/decisions · 2026-05-18 · investing · memory-cycle-v1.1

Deploy memory cycle v1.1 paper trade (post-backtest spec revision) — go / revise / defer / archive?

Founder gate on the post-backtest revision of the memory thesis. Supersedes the v1 deploy gate from 2026-05-17 (archived). All paper; no live capital at risk.

v1.1 kills all mechanical price-based exits per backtest evidence (buy-and-hold beat all 3 variants by 19-91pp on the real-anchor rerun; v1's 90d-proxy spec underperformed by 135-192pp on the 2019-2025 window). Drawdowns become buying signals when fundamentals confirm. Adds SANDISK + INTC per 2026 Q1 smart-money entries by Druckenmiller / Tepper / Tiger Global. Tranche-accumulate MU + SMH + SNDK + INTC, exit only on fundamental anchor break or 2+ smart-money manager exit. Initial deploy = 2R ($10k) split 0.5R each. Paper only.

Why v1.1 exists (the empirical follow-through)

Two backtest reports converged on the same finding:

Diagnosis: the tranche-accumulation logic is value-additive; the mechanical exit logic is the entire source of underperformance. Exiting a multi-year cyclical at a 15% drawdown / 2-consecutive-negative-quarter signal IS selling the bottom by construction.

Corroboration: 2026 Q1 Druckenmiller + Tepper + Tiger Global all opened SANDISK + INTEL positions in the same quarter as the 3-manager Amazon add (per smart-money 2yr backfill summary). Three independent positive-sum managers crossing a structural-memory line at the same time the thesis says we're mid-cycle — exactly the kind of corroborating signal v1's mechanical exits would have steamrolled.

Bull case vs revision risk (the explicit tension)

The bull-case argument for v1.1: the thesis (HBM supply structurally short through 2027, all 2026 capacity pre-sold, MU Q2 fiscal 2026 +196% YoY at 81% gross margins) is empirically validated by Q4 2025 - Q1 2026 data. The backtest shows holding MU through the 2022 cyclical drawdown returned +302% over the next 3 years. Two independent smart-money managers opened SNDK and INTC positions in 2026 Q1, which is the closest thing to external confirmation we get without insider data. The Druckenmiller doctrine — find the great opportunity, bet heavily, let winners run — is the trading discipline matched to a multi-year structural cycle.

The revision risk: v1.1 removes the safety net. No price-based stops, no 15% drawdown exit, no anchor-flip auto-trim. Position size cap goes UP (2R → 3R per name). Drawdowns trigger ADDS, not exits. If the bear case is right and 2026 IS the cycle peak (like 2017 — fully sold out, gross margins peaking, capex committed), then v1.1 will compound losses through the rollover phase before founder review fires the exit. The single fact that would force capitulation per the thesis (hyperscaler 3+ year HBM offtake at fixed price) hasn't happened yet. Smart-money signal is fresh (13F filings <60 days old) and could reverse next quarter. The 4-anchor founder-review gate replaces 5 mechanical exit rules — manual review IS slower than autonomous exit.

The picks: APPROVE locks in the Druckenmiller-doctrine paper-trade shape with the expanded basket. REVISE walks back to a hybrid (e.g. keep tranche entries + a much wider mechanical exit at -30% / 6-month return). ARCHIVE concedes the thesis itself is wrong (not just the strategy). DEFER waits for the next quarter's smart-money 13F to confirm or reverse SNDK + INTC adds before deploying.

Spec changes vs v1

Changev1v1.1
Mechanical price-based exits90d return < -15% phase-marker exitKILLED — none
Tranche-entry logic0.5R/0.5R/1R at -0/-5/-10%Unchanged (value-additive per backtest)
Anchor-based position ADDNoneNEW — drawdowns >20% with bullish anchors = +0.5R add (Druckenmiller doctrine)
Max per name2R ($10k)3R ($15k)
Bucket cap4R ($20k)4R ($20k) unchanged
Ticker universeMU + SMHMU + SMH + SNDK + INTC (SNDK + INTC per 2026 Q1 smart-money entries)
Exit trigger2+ phase markers flip bearish for 2 consecutive quartersSingle anchor flip = founder review via /decisions/; 2+ HIGH severity anchors confirmed = bucket close
Profit trims+50% trim 0.5R; +100% trim anotherRETIRED — let winners run
Quarterly review3-anchor pull + founder surface on phase change4-anchor pull (adds smart-money 13F deltas) + founder surface on any anchor flip

Deploy parameters

ParamValue
ModePaper only (Alpaca paper sandbox)
1R unit$5,000
Initial deploy2.0R total — MU 0.5R + SMH 0.5R + SNDK 0.5R + INTC 0.5R
Max per name3R ($15k) [raised from v1's 2R]
Max bucket (memory total)4R ($20k)
Tranche 2 trigger-5% from t1 avg OR thesis-confirming catalyst
Tranche 3 trigger-10% from t1 avg OR earnings catalyst
Anchor-add triggerCapex revised UP >5% OR DRAM +10% QoQ for 2Q OR HBM cadence accelerating OR new 2+ smart-money manager open OR >20% drawdown with all anchors bullish
Per-trade stopNone (thesis-stop, not price-stop)
Profit trimsNone — let winners run (Druckenmiller doctrine)
Exit trigger (single signal)Founder review via /decisions/ page within 7 days
Exit trigger (confirmation)2+ HIGH severity anchors flip for 1-2 quarters = close ENTIRE bucket
Kill switchFounder channel "halt memory" or "pause investing" = immediate close
Review cadenceQuarterly (Ray runs, files to vault, surfaces to founder only on anchor flip)

Lock in

Approve — deploy paper per v1.1 locked parameters

Ray fires initial 2R tranche (MU 0.5R + SMH 0.5R + SNDK 0.5R + INTC 0.5R = $10k paper) into Alpaca. Logs to positions/. Sets up quarterly 4-anchor watch. v1 deploy gate auto-archives. Use the field for any deploy notes.

Approve + send
Revise — walk back specific parameters

v1.1 is close but a parameter needs adjusting (e.g. keep a wider stop, drop INTC, lower per-name cap back to 2R, smaller initial deploy). Name what; Ray re-files v1.2.

Revise + send
Archive — thesis does not deploy (concede bear case or thesis itself)

If the thesis itself doesn't pencil — not just the strategy. Ray files as status: archived-pre-deploy with reason. No paper trade. Memory bucket closes.

Archive + send
Defer — wait for next-quarter signal confirmation

Hold for next quarterly smart-money 13F (re-confirm SNDK + INTC positions held/added) OR named catalyst (MU Q3 fiscal 2026 earnings) before deploying.

Defer + send