KEEL leads Orbit’s board at +60.3% since it was screened, and 32% of every name the engine has flagged remains above its entry. But the more useful story this month is not the leaderboard — it is what we changed underneath it. June was a risk-off stretch for the high-beta, small-cap names Orbit tends to surface, and rather than ride it out we used the month to cut four positions on thesis and tighten two of the engine’s quality gates.

What Orbit is

Orbit is Luna3.ai’s AI-augmented research engine. It combines twelve algorithmic signals with a gradient-boosted machine-learning model, then layers an agentic LLM analyst on top that reads each top pick’s filings and writes a daily thesis tagged with a conviction score and a catalyst-proximity read. A name’s entry price and date freeze the moment Orbit screens it — so the board below is the setup the engine flagged in real time, never a back-edited record.

The scorecard

Across every screened name, returns since screened span +60.3% at the top to -52.4% at the bottom.

The honest aggregate: 32% of the board sits above its screened entry, with a -5.3% median and a -2.4% mean. This was a down month for the cohort — the same growth-to-value, rate-sensitive flush that pressured small-caps and biotech broadly hit Orbit’s book directly. The scorecard above includes every name the engine has ever screened, including the four we removed this month; we don’t delete laggards from the record. Not every screen works, and after June a majority of the board sits below where Orbit first flagged it.

The top 20% since screened

The top cohort held up at +22.9% equal-weight even through the drawdown, and the names doing the work cluster where they have all spring: AI and compute infrastructure (KEEL, CLSK), defense and drones (RCAT, KTOS), isotope chemistry (ASPI), and a single binary-catalyst biotech (ERAS) that has already cleared its readout. The leaders are the ones with a concrete, near-term catalyst rather than a vague one — that distinction did most of the separating this month.

What we changed this month

Orbit reviews every name on the board on a monthly cadence, and June’s review produced four removals — all on thesis, not price. We cut a biotech whose catalyst had already fired and faded, a name whose deal-close catalyst slipped almost a year into the future, and two lower-quality names the engine’s own risk layer had already flagged.

A removal isn’t a permanent blacklist. A name we cut can earn its way back — but only if it forms a genuinely fresh setup and re-clears the screen on its own; nothing returns just because time has passed. We keep the full record either way, the winners and the names we cut, so the scoreboard stays honest.

Then we hardened the engine itself, in ways that line up with how the screen already works:

• A trap-risk gate. The agentic analyst already grades each top pick for trap-risk; the engine now acts on a high reading, holding a red-flag name back from the leaders and keeping it on the board with a warning rather than letting a low-quality setup sit up front.
• A catalyst-window flag. Orbit’s screen prizes a binary event scheduled inside roughly six months. Every name now carries a small lifecycle tag — on deck when the catalyst is in that window, fired once it has passed, far when it’s still too distant — so a setup whose catalyst has already played out is surfaced for review instead of quietly lingering.

One honest note: not every loss was preventable. The biotech we cut was a legitimately strong, high-conviction setup going into a binary event that simply disappointed — no pre-event filter would have, or should have, screened it out. The fix there isn’t a filter; it’s the faster post-catalyst review the new fired flag now forces. That is the point of running the review every month: the engine should get a little more disciplined each time, especially after a soft one.

How to read the board

Orbit’s screen has two parts — a fundamental and quantitative track screen, and a technical pre-rally screen. The “Source” column tells you which layer first surfaced each name. Tier 1, 2 and 3 are the top conviction layer: names that cleared both parts of the screen and earned an agentic catalyst review on top, with Tier 1 the highest conviction. A Track tag means one of Orbit’s fundamental/quant tracks caught the name and the technical pattern confirmed it, without the agent review the tiers receive. Agent pick is a name the LLM analyst surfaced directly.

The bottom line: June was the kind of month that tells you more about a process than a green one does. The board took the drawdown the broad small-cap tape handed it, the top cohort held, and Luna3 used the stretch to cut what wasn’t working and wire two new guardrails into the engine — the work that compounds long after any single month’s return.

Read next: Intelligence · Trending