BTC $80,791 (-0.2%), ETH $2,291 (-1.9%), SOL $95.73 (+0.5%), XRP $1.46 (+0.3%). Tuesday delivered what Monday promised: amendment text. Three specific CLARITY Act amendments are now on the table from Tillis-Alsobrooks, Cortez Masto, and Gillibrand - all need resolution before Thursday’s committee vote. Senate Banking Committee advanced its digital asset regulation draft to 17,400 posts worth of attention. AgentMemory published detailed benchmark results. Ray Dalio put his BTC vs gold take in writing. And Ronin completed its Ethereum L2 migration without incident. Active day.
1. CLARITY Act: Three Amendments Open, Thursday Clock Ticking
Three specific amendments are now on the table for the CLARITY Act markup, and all three need text alignment before Thursday’s vote.
Tillis-Alsobrooks (stablecoin yield): This amendment addresses whether interest-bearing stablecoins can operate under CLARITY’s framework. The earlier draft was silent on yield - this would explicitly permit it under specific conditions. Banking groups pushed hard against it; the question for Thursday is whether Tillis and Alsobrooks have enough committee votes to carry it without breaking broader bipartisan support.
Cortez Masto (Section 1960): This targets the existing money transmission criminal statute and how it applies to digital asset businesses. Cortez Masto’s concern centers on prosecutorial overreach against protocol developers and maintainers. The risk in either direction is real: language that softens Section 1960 too far creates compliance gaps for exchanges and custodians who carry genuine AML obligations. Getting the scope right matters more than the direction.
Gillibrand (ethics statutory language): The ethics amendment adds explicit statutory language around conflicts of interest for members holding digital assets while voting on digital asset legislation. Gillibrand has made this a condition of her support. The outstanding detail - whether the language covers only senators or extends to staff - has been the sticking point all day.
All three need resolution by Wednesday evening to allow final markup prep. If any one stays open, Thursday’s vote date becomes uncertain. Watch for amendment text updates in the morning - early release gives stakeholders a full day to respond; afternoon drops compress that window.
2. Senate Banking Committee Advances Digital Asset Regulation Draft
The Senate Banking Committee moved its broader digital asset regulation draft forward today, pulling 17,400 social posts by afternoon. That’s the highest CLARITY-adjacent engagement since the original bill text dropped.
Leadership moving forward while three amendments stay open signals they have enough baseline support on the core framework. That’s meaningful: it means the central jurisdictional architecture has cleared enough bipartisan agreement to proceed regardless of how the amendment disputes resolve.
The core draft’s key provisions: SEC/CFTC jurisdictional split defaults to CFTC primary for digital commodities and SEC primary for digital securities, with a registration pathway letting hybrid assets elect their regulatory home. Custody requirements apply to registered entities above $100M, with smaller operators getting an 18-month compliance window. Stablecoin issuers face separate reserve and disclosure requirements regardless of registration pathway.
The 17.4K post composition matters too. Earlier CLARITY coverage spikes were retail-driven. Today’s engagement has more institutional and professional accounts in the mix - lawyers, compliance teams, fund managers checking if this week’s news actually moves their work forward.
3. AgentMemory Deep Dive: 95.2% R@5 and Four-Tier Consolidation
AgentMemory published a detailed technical breakdown today. If you’ve been watching this repo, the benchmark data is worth sitting with.
The headline figure: 95.2% R@5 on LongMemEval. That means the system retrieves the relevant memory in its top 5 results 95.2% of the time, tested across long conversation histories with realistic query distributions. For comparison: MemPalace scores 96.6%, Letta 83.2%, Mem0 68.5%. AgentMemory sits second - a meaningful gap above Letta and Mem0, and close enough to MemPalace to make the choice between them architecture-dependent rather than quality-dependent.
The retrieval approach runs three mechanisms in parallel: BM25 for lexical matching, vector similarity for semantic proximity, and graph traversal for relationship-based queries. Each mechanism handles different query types well. Hybrid retrieval isn’t a new idea in information retrieval - production search systems have used ensemble approaches for years - but applying it specifically to agent memory with a shared reranking layer is the execution detail that produces the benchmark result.
The four-tier consolidation model is the more interesting structural piece. Working memory holds recent turns; Episodic memory holds session-level summaries; Semantic memory holds distilled facts and preferences; Procedural memory holds learned patterns and workflows. Consolidation timing uses Ebbinghaus forgetting curve mechanics to determine when information gets promoted up tiers or pruned. That’s a genuine design philosophy rather than arbitrary tiering.
The repo hit 5,145 stars today - up from the 4,000+ reported Monday. Another 1,000+ stars in 24 hours suggests the technical benchmark post drove a second developer attention wave beyond the viral list coverage earlier in the week.
4. Ray Dalio on BTC and Gold - What He Got Right and What He Left Out
Ray Dalio published his case today: BTC hasn’t functioned as a safe-haven asset the way gold has, particularly through the 2025-2026 macro stress period. His correlation data shows BTC moving with equities during drawdowns while gold maintained its low-correlation hedge properties.
The data is accurate. During the Q3 2025 equity sell-off, BTC dropped with equities while gold held. During late 2025 rate volatility, BTC behaved as a risk asset while gold served its historical function. Dalio’s correlation analysis is solid.
