190 market and economic benchmarks are live · See them
Belief isn’t evidence
Build a book no single bet can sink.
Every desk has more ideas than it can trust. PIER20 is the machine you run yours through. Write the strategy once, grid every loss before capital moves, check it against the bench for overlap, and take the survivors live with the evidence attached.
Alpha decays, and a desk with few uncorrelated sleeves takes its drawdowns all at once. Outrunning that takes three things working together: the throughput to test many ideas, gates strict enough to kill the ones that would fool you, and a bench of survivors that don’t sink together.
Many enter. Few earn capital.I · The fieldII · The gateIII · The sleevesKilled at the gate: its live path, dashedThe sum of the survivors: one steady line
Maintain your edge · Velocity
Replace what decays before the book feels it.
It is the nature of alpha to decay. No edge is permanent. Ideas were never the bottleneck. Testing them quickly and reliably is. The field refills only as fast as you can qualify.
Know your losses upfront · Validation
Don’t let shiny backtests flatter you.
A backtest is one pass through one history. Out-of-sample runs, Monte Carlo paths, and formal checks find the defects a backtest cannot. Each finding closes a loop: find the flaw, fix it, run again. Risk falls with every pass.
Never sink together · Variety
Catch the bets that look different but aren’t.
Two sleeves can share no code and still share every drawdown. Screen each survivor against the bench for overlap before it joins, and watch for pairs that start moving together. The bench counts exposures, not strategies.
02 · The engine
Test more strategies and go live without code rewrites.
The loop from idea to live has two speed limits: machine time, and every rewrite between phases. PIER20 removes both.
One artifact carries a strategy through every gate.
momo_14d.v3
strategy momo_14d {
universe: SPX_500
signal: zscore(ret_20d) > 1.4
size: vol_target(8%)
gate: V3 { costs, oos, monte_carlo, overlap }
}
>> backtest · full history · costs on ✓ 3.2s
>> sim: oos + monte-carlo ✓ PASS
>> paper: on the tape · no capital ● RUNNING
Write onceRun through every gate
Nothing lost in translation · One artifact
Stop rewriting strategies between phases.
Most desks write a strategy twice: once for research, once for production. The rewrite costs weeks and adds defects the backtest never saw. In PIER20 notation, one artifact runs backtest, simulation, paper, and live. What you test is what you deploy.
Answers while the idea is fresh · Throughput
Never wait on the machine.
A full-history backtest with costs on returns in seconds, not overnight. Simulation and paper run on the same engine, against the same artifact. Live execution answers in under twenty nanoseconds. The queue is never the reason a strategy waits.
50M ticks/sBacktest throughput
<20nsLive execution
1 artifactBacktest to live
03 · The gates
Filter out strategies before they can cost you in the market.
Qualification is a sequence of gates, stricter and smarter as they go. Their job is to surface every way a strategy can lose money while losing money still costs nothing.
Qualification determines what belief cannot.
Survival through the gatesper 100 strategies · illustrative
Gate one · Backtest
Has it worked before?
A backtest answers one question: did this make money in the past? A pass earns the strategy a harder test, and nothing else. Past performance qualifies. It does not predict.
Gate two · Simulation
Does it survive markets that never happened?
Out-of-sample and Monte Carlo runs put the strategy through conditions no history contains. Formal checks go further: when models author strategies, some learn to game the test instead of the market. The gate catches both.
Gate three · Paper
Does it behave in the real market?
Paper trades the live market without capital. This is first contact with real water. This is where behavior drift shows up. If the live pattern does not match the simulated one, the strategy goes back.
04 · The archive
Maintain a central corpus of your priors, so mistakes never repeat.
Every trial adds to the corpus: the strategy, the run, the verdict. Over time it becomes a map of the water: where edges held, where they sank, and which bets the book already carries.
Every trial becomes institutional memory.
Strategy
Gate reached
Verdict
What killed it
Overlap vs bench
fx_carry_09
Live
Live
—
low
crowd_rev_4
Live
Live
—
flagged · moves with fx_carry_09
momo_14d
Paper
Simulation testing
—
low
idx_lead_7
Paper
Killed
live drift from sim
—
vol_skew_2
Simulation
Killed
costs ate the signal
—
mr_pairs_51
Backtest
Killed
failed the full history
—
gold_mom_3
—
Killed
already tried · run 0112
—
Nothing tested twice · The record
Has this been tried?
The archive keeps every run: the strategy, the verdict, and what killed it. New ideas get checked against the record before they cost a loop. The memory stays when analysts move on.
Edge cases on file · The sinkholes
Where do strategies sink?
Every kill records the condition that broke the strategy. Those edge cases accumulate into a map of where losses live: crowded exits, regime turns, costs that eat the signal. A new idea can be read against that map before it enters the water.
No hidden overlap · The bench
Does the book already hold this bet?
Before a strategy joins the book, the archive compares its exposure with every live sleeve and flags the pairs that move together. The check repeats while they run, because correlation can arrive later.
05 · Reports
Essays with the working shown
Long-form research from the desk: methods, data, and the parts that didn't work. Written to be checked, not skimmed.
Each series comes from a primary source and updates on its own cadence. Every chart shows its formula, its source series, and its latest observation date.
One artifact runs backtest, simulation, and live. No rewrites, no handoffs, no translation gaps. What you test is what you deploy.
The engine backtests at 50 million ticks per second. Live execution answers in under 20 nanoseconds.
You describe your strategy and guide the AI. It writes the code, runs the tests, and shows where the strategy fails and what to improve. The archive keeps the record of every run.
Every strategy passes simulation before it goes live. Out-of-sample and Monte Carlo runs find the drawdowns a backtest cannot. A more resilient strategy makes the return path smoother.
Is my strategy IP safe?
Strategies stay inside your deployment. The notation, results, and archive are yours; nothing leaves your environment, and self-hosted installs keep the whole loop on your metal.
Can I self-host the engine?
Yes. The engine, archive, and notation tooling ship as a self-hosted stack; the managed version exists for desks that don't want to run infrastructure.
What does the AI actually do?
Your own LLM works across the notation, engine, results, and archive: writing strategies, debugging finance-native errors, reviewing findings, generating variants, and suggesting the next improvement.
How fast is onboarding?
One strategy through PIER20 in your first session. No platform migration, no portfolio migration. Bring one tested idea and run it.
Final call
Run one idea in minutes.
Put one tested idea or thesis through PIER20 and watch it become a finance-native notation, run through the evidence stack, and come back with real feedback. No migration of any kind.