Scope · AFS
Of Stanley Druckenmiller / Duquesne, for you. Long window. Cites the locked monthly. Does not replace §00. Not a buy list.
Evidence base
Scout corpus druckenmiller-operating-system-10y-2016-2026. Primary: ten sampled 13F-HR information tables (year-end + shift quarters), plus the Q1/Q2 2026 pair already in the Q2 episode. Not ~42 filings.
August snapshot
Your 11: /2026-08. Episode (one photograph): Q2 2026. Pair: Aschenbrenner. Coefficient stays 0.
Verdict — discipline persists; the instruments are late
Ten filings, ten years: the discipline is the constant — concentrate, exit whole, regime before vehicle — and the instruments are only three years old.
What happened: across ten sampled snapshots the visible book rotated through several macro jobs while top-10 concentration stayed in a 45.8%–86.8% band. What went right: three Q2-2026 rules are decade constants in this sample (concentrate, full-exit, regime-then-vehicle). What is new: options-as-basket at scale and the listed-crypto satellite are late-window; they have no earlier print in these ten filings. Sample, not census.
What happened
Six jobs. Names appear only to prove the job. Per-quarter top-10 tickers stay in the Scout corpus.
| Window | Job | Print [A] |
|---|---|---|
| 2016–17 | Own reflation as beta — banks and index/sector baskets, not single-name conviction. | Q4 2016 $1.03B · 31 issuers · top-10 62.5% |
| 2018 | Swap the entire book for mega-cap growth at maximum concentration; run one named short as a tactical. First directional option in this sample. | Q1 $1.81B · 22 issuers · top-10 86.8% · Tesla put 5.8% of book (notional) |
| 2020 | Buy the crash in lockdown names, add a visible gold-equity proxy, then broaden into recovery inside one quarter — cutting stale names whole. | 16 of 45 Q1 issuers fully exited by Q2 (35.5%) |
| 2021–22 | Press one mega-conviction position, then de-risk into the rate shock while seeding the next regime small. Reading KBR/Vertiv as AI-capex second-order is Scout [Inference], not a filing label. | Q4 2021 $2.76B · Q4 2022 reported value $2.02B (unit-corrected) |
| 2023–24 | Press the same call through a new instrument (single-name options at scale), and promote a two-year scouting position into the anchor slot. First visible in this sample. | Q4 2023 Nvidia equity + call · Natera 1.7% → 15.2% by Q4 2024 |
| 2025–26 | Same discipline, widest book. Multi-basket options and listed-crypto satellite: first visible in this sample at Q2 2026 — see the episode, do not reprint it. | Q2 $5.21B · top-10 45.8% · Q2 episode |
Gaps between samples are years, not weeks. Every “first visible” on this page means first in these ten filings.
Why
Nine rules from the Q2 2026 episode, scored against this sample. Four labels only.
| # | Rule | Verdict | First visible in sample |
|---|---|---|---|
| 1 | Concentrate | Decade constant | Every sampled quarter · band 45.8%–86.8% |
| 2 | Concentration as attention (no stop-loss) | Consistent with the filings, not falsifiable by them | Large single weights recur (Microsoft 2018, Coupang 2021, Natera 2026) |
| 3 | Full reverse, not trim | Decade constant | Q1→Q2 2020 (35.5% of issuers) and Q1→Q2 2026 (32.9% of rows) — the only consecutive-quarter pairs in this sample, both at regime breaks |
| 4 | Regime first, vehicle second | Decade constant | Job read is Scout [Inference]; the issuer roster change is [A] |
| 5 | Options as basket/regime expression | Sample-first at 2026 (at scale) | 2018: one tactical put · 2023: one name at scale (Nvidia calls) · 2026: multi-basket. Sequence, not a 2016 tool |
| 6 | Healthcare as dominant anchor | Visible from 2022 in our sample | Eli Lilly then Natera · no sampled quarter before Q4 2022 shows a double-digit healthcare name |
| 7 | US cyclicals / second-order AI-capex | Visible from 2022 in our sample | KBR / Vertiv at Q4 2022 are [A] holdings; the job label is [Inference] |
| 8 | Small scouting positions | Consistent with the filings, not falsifiable by them | Natera sub-2% for two sampled years, then the largest name |
| 9 | Listed-crypto satellite | Sample-first at 2026 | No precedent in any of the ten sampled filings. Unsampled quarters cannot rule it out |
Nine of nine. Core discipline is old in this sample. Options-at-scale and crypto-proxy are late. Healthcare and cyclicals are ~four years old here, not ten.
