About

Latent Lines is the publication of an experiment. It is one person reading the world, building tools, holding positions, and showing the work.

Who is writing

I am Mike Singer. I studied philosophy. I am not a finance professional and have no formal investment-management training.

I have been building Time Detective's, a project to bridge entertainment and history. Five years ago I began building NewsPlanetAI -- an apparatus that reads several hundred news articles per day and surfaces the slow geopolitical patterns underneath the daily noise. Both tools were built first for myself, then for people I have shared them with.

This publication exists because I wanted to test a question the financial press largely answers in the negative: can a layman with very little capital, using accessible AI tools and a patient method, grow a portfolio by reading the world rather than reading the market? Five years and a series of conviction trades later, the answer is in progress -- which is why the portfolio is public.

The method, briefly

NewsPlanetAI surfaces patterns. Per-ticker wikis carry the structural facts -- regulatory filings, contract awards, who pays whom. Per-counterparty wikis (the Department of War, the NRC, the Space Development Agency, China's MOFCOM) map which holdings move together when one of those nodes shifts. A set of hard rules governs how positions are sized, trimmed, and added to. Personal intuition, scuttlebutt, primary-source reading, and a philosophy of waiting fill the rest.

The weekly brief is the synthesis in motion. It is what comes out of the apparatus on any given week.

How performance is measured

Two numbers, both shown on the hero block:

The full math, including the convention used to separate price-driven returns from contribution-driven moves, is documented in the project repository.

How we benchmark

The chart on the landing page can overlay two composite baskets of ETFs alongside the portfolio line. Each basket answers a different question.

CMU-Broad -- "would passive indexing have beaten me?"

A weighted basket of broad US market ETFs. If the portfolio underperforms this line, the right move would have been to buy the index and go fishing.

ETF Weight What it tracks
SPY 40% S&P 500 -- the broadest opportunity-cost baseline
IWM 25% Russell 2000 -- small-cap, where uranium, copper, and rare-earth names live
DIA 15% Dow Jones -- the industrial / old-economy signal
QQQ 10% NASDAQ-100 -- the tech and liquidity-regime signal
SGOV 10% Short-term T-Bills -- the cash floor, the truly passive alternative

TMU-Thesis -- "am I beating the sector universe my thesis claims to understand?"

A weighted basket tilted toward the sectors the portfolio actually expresses convictions in. If the portfolio underperforms this line, a dumb sector-ETF basket would have done a better job than the active picks. This is the honest signal for whether the thesis is doing work.

ETF Weight What it tracks
IWM 30% Small-cap base -- most positions live here
SPY 20% Broad US equity
XLE 15% Energy -- the uranium / energy-sovereignty thesis
XME 15% Metals and mining -- copper, rare-earth, silver
ITA 10% Aerospace and defense -- orbital infrastructure
QQQ 10% Tech / liquidity regime

The formula in both cases is the same. For each ETF i in the basket, the cumulative ratio from the anchor date t₀ to time t is ratio_i(t) = close_i(t) / close_i(t0). The composite is the weighted sum: composite(t) = Σ weight_i × ratio_i(t). Converting to a percentage gives the line you see on the chart. Adjusted closing prices are used throughout, which incorporate dividends and splits -- the correct practical proxy for total-return when index total-return series aren't directly available.

A note on what's reconstructed. The live CMU system anchors to May 11, 2026 -- the day the framework was instrumented. The benchmark overlay on the chart extends the same composite math backward to October 31, 2023 so that "since active trading" has a benchmark to compare against. Pre-May-2026 values are reconstructed from ETF closes for that purpose; they are not retroactive instrument readings. The live system, and the live CMU unit ratio shown in the hero block, remain unchanged by this.

The full spec, including the design rationale for separating CMU-Broad from TMU-Thesis (and the deliberate exclusion of XLE / XME / ITA from CMU-Broad to keep the passive baseline thesis-free), is in project/docs/CMU_SPEC_v01.md in the project repository.

Skin in the game

Every position discussed in the brief is one I hold. The full holdings list is at /disclosure. I do not accept paid promotion of any security. I do not trade in the 48 hours surrounding publication of any post that names a specific ticker.

Cadence

Weekly minimum, irregular preferred. The publication's tempo is the synthesis's tempo, not the market's.