Daniel Oliveira

$ARE valuation is ridiculous now. Mostly because momentum traders and algos just look that occupancy and rent forecast for the full year were revised down. What they don’t see is the footnotes that explain this happens because they no longer intend to sell a range of properties. Also capitalized interest is dropping because construction is ahead of time and thus FFO for Q4 was revised down. For me this is the most important and bullish insight of the full report. “Our prior guidance for occupancy percentage as of December 31, 2026 and 2026 same property performance net operating income changes assumed a benefit of approximately 2% and 3%, respectively, related to a range of assets with vacancy that could potentially be sold in 2026 and/or qualify for classification as held for sale by December 31, 2026 but had not yet met such criteria as of December 31, 2025. Our updated guidance for these metrics assumes a reduced benefit of approximately 1% and 2%, respectively, related to a range of assets with vacancy that could potentially be sold during 2026 and/or qualify for classification as held for sale by December 31, 2026, **primarily due to our revised expectation that we may no longer sell as many assets with significant vacancy driven, in part, by positive leasing prospects for certain spaces that are currently vacant**”

Life science REITs are facing a brutal near-term backdrop, but that is exactly why they are interesting. The oversupply is real, but the long-term demand drivers remain powerful, especially if AI accelerates healthcare innovation and biotech formation. $ARE

Seems like $EBS finally has a pipeline for strong revenue growth. This 2.5x P/EBITDA makes no sense, they are literally buying 2% of shares outstanding EVERY quarter. Almost 20% short ratio. Hope to see a short squeeze, but even if it doesn’t happen my calls are going to print tomorrow.

$ARE I think this footnote makes a nice summary of how the market has already bottomed and is set for a recovery by the end of 2026. “Our prior guidance for occupancy percentage as of December 31, 2026 and 2026 same property performance net operating income changes assumed a benefit of approximately 2% and 3%, respectively, related to a range of assets with vacancy that could potentially be sold in 2026 and/or qualify for classification as held for sale by December 31, 2026 but had not yet met such criteria as of December 31, 2025. Our updated guidance for these metrics assumes a reduced benefit of approximately 1% and 2%, respectively, related to a range of assets with vacancy that could potentially be sold during 2026 and/or qualify for classification as held for sale by December 31, 2026, **primarily due to our revised expectation that we may no longer sell as many assets with significant vacancy driven, in part, by positive leasing prospects for certain spaces that are currently vacant**”

@brunorfacoelho @CoachDanGo 😀

@CoachDanGo @daniel_not_nerd they’re going for you cheat code

The five levels of health: Level 1 - No watch Level 2 - Apple Watch Level 3 - Whoop Level 4 - Garmin Level 5 - Casio

@KobeissiLetter @grok what is the source of the data? Isn’t short interest just reported twice a month?

We are witnessing a historic short squeeze right now. The S&P 500 has now added nearly +$6 TRILLION in market cap since March 30th. In just 5 trading days, hedge fund short exposure to US ETFs has gone from the highest since May 2025 to lower than 97% of cases over the last 5 years. Meanwhile, the capital that was sidelined amid the Iran War is quickly rotating back into AI stocks. The reality is that AI names have only gotten bigger amid the volatility of the Iran War. Stocks like Nvidia and Apple were nearly half as cheap as Costco and Walmart on a Forward P/E basis after the recent correction. All while 4% inflation is back and investors are searching for any source of yield as a hedge. Record highs are on the horizon.

$ARE price action lost touch with reality, the stock can double from the current price + dividends.

$ARE great report on the life sciences real estate sector cushmanwakefield.com/en/insights/li…

@KobeissiLetter Doesn't hold if you do from 2005 until today

There is a near-perfect correlation between US oil prices and US CPI inflation, as shown in our below analysis. As WTI crude surges above $112/barrel, we believe the US economy is bracing for 3.5%+ CPI inflation, particularly if current prices persist through April. Asset owners will be the only winners in the long-run.

@bcherny Will this also be open-sourced?

Today we're excited to announce NO_FLICKER mode for Claude Code in the terminal It uses an experimental new renderer that we're excited about. The renderer is early and has tradeoffs, but already we've found that most internal users prefer it over the old renderer. It also supports mouse events (yes, in a terminal). Try it: CLAUDE_CODE_NO_FLICKER=1 claude

@Fried_rice It did not rewrite anything; it just migrated some files to python, over 90% of the code is missing, and that project is not runnable.