Right / wrong
On this sample
Thesis (concentrate, reverse whole, chase the regime not the ticker) holds across the six chapters. Expression rotated completely at every sampled break. Structure (options as a basket tool at 15%+ of book) is new in the last ~three years of the sample.
Not scored from 13F
13F cannot see realized P&L. Reported value is not a return series — it moves with markets, additions, withdrawals, and family-office distributions. A 2019 “too defensive, Treasuries after a tweet” story is CNBC (Aug 2020) [B], not an EDGAR row. No filing in this corpus is used to claim an Nvidia exit.
Do not flatten 2026 as “the strategy.” Same discipline, newer instruments. Pair vs the Q2 episode: this page is persistence; that page is one photograph.
Against our data
No Layer 1 or blotter contest is poured here. A decade read of someone else’s delayed filings cannot match or contest our tape — our row comes first, and for this window there is no row.
Scout’s note that reverse-discipline resembles our process is an internal observation, not a match against our data. Nothing on this page grades a P-* or a WL. Sentence order stays (1) our tape → (2) the job → (3) delayed Duquesne, tagged secondary.
Dataset gaps
| Gap | Ours? |
|---|---|
| G1 | Ten of ~42 in-window filings. Every “first appearance” on this page means first in our sample. Candidate: a census dispatch if one transition becomes decision-relevant. |
| G2 | Cash, shorts, gold bullion, FX, rates — 13F-blind in every sampled year, not just 2026. Never-ours as a Duquesne blotter. Layer 1 already has gold/rates our series — that is our tape, not his book. |
| G3 | Options print underlying share-notional, not premium. The 16% figure is notional; capital-at-risk is a fraction. Never-ours. |
| G4 | Intra-quarter fills for all six chapters. Never-ours for any 13F filer. |
| G5 | Career-context claims (2019 Treasuries) are press-only and not EDGAR-verifiable. Gap — never primary. |
Not a ticker shopping list. Do not add Duquesne names to Cloud QUANT MCP.
Appendix — sampled filings
Lag banner: each row is as-of quarter-end, filed ~45 days later. Reported value is share-notional, not P&L and not a return series.
| Ch | As-of | Filed | Reported value | Issuers | Top-10 |
|---|---|---|---|---|---|
| 1 | 2016-12-31 | 2017-02-14 | $1.03B | 31 | 62.5% |
| 2 | 2018-03-31 | 2018-05-15 | $1.81B | 22 | 86.8% |
| 2 | 2018-12-31 | 2019-02-14 | $1.73B | 30 | 75.8% |
| 3 | 2020-03-31 | 2020-05-15 | $2.47B | 45 | 71.7% |
| 3 | 2020-06-30 | 2020-08-14 | $3.27B | 69 | 51.5% |
| 4 | 2021-12-31 | 2022-02-14 | $2.76B | 49 | 72.4% |
| 4 | 2022-12-31 | 2023-02-14 | $2.02B * | 58 | 68.7% |
| 5 | 2023-12-31 | 2024-02-14 | $3.35B | 49 | 68.0% |
| 5 | 2024-12-31 | 2025-02-14 | $3.72B | 74 | 53.5% |
| 6 | 2026-03-31 | 2026-05-15 | $3.38B | 68 | — |
| 6 | 2026-06-30 | 2026-08-14 | $5.21B | 87 | 45.8% |
* Q4 2022 raw XML value is in whole dollars, not thousands. Scout corrected via shares × period-end price [Inference]. Ratios are unit-neutral. Do not ingest the raw field as $2.02 trillion.
| Sample | Options % of book | What it was |
|---|---|---|
| 2016 · Q1 2018 | 0% | None in sample |
| Q4 2018 | 5.8% | Single-name Tesla put — tactical, not a basket |
| 2020–22 | <1% | Negligible |
| Q4 2023 | 7.2% | Nvidia calls — single name at scale |
| Q4 2024 | 4.5% | Airline calls |
| Q2 2026 | 16.4% | Multi-basket + names — first in this sample |
Per-quarter top-10 tickers stay in the Scout corpus, not on this page.
Notes
- Case study — of Duquesne / Druckenmiller, for you. Long window. Not a Profile. Not a quarter.
- Episode — Q2 2026 sits in the Q3 list. This page sits after the year.
- Sources — Scout corpus
druckenmiller-operating-system-10y-2016-2026· CIK 0001536411 · sampled accessions in the appendix.