Luna3.ai Screener

Aschenbrenner runs $5B+ exclusively on the AI compute buildout thesis (power, optical, BTC-miner→AI pivots). When he opens a NEW position it's a high-signal data point — he's the canonical thesis holder. Cross-reference these against the screener tracks below.

Catches names Track D misses: bitcoin miners with DC-ready facilities pivoting to AI hosting (CORZ/IREN/APLD pattern) and power generators feeding the AI buildout (BE/EQT/BW). Their 10-Ks talk "hashrate" or "power purchase agreement," not "data center." Wider mcap range ($300M-$15B) since these names cluster mid-cap.

Captures the full Aschenbrenner archetype validated 2026-05-07 against his Q4 2025 13F: capital-intensive growth at premium price. His holdings cluster at capex/rev 0.38 (89th pct vs baseline 0.04), P/S 10.5x (87th pct vs 2.2x), rev growth +25% (82nd pct vs +5.7%), with debt-funded buildout (D/E 0.85, low cash) and reinvestment (ROE 3.8%). Three thesis kw clusters: data_center (+74.6pp), power_gen (+49.2pp), convertible_growth (+55.5pp). Distinct from Track A (value, P/S<3) and Track G (keyword-only). Re-validation scheduled 2026-05-15 against Q1 2026 13F.

Zero stocks met all hard gates this scan: Price $5–$100 · MCap $300M–$3B · Rev growth >15% · P/S <3 · ROE >15% · Rel Vol >1.5× · Analyst <5. This is normal — the filter set is reverse-engineered from the pre-run metrics of actual winners (SNDK, FIX, CRDO, VRT, BE, IESC, LITE), so most days return 0–2 names.

This tab tracks themes too new for the calibrated screener (Track A/D/E/G). Three layers update on different cadences: Sector ETF momentum refreshes daily — "is this theme heating up RIGHT NOW?" ARK / thesis-fund Watch refreshes quarterly via 13F (45-day SEC lag) — speculative conviction. Emerging keywords (quantum, humanoid, fusion, agent_ai, llm, space_launch) appear as info-only columns on Track D / G cards in the Screener tab. A theme graduates from this tab to a hard gate when Q3+ calibration shows >=30pp differential.

>+15% over 30 days = theme is heating up. Drill into the ETF's holdings to find specific names benefiting. Persistent >+30% YTD = the bull case is being priced in — expect calibration to surface its keywords next quarter.

ARK opens speculative positions ~quarters before mainstream funds. Use as leading indicator for quantum / robotics / space / genomics themes. 8 NEW positions in latest 13F.

What's structurally invisible: private companies (Anthropic, OpenAI, SpaceX, Stripe) have no SEC filings, no insiders, no 10-Ks. They never appear in any track. Their effects propagate via public proxies (NVDA, MSFT, AVGO — all caught by Track D when their narratives align).

Stocks showing the technical signature that preceded prior 1,000% gainers (SNDK, BE, AAOI, IREN, HUT, CIFR …). Two layers: Featured Setups are manually curated — capture the catalyst-on-deck nuance the algorithm can't see (FDA dates, ASCO readouts, hyperscaler announcements). Auto-Detected scans every tracker stock against 10 algorithmic features — volume dry-up, pocket pivot, drawdown depth, ATR compression, Stage 1→2 transition, distribution-day count, RSI entry zone, theme density. A name graduates off this list when it runs >50% from the pivot price.

Same concept as the Signal Tracker on the Screener tab, scoped to Technicals. Each symbol that's ever appeared in Featured Tier 1–3 or Auto-Detected Patterns stays in this list permanently — even after it's removed from the active watchlist — so lifetime return from first scan is always visible. Active = currently in Featured or Auto-Detected. Removed = previously tracked, now off the active list (most likely auto-graduated, agent-removed, or thesis broken).

Source badge: tier-1 · tier-2 · tier-3 = Featured Setups · Track D/E/G/H/A/B/C = algorithmic scan track. State: ACTIVE currently in a list · REMOVED previously tracked. Lifetime return = (Now − Entry) ÷ Entry from first_seen date.

Live columns (Now $ · vs Pivot · Status · Live score) refresh on every screener run (Tue–Sat 06:30 Melbourne post-US-close). Curated columns (Theme · Pivot · Hits · Catalyst · Date) come from tools/screener/featured_setups.json. Status: Basing = below pivot · Coiling = within 10% of pivot · Breaking Out = at/above pivot · Running = +10% past pivot · Broken (red bar) = price <= setup × 0.7. Auto-graduate at +50% from pivot — setup moves to Avoid list automatically. Hover any row for the qualitative thesis.

Each row is scored against 10 features: A drawdown ≥50% · B Stage 1→2 (50/200MA align) · C ATR compression · D basing ≥4 weeks · E volume dry-up (RelVol<0.7) · F pocket pivot · G no distribution (≤2 days) · H recent 8-K + DC keyword (hyperscaler tell) · I hot theme (DC kw ≥5) · J RSI entry zone (40–65). Tier 1 = 8–10 hits · Tier 2 = 6–7 · Tier 3 = 4–5.

Hard disqualifiers (skip even if 8+ hit): active SEC investigation · unresolved going-concern audit · active ATM dilution mid-base · >70% single-customer concentration with declining revenue · major lockup expiring next 30 days. Manual review of Featured setups handles these; Auto list does not — verify before acting.

Live updates: Prices load fresh from Sina Finance API on every page open. During HK market hours (09:30-16:00 HKT, Mon-Fri) the Price + Return columns continue polling every 5s. A-shares are real-time; HK is 15-min delayed (HKEX licensing constraint — affects all free providers). Outside market hours the initial load shows the latest available close from Sina. 实时更新:每次打开页面时价格均从新浪财经 API 即时加载。在港股交易时间(港股 09:30-16:00,周一至周五),价格与收益列每 5 秒持续轮询。A 股为实时数据;港股延迟 15 分钟(港交所许可限制 — 影响所有免费数据源)。非交易时段,初次加载显示新浪最新可用收盘价。

Every stock currently passing either the technical-pattern tiers (T1/T2/T3) or the fundamentals tracks (A-HK / A-CN / D-CN / D-CN-Optical). The Return column shows lifetime gain from first_seen (live-updated during HK hours), so you can see how the screen has performed since it first flagged each name. Stocks that fall out of the screen are removed from this view (still tracked internally for screener state). 当前通过技术形态层级(T1/T2/T3)或基本面赛道(A-HK / A-CN / D-CN / D-CN-Optical)的所有标的。收益列显示自首次出现起的累计涨幅(港股交易时段实时更新),可查看筛选模型自首次标记起的表现。标的若不再通过筛选,即从本视图移除(内部仍追踪以维护筛选状态)。