BALLAST.
All-weather portfolios, run in the open

A paper fund that publishes its kills.

Ballast runs five all-weather paper books, priced every trading day by an AI operator and committed to git as they happen. The research program behind them answers to a pre-registered statistical gate, almost everything we have registered has died at it, and the whole score is public including the survivors we decided not to trade. What we can honestly claim today is disciplined risk premia and an operation you can audit. Nothing more is claimed until the gate and the forward record earn it.

Ballast is an all-weather portfolio with an AI operations team. It holds a risk-balanced book designed so no single market regime can capsize it, prices itself every day, and explains every dollar of gain or loss in plain English. Its entire history is public and timestamped from day one, so you never have to take our word for anything.

The score we keep against ourselves

Strategy variants gated
47
Hypotheses killed
8
Forward experiments
1
Citable edges
0

Every strategy idea we test is pre-registered with its thresholds before the result exists, then scored by a deflated Sharpe gate that corrects for how many things we tried. The gate is calibrated against pure noise: in its latest full null battery it shipped 0 of 100 strategies that had no edge by construction. It has killed 8 of the 9 hypotheses pre-registered and published so far, including one it first shipped and we later took back in public. The one that cleared it is held off by default, with the findings that stop it being a tradeable claim printed in the same row as the result. And it cuts the other way too: our own factor regression says the aggressive book's backtested return is mostly plain market exposure (market beta 0.51, R squared 0.62), which we print rather than bury. That is the product: an operation that will not let us lie to ourselves about our own results.

Read the kill log →   Check the record yourself →

Three books, one architecture

Same all-weather engine, three risk levels. Pick the temperament that fits you. These are model portfolios you can follow, not accounts we manage, and every figure below is paper.

Ballast Aggressive Paper

Full throttle, still all weather.
Largest: Long duration 20%, Cash and bills 20%
Since inception
+4.96%
NAV
$104,963
Posture: 1.49x gross, 18% vol target
See the full book →

Ballast Balanced Paper

The whole map, no borrowing.
Largest: Cash and bills 27%, Long duration 22%
Since inception
+1.52%
NAV
$101,522
Posture: unlevered, 10% vol target
See the full book →

Ballast Conservative Paper

Ballast first, growth second.
Largest: Cash and bills 37%, Long duration 25%
Since inception
+1.15%
NAV
$101,148
Posture: unlevered, 6% vol target
See the full book →

Forward experiments, unproven

These books start from the aggressive all-weather book, then add strategy overlays chosen by a backtest search. They run forward in the open like the rest, but the honest re scoring of that search (see each book's honesty panel) found the overlays' edge does not clearly survive deflation and out of sample testing. Read them as an experiment, not a proven upgrade.

Ballast Systematic Aggressive Paper Forward experiment, unproven

Forward experiment, unproven.
Largest: Long duration 20%, Cash and bills 20%
Since inception
+5.31%
NAV
$105,309
Posture: 1.49x gross, 18% vol target
See the full book →

Ballast Short Vol Watch Paper Forward experiment, unproven

Forward experiment, unproven.
Largest: Long duration 20%, Cash and bills 19%
Since inception
+5.44%
NAV
$105,437
Posture: 1.49x gross, 18% vol target
See the full book →

New, in incubation

This book cleared our pre registered statistical gate, which is rarer here than it sounds: most of what we test we kill in public. Clearing the gate does not make it proven. Its forward record starts at inception and is nowhere near long enough to say anything, we did not backfill a day of it, the forward book is referenced to a different venue than the one the gate scored, the yield it harvests is documented as decaying, and the venue solvency risk a real position would carry is invisible to every statistic we ran. It runs forward on paper so the record can answer what the backtest cannot. Where it sits on the ladder →

Ballast Crypto Basis Carry Paper Forward experiment, unproven

New in incubation. Young record, honest limits.
Largest: Crypto basis carry 100%
Since inception
+0.55%
NAV
$100,547
Posture: 2.42x gross, no vol target
See the full book →

Why this is different

A record you can check
The NAV history is append-only, source-tagged, and committed to git as it happens. Retail finance is drowning in backtests. This is a live daily record with a paper trail anyone can audit, down to recomputing a single day's NAV by hand. Try it →
A kill log, not a highlight reel
Every strategy hypothesis is pre-registered before the result exists and every verdict is published, including the one we reversed on ourselves when we tightened our own gate. The kill log →
A research ledger you can hash
The pre-registration ledger is hash-chained, so editing, reordering, or deleting any past entry breaks a committed head you can recompute yourself, and the head is staged for external timestamping. How to verify it →
We tell you how much is just beta
Our own regression on the aggressive book's backtest puts market beta at 0.51 with an R squared of 0.62. Most of the return is compensated market risk, and we say so before you ask. The numbers →
We know where it stops scaling
A square-root market-impact model puts the aggressive strategy's honest ceiling near $56 million. The limit is how much its least liquid instrument trades in a day, not edge erosion, and that distinction is a claim an allocator can check. The capacity table →
An AI operator that shows its work
Ballast is constructed, marked, monitored, and rebalanced by an AI agent system with a human overseeing risk. We do not claim to predict markets. We run the operation and publish all of it.

Read how the portfolios are built →

Our gate, pointed at your backtest

The statistical machinery that kills our own ideas also runs as a standalone audit for outside strategies. Send a daily returns series plus the variants you actually tried, and it returns a certificate: the Sharpe that survives after your trial count is corrected for, walk-forward out of sample performance, the probability of backtest overfitting, and the smallest edge your sample length could even detect. The verdict applies the identical significance floor we hold our own research to.

What the overfitting audit covers →

Free to follow, more for subscribers

The NAV curves, the monthly summary, and the verification page are free. The full weekly letter, the exact current weights, rebalance issues, and alerts go to subscribers on a regular cadence. Same content for everyone, no book picked for you.

Free: NAV curves, the monthly summary, and the verification page. No spam, and nothing here is personalized advice.

See subscription plans →

Paper, and labeled paper everywhere. Ballast holds no real money. It is a published model, not a managed account, and it is not investment advice. That posture is deliberate and it is explained in full on the disclaimers page.