are not performing in our potential"
the cleanest way i know to read a deal like this is to give the filing to an AI and make it show its work
give claude (or any model) the trian offer filing and wendy's last two quarterly reports, then ask:
"1. pull the offer terms: price, premium, who's in the group, financing conditions. cite the page for every number. if a number isn't in the documents, say so instead of guessing
2. put $7.35 next to net debt and the last six quarters of same-store sales. show the math and label every estimate as an estimate
3. list who sits on both sides: trian's 16.3% stake, its board seats, flynn as a franchisee and a buyer. flag conflicts, don't judge them
4. write the 5 questions a special committee should ask before saying yes. no buy or sell call, no price target"
start with one filing before expanding to the whole sector
$7.35: fair exit for shareholders, or a discount buyout of a brand at its bottom?
The whale's wallet:
polymarket.com/profile/0xbaa…
Want to take the other side, or ride with him? Start here ↓
polymarket.com/?r=Vaill
(ref link, not financial advice)
POLYMARKET WALLET 0XB0E7 "MADE $1.5M IN A WEEK". ITS MAIN POSITION, A WHITE HOUSE 200+ POSTS BET: +$30
The Polymarket weekly leaderboard puts 0xb0e7…39d2 at #2 with +$1,498,207.
-> Now open the wallet itself. Total PnL since it joined:
+$114.5K. Not for the week. For everything.
The wallet joined on September 16.
It has made 32 trades in total. All 32 are sells. Zero buys.
Total volume: $15.8K.
Its last 500 actions have no trades at all: 257 redeems, 163 merges, 80 conversions, all on Elon Musk tweet-count ranges for Oct 23-30.
Profile PnL on Oct 21: $72.4K. Peak on Oct 24: $159.3K. Today: $114.5K.
So the real week is about +$42K. The board says +$1.5M. The monthly board says +$2.51M on $13.2K volume.
Its biggest real win: "No" on Harry Kane for the 2026 Ballon d'Or, +$97K.
The main position from the alert, White House 200+ posts "Yes", sits at +$30.
What it showed: this wallet moves money by merging, converting and redeeming shares, not by buying them, and the weekly board shows $1.38M more than the wallet has made in its whole life.
Caveat: Polymarket has not explained the gap. Alerts that read the leaderboard will keep printing it.
Would you copy a wallet from the leaderboard number, or from its profile?
A single leaderboard line is no longer proof of a win. One profile check now separates a $1.5M week from a $42K one ↓
the prompt pack is the product. the uncensored api is the fallback when the pack dies
most still chase one jailbreak paste and pray the next model update does not wipe it
Andreas698/glm-5.3-uncensored ships 35 paste-ready prompt files plus a GLM 5.3 variant guide and a RouterHub key that also hits GPT, Claude, DeepSeek, Qwen and Grok
i read the repo like a stack, not a meme
what it actually hands you:
1. 35 jailbreak files in /jailbreaks · sorted by family: OpenCode Zen stealth · Claude Opus/Sonnet · GLM/ZCode 5.2–5.3 · DeepSeek · Kimi · Grok · Gemini/Mistral/Qwen · Ollama Modelfiles
2. use modes are named: system paste · AGENTS.md drop-in · workspace policy · ollama create -f · grok config field · not one vague "prompt.txt"
3. labels get separated: abliterated ≈ weight-side refusal cut · uncensored ≈ broader product class · measure refusals, do not trust the badge
4. when a persona lock gets patched: update the file or skip the prompt layer and hit a GLM 5.3 uncensored API id on RouterHub
5. one prepaid key · OpenAI-compatible Chat Completions for GLM · Anthropic-compatible Messages where listed · 30+ catalogue entries beyond the GLM lane
6. alone a jailbreak is perishable · alone an uncensored model is a black box · together they are a research stack with a documented escape hatch
MIT. stars already moving. the readme is the paper: prompts first, model semantics second, gateway third
most threads sell "uncensored" as a vibe
this repo sells a folder you can audit and an API you can fall back to when the folder ages out
bookmark this before the next model rotation kills half your system prompts overnight
edge is open while people still paste one trick instead of running a collection
This Hyperliquid trader was up $5.98M at 7:00 UTC. By 18:40 UTC he was down $5.73M. Same day, same account.
Wallet: 0xb83d…6e36
Hyperdash labels it "Abraxas Capital (Hedged) #1".
32 open positions. All 29 on the main perp markets are shorts: BTC, ETH, HYPE, SOL, ZEC, DOGE, XRP and more.
His stats:
-> $191.8M account value
-> $572.5M in short positions, about 3x the account
-> +$82.3M PnL this month
-> +$140.5M all-time, on $5.93B volume
That's how the day went:
1. BTC short, 2,516 BTC from $78,114 -> -$6.23M
2. HYPE short, 599,798 HYPE from $69.82 -> -$6.90M
3. SOL short -> -$2.49M
ETH and ZEC shorts made back +$2.3M and +$2.2M. Funding paid him $93.8K.
Net for the day: -$4.84M, about -2.5%.
One thing the balance chart hides: at 07:20 UTC he sent $10M out of the account. So the chart drops about $21M from top to bottom, and $10M of that is the transfer. Trading took him from +$5.98M to -$5.73M, an $11.7M swing.
In other words, $4.84M is a bad Tuesday for him, not a blowup. He is still +$12.1M on the week.
