Divyanshu (@trydiv)

@trydivNov 1

markets close but conviction doesn't, so we built Position investing has always started with a ticker but conviction rarely does, a place where you express a view on the world and own the companies shaping it, on markets that never close

574

Divyanshu · Nov 12, 2026 · 3:13 AM UTC

@trydiv4h

nvidia closed +2.9% at $173.73. then the wsj report on the board preparing to remove jensen hit after the bell the $NVDA perp on @tradexyz: ▸ $337m traded 5-6pm ET, its biggest hour in 3 days ▸ low $156.14, ~10% under the close ▸ still -6% now, while the other 8 chip perps sit between -1.2% and +3.7% ▸ funding ~69%/yr, longs paying dip buyers are paying up to hold it overnight

Small-multiple line charts of nine chip perps on trade[XYZ] since the US cash close on Nov 11, same y-scale. Nvidia is down about 6% with a low of -10.2% at 5:15 PM ET; AMD +3.7%, Broadcom +2.4%, and TSMC, Micron, SanDisk, SK Hynix, Samsung and Intel all within about 1%.

ALT Small-multiple line charts of nine chip perps on trade[XYZ] since the US cash close on Nov 11, same y-scale. Nvidia is down about 6% with a low of -10.2% at 5:15 PM ET; AMD +3.7%, Broadcom +2.4%, and TSMC, Micron, SanDisk, SK Hynix, Samsung and Intel all within about 1%.

140

Divyanshu · Nov 11, 2026 · 5:04 PM UTC

@trydiv15h

bitso shipped 200+ perp markets this week. pump.fun shipped limit orders, twap and cross margin neither built a matching engine. both are "powered by hyperliquid"

Pump.fun — Launch and trade memecoins on Solana

Pump.fun is the fairest way to launch and trade memecoins on Solana. Coins are instantly tradable on a transparent bonding curve — no liquidity to seed, no presales, no team allocations.

pump.fun

Divyanshu · Nov 11, 2026 · 1:57 PM UTC

@trydiv18h

the most useful part of anthropic's fable 5.2 prompting guide is one sentence: "This codebase is yours: what it's for and how it's built are your call, and you'll keep building on it." what it does in their own tests ▸ merges repeated code in ~half the tries, under a tenth without it ▸ ~1.5x the output tokens ▸ changes how an existing feature behaves in 1 of every 6–10 tries so it goes in your own repo's CLAUDE.md, never the global one that also loads on client code

Excerpt from Anthropic's Prompting Claude Fable 5.2 guide, section 'Say who owns the code', with highlights: the sentence 'This codebase is yours: what it's for and how it's built are your call, and you'll keep building on it'; merged repeated code in about half of tries with it vs under a tenth without; about 1.5 times the output tokens; changed an existing feature's behaviour in one try in six to ten.

ALT Excerpt from Anthropic's Prompting Claude Fable 5.2 guide, section 'Say who owns the code', with highlights: the sentence 'This codebase is yours: what it's for and how it's built are your call, and you'll keep building on it'; merged repeated code in about half of tries with it vs under a tenth without; about 1.5 times the output tokens; changed an existing feature's behaviour in one try in six to ten.

Divyanshu · Nov 11, 2026 · 3:34 AM UTC

@trydivNov 11

anthropic's ipo range: $70–78 the hyperliquid perp is at $104.21 rn ▸ ~34% above the top of the range ▸ ~$1.88tn fully diluted vs $1.405tn at $78 ▸ $110m traded in 24h, $103m open interest not a mispricing tho. xyz:ANT converts to where ANT actually trades on nasdaq, not to the ipo price so $104 is the market's bet on the day-one pop, not on the range

Line chart of the trade.xyz ANT pre-IPO perp on Hyperliquid, hourly close since Oct 26, 2026. It traded between about $95 and $120, and sits at $104.21 on Nov 11, about 34% above Anthropic's $70 to $78 IPO range, shown as a grey band. Volume spiked to $28m an hour on Nov 10 when the range was filed.

ALT Line chart of the trade.xyz ANT pre-IPO perp on Hyperliquid, hourly close since Oct 26, 2026. It traded between about $95 and $120, and sits at $104.21 on Nov 11, about 34% above Anthropic's $70 to $78 IPO range, shown as a grey band. Volume spiked to $28m an hour on Nov 10 when the range was filed.

Anthropic

@AnthropicAINov 10

Anthropic has launched the roadshow for its proposed initial public offering: anthropic.com/news/roadshow-… Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.

