Building While They Were Breaking: What We Shipped in Q2 2026
By Messy Virgo — July 9, 2026
Q2 2026 was not a quiet quarter in crypto. Liquidations stacked up. Builders rage-quit on X. Every other project promised autonomous agents by Friday.
We kept shipping anyway.
This is not a victory lap. It is a receipt — what landed in git, on the blog, and in public fund surfaces between April 1 and June 30, 2026, and why it matters if you care about AI-managed funds that can actually be trusted.
Note: Our funds remain pre-live and human-reviewed. Q2 advanced the workflow; it did not flip a switch to autonomous production capital. Proof before autonomy — always.
The quarter in one sentence
Messy Virgo turned its AI fund thesis into a public proof stack: staged screening with dated artifacts, recurring fund updates, governed council sessions, and a decision-to-execution bridge — all inspectable on the website while the code kept moving underneath.
By the numbers
Evidence from git history across our core repos (messyvirgo-platform, messyvirgo-com, messyvirgo-org) for Q2 2026:
| Metric | Q2 total |
|---|---|
| Commits (all three repos) | 617 |
| Platform commits | 531 |
| Website commits | 80 |
| Platform commit growth | Apr 92 → May 156 → Jun 283 |
| Commit themes (platform) | council 138 · fund 81 · cli 59 · web 44 · agent 38 · screening 27 |
On the public surface we also shipped 13 weekly build log entries, 16 fund-update cycles, multiple long-form essays, and monthly treasury transparency posts — not as marketing filler, but as the proof layer the homepage promises: Funds are the product. Research tools are the engine. Updates are the proof.
April: Screening became a workflow
The quarter opened with a structural shift: screening stopped being a concept deck and became a dated operating pipeline.
We published the two-part screening series — Part 1: from catalog to daily substrate and Part 2: daily runs, aggregates, and configurable playbooks — describing how Messy prepares the market before a fund ever screens:
- Monthly token catalog maintenance
- Weekly context refresh (macro, narrative, security)
- Daily due diligence on hundreds of Base tokens
- Saved screening runs and aggregate candidate views per fund sleeve
Platform work matched the story: token catalog split from weekly screening, per-sleeve UTC run days, aggregation presets, and inspection surfaces for historical comparability.
Why it mattered: A fund agent cannot be audited if its inputs are ad hoc. April gave Messy repeatable artifacts — the same question, the same date, the same shortlist, week after week.
May: The proof layer went live
May was about making the engine visible.
Public fund surfaces
We established the weekly Fund Update rhythm on the website — frozen snapshots of macro regime, narrative momentum, screening aggregates, and (later) council notes. Guru Lotus read-only fund views landed. The homepage reframed around a simple sequence: research engine → pre-live funds → allocation logic → future AI-managed funds.
AI evals, not AI demos
On May 23 we published How We Benchmark AI Agents Before Production — documenting three-cycle, production-path evals for context news across eight model configurations, selecting Claude Haiku 4.5 for operational fit (memory, repair friction, recall, runtime stability) rather than one-off prose quality.
Many projects demo outputs. We documented how a model survives the workflow around those outputs.
Transparency as product
The monthly Every Month, We Buy More Of The Same Token You Hold series continued — May costs settled without DAO token dumps, with co-founders ending the month with more MESSY exposure, not less.
We also shipped trust and governance copy updates, May raffle rules, and the long-horizon essay An AI End-State: What Messy Virgo Becomes.
Why it mattered: May proved Messy could run a public operating cadence — not just ship features behind a login.
June: The council became real
June was the defining month.
The decision pipeline went public
We published the two-part decision pipeline arc:
- Part 1: The AI Investment Council — PM agents, risk function, chair, motions, votes, recorded resolutions
- Part 2: The Decision-to-Execution Bridge — the deliberate gap between what the fund should hold and what trades are safe to execute right now
The bridge is the part most AI trading stacks skip. We did not.
Workflow-traded micro test funds
By the June 27 fund update, Guru workflow-traded micro test funds (mvf-base01, mvf-base02) were surfacing in public:
- Council session counts and rotation outcomes
- Full council minutes on fund pages
- Screening → council review → signed execution as a named path
Platform commits in June concentrated on council runtime, board-minutes protocols, session inspector UI, Guru Lotus execution bridging, macro analyst strategy, Velora route selection, and PM agent tooling — 138 Q2 commit subjects mentioning council alone.
Strategy and positioning essays
June content also included If AI Can Build Anything, Why Isn't Everyone Rich?, Why Messy Virgo Is Built to Last, State of the Stack, and the trilogy capstone In a World Where AI Builds Everything, Trust Is the Only Moat.
Why it mattered: June is where Messy's core claim shifted from AI-assisted crypto research toward governed AI fund operations with inspectable decision records.
Transparency and governance (the non-agentic wins)
Not every Q2 milestone was a model release. Some of the most credibility-building work was operational:
- Monthly treasury transparency — April, May, and on-chain June treasury movements
- May raffle final distribution — reproducible allocation methodology for 2,000,000 MESSY across eligible wallets
- Swiss association updates — registered office relocation to Glattbrugg and governance copy aligned across the site
For a project claiming AI-managed funds, how you handle treasury, legal consistency, and community accounting is part of the product — not a footnote.
What we did not claim
Worth stating plainly:
- We did not launch fully autonomous production capital in Q2
- We did not remove the human approval gate before execution
- We did not ask holders to trust a black box
The homepage still says funds are pre-live and human-reviewed. The decision-to-execution material says the same. The right framing for Q2 is: major progress toward AI-managed funds, with more of the workflow proven in public.
Where to dig in
If you want the primary sources:
| Surface | Link |
|---|---|
| Weekly build log | /buildlog.html |
| Latest fund update | /fund-update/ |
| Screening series | Part 1 · Part 2 |
| Decision pipeline | Part 1 · Part 2 |
| AI eval discipline | Benchmarking agents before production |
| Treasury | /treasury.html |
Q3 starts here
Q2 closed with CLI-native council runs, Guru full-exit execution, paper-provider parity, recency-weighted screening, and the first version of a seven-rule holding-review doctrine for exit discipline — documented in the June 30 – July 6 build log.
The market will keep breaking things. We will keep shipping the verification layer.
If you are an allocator, a builder, or a holder who wants to see the work — not the pitch — the proof stack is public. Start with the build log and the latest fund update. That is the quarter, in the open.