Claude code source code has been leaked via a map file in their npm registry! Code: …a8527898604c1bbb12468b1581d95e.r2.dev/src.zip

Also two anti-distillation mechanisms: 1. Fake tool injection: When sending requests to the API, Claude Code sets anti_distillation = ['fake_tools'] on the request. This is only for first-party CLI users, gated behind a GrowthBook flag (tengu_anti_distill_fake_tool_injection). The API presumably injects fake/decoy tool definitions or outputs into the conversation so that anyone training on Claude's outputs (distilling the model) would learn garbage tool schemas alongside real ones — poisoning the training data. 2. Connector text summarization: The API has a connector_text content block type (same level as text, thinking, tool_use). When the model produces text between tool calls (the "reasoning" you see in the terminal), the API can: - Buffer that assistant text - Summarize it server-side - Return the summary with a cryptographic signature (same mechanism as thinking blocks) - On subsequent turns, the original gets restored from the signature

dug through the leaked claude code source. 44 hidden feature flags. 20+ unshipped features... background agents, multi-agent orchestration, cron scheduling, voice mode, browser control. it's all built.. no wonder they releasing a new feature every 2 days.. everything already done..

Here's what Claude Code source reveals. 512K lines of TypeScript. Here's how it actually works. When bash runs long, your terminal shows live output every second. The model never sees it, progress gets stripped before going back to the API. It blocks until the command finishes and gets the final dump. That's why "run in background" exists. The system silently injects "reminders" into tool results and messages, not from you, not from the model. Available tools, pending tasks, plan rules, memory. When you reject a tool call, it appends a hint telling the model your next message probably contains a preference worth saving to memory. It learns from your "no". Tools run while the response is still streaming. Each tool declares if it's concurrency-safe. Search/read operations run in parallel, file edits get exclusive access. Bash failure kills all sibling tools. Results still return in original order. The entire architecture optimizes for prompt cache hits. Most tools aren't sent to the API at start, model only sees names, one bootstrap tool searches for schemas on demand. Between turns, old tool outputs get deleted via cache edits without invalidating the cache. System prompt splits static/dynamic at a boundary marker so the static half stays cached across conversations. Tool availability sent as deltas, not full lists. After each turn, a forked agent (shares prompt cache, cheap) scans the conversation and saves anything worth remembering to disk. Sandbox defends against a real git exploit where planted files trick git into treating a dir as a bare repo to execute code outside the sandbox. Defense: scrubs those files after every sandboxed command. Now the hidden stuff. Every user has a secret companion pet. User ID hashes into one of 18 species — dragon, axolotl, capybara, one called "chonk." 1% legendary, 1% shiny, stats: DEBUGGING, PATIENCE, CHAOS, WISDOM, SNARK. Full ASCII sprites. Feature-flagged off. Can't cheat it —species regenerates from your hash every session. KAIROS (unreleased, 57+ references) — autonomous agent mode. Runs between your interactions. Daily logs, GitHub webhook subscriptions, push notifications, persistent cron jobs. Sleeps timed to prompt cache TTL to minimize cost. ULTRAPLAN — 30-min planning that teleports work between local and cloud via "CCR" (Claude Code

@BillAckman The problem with taking Abu Musa, Greater Tunb, and Lesser Tunb is that it puts US military personnel at risk, which is going to take out a lot of US lives. This post that you like so much never mentions human lives.

By far the best piece I have read on U.S. strategic options in Iran.

@melvynx People are not lying, @claudeai has been adjusting the token budgets quite frequently this last week. On Monday I hit my limits on my Max 20 plan with a few prompts, yesterday I was running multiple terminals and I couldn’t even use half of it.

I just use non-stop my Claude code sub all day, didn't even reach 100% of session usage. People are lying so much on X.

Seems like @claudeai is reducing daily and weekly budgets for everyone without informing users… This is what happens when pricing is not transparent and doesn’t mention how much usage are we actually buying.

$9-10 trillion in U.S. debt matures in the next 12 months. If refinanced today at 4.4% instead of 2.3-2.6%, that's an extra $180-200 billion yearly in interest costs. That's 4% of the federal budget. Trump needs to end the Middle East conflict to bring oil prices down so the Fed can safely cut rates without risking inflation and help manage this massive debt rollover. Iran doesn’t want a deal, Trumps wants one and actually needs it.

BREAKING: Iran says it has reviewed the US' peace plan proposal and it considers the ask to be "excessive." Iran also says it will not allow President Trump to dictate the timing of the end of the war.

$ARE price being artificially suppressed with very low volume in the past weeks

@Al_Grigor Have the same issue since yesterday. Anthropic is ignoring customers and not justifying. They could be honest and assume they have a bug or that they just reduced the budget.

I hit my limits very quick this week - even with 20x pro plan. It makes my claude code unusable A good reason to do more stuff with Codex!

@trq212 @trq212 have you guys fixed the usage counter? It has been broken since this morning. I am on 20x, and one single prompt can take 10% of my 5H usage

Claude Code can control your computer now!