Anthropic launches roadshow for proposed initial public offering

Today Anthropic, PBC launched the roadshow for its proposed initial public offering of its Class A common stock.

anthropic.com

Divyanshu · Nov 10, 2026 · 7:59 PM UTC

nyse didn't put stocks on avalanche matching stays on pillar. avalanche is the ledger for the clearing firm, on nyse's own custom chain built w/- avalanche tech, not the public network. still pending approval the back office went onchain first. that was always the slow part

Divyanshu · Nov 10, 2026 · 5:03 PM UTC

claude fable 5.1 → 5.2, what changed on the price sheet: • input: $10 → $8 /mtok • output: $50 → $40 • cache reads: $0.25 → $0.20 • new fast mode: $16 / $80 (2x base) the line that matters for agents is cache reads fable reads cache at 2.5% of its input price. opus 5.5 is at 5%, most older claude models at 10% so fable 5.2 cache hits now cost the same $0.20 as opus 5.5, a model with half its input price. fortune has gpt-6 astra at $1 long agent loops are mostly re-reading context. on that part of the bill the flagship isn't the expensive one anymore

Divyanshu · Nov 10, 2026 · 2:02 PM UTC

ran pyth's off-hours cut on trade.xyz for last week (nov 3–9) • 14 most traded us-listed names, $10.6b volume • 47% of it traded outside nyse regular hours • pyth's october number across venues: 49.8% • sk hynix adr (SKHY) was the outlier at 62% one venue, one week, still within 3 points of the cross-venue month my guess on SKHY: its home market trades while new york sleeps half the stock-perp tape already happens when the nyse is shut

Ranked bar chart: share of volume traded outside NYSE regular hours on trade.xyz stock perps, Nov 3 to 9 2026, 14 most traded US-listed names, 10.6 billion dollars total. 47% overall. SK Hynix ADR highest at 62%, Circle lowest at 31%. Dashed line marks Pyth's October cross-venue figure of 49.8%.

ALT Ranked bar chart: share of volume traded outside NYSE regular hours on trade.xyz stock perps, Nov 3 to 9 2026, 14 most traded US-listed names, 10.6 billion dollars total. 47% overall. SK Hynix ADR highest at 62%, Circle lowest at 31%. Dashed line marks Pyth's October cross-venue figure of 49.8%.

Pyth Network 🔮

@PythNetworkNov 9

RWA perps in October: → $829.4B in volume, +29% on September → 49.8% of US single-stock perp volume traded while NYSE was closed → $14.6B open interest, a new high → 88.5% priced by Pyth Read the full report ⤵️

October: The Market After The Bell

Article · @PythNetwork

Divyanshu · Nov 10, 2026 · 3:28 AM UTC

korean memory longs are paying up on trade.xyz rn at this hour's funding rate: • sk hynix (SKHX) longs: ~0.30%/day, ~108%/yr • samsung (SMSN) longs: ~0.36%/day, ~133%/yr • brent shorts pay the other way, ~66%/yr • 8 of the top 20 markets by oi sit at the 5.5%/yr baseline mark is ~0.3% over the oracle on both. seoul is open, nyse isn't, and longs are paying to stay in funding resets every hour, so this is a snapshot, not a yield

Terminal showing a real Hyperliquid API response for trade.xyz SK Hynix (SKHX) and Samsung (SMSN) perps, pulled Nov 10 2026 7:55 AM IST. Hourly funding 0.0001229 and 0.0001518, which is about 0.30% and 0.36% a day, or 108% and 133% a year at that hour's rate, versus a 5.5% a year baseline.

ALT Terminal showing a real Hyperliquid API response for trade.xyz SK Hynix (SKHX) and Samsung (SMSN) perps, pulled Nov 10 2026 7:55 AM IST. Hourly funding 0.0001229 and 0.0001518, which is about 0.30% and 0.36% a day, or 108% and 133% a year at that hour's rate, versus a 5.5% a year baseline.

Divyanshu · Nov 9, 2026 · 5:16 PM UTC

Replying to @DigitalAssets

@DigitalAssets just called @solana "an emerging piece of financial market infrastructure" the useful part of the report isn't the headline it's the denominators

Card on Fidelity Digital Assets' Solana report: 15% of exchange trades counted by trade count not dollars; revenue from competition to be first; tokenized asset volume under $100m to about $5b a month; $590b+ Solana perp volume; trade.xyz $2.18b in 24h.