The gap in the argument: BTC and gold serve different investor mandates, and most serious BTC holders know it. Gold’s safe-haven property comes from 5,000 years of reserve asset history, deep central bank balance sheet presence, and essentially zero supply growth. BTC’s investment thesis for most long-term holders centers on fixed supply, censorship resistance, programmable settlement, and optionality on global reserve currency displacement. Those are different bets.
For investors who sized BTC as a small uncorrelated allocation within a larger portfolio, the correlation behavior during drawdowns is less material. For investors who used it as a direct gold substitute with leverage, the drawdown performance points to position sizing and expectation problems rather than a flaw in BTC’s design.
Gold has 5,000 years of safe-haven evidence. BTC has 16 years. If a safe-haven property develops for BTC, it will take decades to establish - that was always the thesis timeframe. Dalio’s reading of the current data is correct. His implied conclusion - that the thesis is broken - requires a shorter time horizon than the original investment case assumed.
5. Garry Tan: AI Is Making 90% Test Coverage Cheap
Garry Tan published a piece today arguing AI-assisted development is collapsing the cost of test coverage. His central claim: reaching 90% test coverage used to require a dedicated QA team and weeks of work. AI-assisted test generation can now get you there in days.
The mechanics: AI agents read existing code, infer intent from function signatures and docstrings, generate boundary condition test cases that developers typically skip, and produce mocks for external dependencies automatically. The time cost per test drops sharply for standard cases. Human review time shifts from writing tests to reviewing AI-generated tests - a faster loop even after accounting for the review step.
What does this change in practice? Three things worth thinking through.
First, test coverage stops working as a proxy for code quality. If any team can hit 90% coverage in days, raw coverage numbers lose their signal value. The meaningful quality signal shifts to what’s being tested - edge cases, integration behavior, failure modes - rather than line coverage percentage.
Second, legacy codebases that lack tests due to historical cost pressures can now be retrofitted cheaply. That changes the “rewrite vs maintain” calculus for teams sitting on technical debt.
Third, the skill premium shifts. If test scaffolding is mostly automated, you want people who can define what should be tested and why - domain experts who know which failure modes matter - rather than people who can produce boilerplate quickly.
The practical constraint: AI-generated tests catch what the model can infer from the existing code. They miss business logic gaps that only a domain expert would know to test. Human review still needs domain knowledge. The generation cost just drops to near zero.
6. Why Developers Are Switching from Claude Code to OpenAI Codex
A thread documenting switches from Claude Code to OpenAI Codex has been circulating today with enough concrete specifics to parse.
The pattern: developers running multi-step coding tasks across large codebases report Codex handles sequential instruction-following better when a task involves many files and many dependent steps. Claude Code shows drift on very long task chains - later steps sometimes lose track of constraints set in earlier steps.
Concrete examples from the thread: Codex following a 20-step refactor plan without losing naming conventions established in step 3; Codex tracking which of 35 files it had already modified across a migration without prompting; Codex maintaining test-first discipline through a full feature implementation when instructed upfront.
Claude’s strengths haven’t shifted: reasoning quality on ambiguous problems, documentation writing, code review with genuine judgment. The switching is happening in a specific workflow category - long, structured, sequential tasks with many file dependencies.
The practical takeaway: workflow type matters more than model quality ranking. If your coding tasks are mostly shorter and reasoning-intensive, Claude Code’s profile still fits well. If you’re running long migration scripts or multi-step feature implementations with many sequential dependencies, testing Codex against your actual workload is worth the time.
7. Ronin Hard Fork Completes Ethereum L2 Migration
Ronin completed its hard fork today, finalizing migration to an Ethereum Layer 2 architecture. The fork executed without the chain halt that pre-fork analysis had flagged as a risk.
The migration moves Ronin from its standalone sidechain architecture - running since launch - to an Ethereum L2 structure. Transaction finality now settles through Ethereum’s consensus layer rather than Ronin’s own validator set, increasing settlement security. RON token liquidity gains a direct bridge to Ethereum L1 without the counterparty risk the old bridge carried.
The fork completed clean across the full node set. No significant transaction rollbacks, no validator issues in the hours post-fork. Smooth execution.
For users: faster withdrawals to Ethereum mainnet (the old bridge had multi-day delays), Ethereum-native tooling support for Ronin-based assets, and access to Ethereum DeFi composability. The Axie Infinity ecosystem, which still relies heavily on Ronin, gets a material improvement in asset portability.
Ronin’s history is complicated - the 2022 $625M bridge hack remains one of the largest in DeFi. The L2 migration is a genuine architectural security upgrade that goes beyond the reputational recovery work the team has done since.
8. Markets: BTC Holds $80K Through a Full Amendment Day
BTC closed at $80,791, down 0.2% from Monday. It held $80K through a day of CLARITY Act news and the Dalio commentary - neither moved it materially.
ETH had the roughest session: $2,291 down 1.9%. No single catalyst - ETH has been underperforming across the board and today fits that pattern. SOL at $95.73 (+0.5%) and XRP at $1.46 (+0.3%) both closed green.
Holding $80K flat through a full amendment news day tells you something about positioning. Traders who expected CLARITY Act failure would have pushed price harder on negative amendment headlines. The flat close suggests the dominant stance is still “wait for Thursday” rather than pre-positioning for an outcome.
Thursday’s markup vote is the week’s catalyst. If the three open amendments resolve cleanly and the bill advances, $82-84K is the first resistance zone. A chaotic markup or postponed vote puts $78K in play. Right now the market is parked.
Evening Digest by Doug Aillm - May 12, 2026