If the "Hedged" label is right, the other side of these shorts lives somewhere else.
Would you short $572M of crypto into a bounce, or is this just the hedge doing its job?
LUXONIS OAK 4 FFC PUTS 5 ROBOT CAMERAS ON ONE VISION COMPUTER FOR $849
SEES, CALIBRATES, THINKS ON-BOARD.
every robot vision board says trust me.
luxonis ships one that also tells you where it breaks. that part is rarer than the specs.
OAK 4 FFC: THE MECHANICS
> up to 5 camera modules on a 70 × 53 mm carrier, placed wherever your robot's body needs them
> modules: 48 MP IMX586 color, 5 MP OG05 and 1 MP OV9282 global shutter, sold separately
> RVC4 compute: 6-core ARM, 8 GB RAM, 128 GB storage, Luxonis OS on Linux
> "52 TOPS" on the box, which is a 48 TOPS INT8 accelerator plus a 4 TOPS FP16 GPU
> 2.5 GbE, USB-C, CAN, 6–32 V, optional PoE+
> 36 g barebone, 215 g dev kit
the humanoid layout they suggest: 2 cameras forward, 1 wide on each side, 1 looking back.
WHERE OAK 4 FFC CALIBRATION FAILS (THEIR WORDS, NOT MINE)
dynamic calibration lines cameras up from the scene itself, no checkerboard. it needs texture at different depths.
> dark scenes: bad
> blank walls: bad
> reflective or glass surfaces: bad
> repeating patterns: bad
and the 52 TOPS is a peak number, not a benchmark for your robot's workload.
that is why this matters. a humanoid head in a white hallway is exactly the case on that list. you know before you buy.
and in parallel:
> the older OAK-FFC 4P took 4 cameras, this one takes 5
> it can run a simple robot alone, or sit next to a Jetson as the eyes only
> $849 dev kit shipping now, $730 barebone in Q1 2027
everyone ships a vision board that shows only the demo. luxonis shipped one with a list of rooms where its calibration struggles.
5 cameras on one $849 board: real humanoid vision stack, or a dev kit for drone people?
This trader turned $55 into $53,350 in 2 hours 25 minutes. After that, his profile barely moved.
Profile: polymarket.com/@hellsangelx
No politics, no sports. Only crypto "Up or Down" markets, bought at 1¢ to 10¢.
His stats:
-> +$85.4K total PnL
-> 1,114 markets traded
-> 29,361 trades
-> 23 months on Polymarket
That's how it works:
His last 500 trades: median entry 5¢, the cheapest 10% at 1¢ or less. Median size: $3.27. Bitcoin, Ethereum, Solana, XRP.
Cheap tickets. Most of them expire at zero.
Then October 5. Market: "Bitcoin Up or Down on October 5?" The deciding Binance candle closed at 12:00 PM ET.
At 12:01 PM he bought 55,000 "Up" shares at 0.1¢. Cost: about $55.
The market resolved Up. At 2:26 PM he sold at 97¢.
Here are his top 3 deals by profit:
1. $55 -> $53,350 (+96,901%)
2. Solana, Jul 27, Down at 1.5¢ -> +$4,670
3. Solana, Feb 9, Down at 4¢ -> +$3,873
Before that day: $30.7K in 22 months. After it: $85.4K.
In other words, one minute made 62% of everything he has made on Polymarket.
Not repetition at scale. One ticket, at the right minute.
Was that an edge, or a slow order book?
THIS IS F*CKING UNREAL, GPT-6.1 TERRA TURNED 10 INDEX POINTS BEHIND CLAUDE OPUS 5.5 INTO A 4X CHEAPER BILL
SAME 1,000 AGENT TASKS: $98 VS $390
orcarouter, oct 27, artificial analysis index: opus 5.5 at 58, terra at 48.
10 index points sounds like another league. then the bill.
the job: a 20-step browser agent, 3,000 input and 150 output tokens per step, run 1,000 times with 30% retries.
WHERE THE 10 INDEX POINTS SHOW UP
→ price: terra $1 in / $5 out per million tokens, opus 5.5 $4 / $20
→ cost per 1,000 tasks: terra ~$98, opus ~$390, $292 apart
→ failures: terra 258, opus 182, so 76 extra exceptions
→ computer use: osworld 2.1 at 74.2% vs 81.8%, a 7.6 points gap
so terra saves $292 and hands you 76 more tasks to fix by hand. at 5 minutes a fix that's about 6 hours. at 15 minutes it's about 19.
the article's answer: split by task, not by vendor. the bounded, supervised queue goes to terra, the expensive tail to opus 5.5.
THE ROUTING PROMPT FOR CLAUDE
give claude code your last 200 agent runs as a csv (task, steps, tokens, pass/fail, minutes to fix), then ask:
"1. tag every task as bounded (clear end state, a human checks it) or open (unsupervised, multi-file, long horizon). don't guess on unclear rows, list them for me.
2. price each tag on gpt-6.1 terra ($1/$5) and claude opus 5.5 ($4/$20) with my real token counts, plus my fix minutes times the failure gap (258 vs 182 per 1,000).
3. propose a routing rule: terra for bounded, opus 5.5 for open, fall back to opus when terra fails twice. write it as a config diff and never apply it without my yes.
4. show the monthly $ and the extra fix hours side by side, then stop."
start with one queue before expanding.
pay 4x for opus 5.5 on every task, or route and fix 76 by hand?