ALT Card on Fidelity Digital Assets' Solana report: 15% of exchange trades counted by trade count not dollars; revenue from competition to be first; tokenized asset volume under $100m to about $5b a month; $590b+ Solana perp volume; trade.xyz $2.18b in 24h.

more replies

Divyanshu · Nov 9, 2026 · 5:16 PM UTC

and the menu is widening • tokenized asset volume on solana: under $100m a month → nearly $5b • solana perp dexs: $590b+ notional in 12 months • the venue fidelity charts them against: @HyperliquidX this morning one hip-3 dex there, @tradexyz, did $2.18b in a single 24h window

Divyanshu · Nov 9, 2026 · 5:16 PM UTC

fidelitydigitalassets.com

my take price exposure goes where market makers will pay to be first ownership settles where the register sits tokenized stocks need both, and chains are already specializing into one or the other report: fidelitydigitalassets.com/re…

Solana's Competitive Advantage in On-Chain Trading

Education and Insights Explore Solana's competitive position as a trading platform, including network scale, execution quality, tokenization, and implications for SOL.

Divyanshu · Nov 9, 2026 · 3:02 PM UTC

Replying to @METR_Evals

@METR_Evals spent 11 weeks with @AnthropicAI's internal data asking one question is ai speeding up ai research? their answer: anthropic is likely improving models on well-specified tasks at ~2x the pace of march 2024 – october 2025 and most of that speedup is ai doing the r&d

Card summarising METR's assessment of AI R&D acceleration at Anthropic: about 2x pace on well-specified tasks, 1.9x AECI growth (95% CI 1.5-2.8x), about 80% attributed to AI, about 12x self-reported staff uplift, no clear bottlenecks for 12 months.

ALT Card summarising METR's assessment of AI R&D acceleration at Anthropic: about 2x pace on well-specified tasks, 1.9x AECI growth (95% CI 1.5-2.8x), about 80% attributed to AI, about 12x self-reported staff uplift, no clear bottlenecks for 12 months.

more replies

Divyanshu · Nov 9, 2026 · 3:02 PM UTC

the caveats are the real read • anthropic redacted a majority of the evidence metr found most informative • robustness checks land anywhere from 1.1x to 2.9x • add non-anthropic models to the trend and it drops to 1.4x • model 4 already scores >50% on every benchmark it was tested on, the ruler is running out

Divyanshu · Nov 9, 2026 · 3:02 PM UTC

my take the frontier isn't only a compute race anymore, it's a loop where the model helps build the next one if you're building on agents, the model you ship on today is the slowest one you'll ever use, design for that report: metr.org/blog/2026-11-05-ant…

Independent Assessment of AI R&D Acceleration at Anthropic in October 2026

METR's high-access assessment of the degree to which AI is accelerating AI research and development at Anthropic.

metr.org

Divyanshu · Nov 9, 2026 · 11:28 AM UTC

metr spent 11 weeks checking whether claude is speeding up anthropic's research to sort the researcher interviews, they asked claude claude's verdict on whether claude helps: • yes: 2 • leans yes: 3 • borderline / unclear: 4 • no: 2 more modest than most founders tbh

Divyanshu · Nov 9, 2026 · 9:57 AM UTC

pro 100 = 10x pro 200 = 20x pro 500 = 50x so pro 50 = 5x the only thing in ai still scaling linearly is the pricing page

Tibo

@thsottiauxNov 8

Would a Pro 50 make sense? Wdyt?

Divyanshu · Nov 9, 2026 · 6:56 AM UTC

International Cyber Digest

if you use claude code, check your repos today "yes, and don't ask again" saves the approved command as an allow rule in .claude/settings.local.json if that command had a password in it, so does the file claude code keeps the file out of git through your global git excludes, which live on your machine, not in the repo apiiro found 1,379 secrets in 1,122 public copies of it → git log --all -- .claude/settings.local.json → rotate anything that ever sat in it → add the path to the repo's own .gitignore → approve patterns, pass secrets via env vars

@IntCyberDigestNov 8

‼️ Click "don't ask again" in Claude Code and it can save the approved command, password included, in plain text in a project file. Researchers found 1,122 public copies of that file on GitHub holding secrets, including 379 database passwords and 23 AWS keys. Of 16 coding agents compared, only Claude Code saves the full command this way by default. Bug reports about it were closed as "not planned" by a bot.

Image from International Cyber Digest (@IntCyberDigest)

121

Divyanshu · Nov 9, 2026 · 4:47 AM UTC