Proof Dossier
This is Aether's evidence locker, on public display. Sealed replay receipts, per-card evidence with the drawdowns left in, and the gates every capability must earn its way through.
Most performance pages are curated. This one is sealed.
Every number below was graded by machinery that cannot be talked out of a verdict — built by the same team it polices.
The Second Product
AI produces work that looks finished when it isn't — polished code, confident calls, green dashboards over hollow answers. We caught our own AI agents doing exactly that, and built the machine that catches it. Every artifact now travels through this pipeline. The oracle is its first customer.
The rest of this page runs the system's own numbers through that pipeline — including the one it holds back.
The current measure
What "proven" costs here
A cell earns the proven grade only when its statistical floor — the pessimistic end of its confidence interval — clears 50% with positive expectancy after fees, on accounting that grades every ambiguous bar against us. Cards that trade selectively publish smaller samples and wait longer for the grade. The system does not round up for anyone.
Minted through the upgraded engines · Books sealed 2026-08-31 (True Edge) and 2026-09-02 (LHCS / Curve, MODEL, THE VOICE) · Dollar figures re-stated 2026-10-05 under the settle-at-horizon-end rule · engine commit stamped on every card · Full per-trade receipts in the Evidence Console
The vintage receipts
These four books are a retired measure from an earlier engine generation, replayed with a $100,000 research account per card. They stay on this page because deleting old evidence is how numbers get laundered. They are not the current measure and may not be cited as this system's capability: every number below has been superseded by the current sealed books in the Evidence Console, and in some cases the current number is lower, in others higher.
Superseded-vintage disclosure. These books measure the engines as they stood at that run. A replay grades the engine that produced it — nothing newer — so these numbers are provenance, not capability. The current sealed measure (a weekly-cadence walk of roughly thirteen years; True Edge's book sealed 2026-08-31, the other three re-sealed 2026-09-02; dollar figures re-stated 2026-10-05 under the settle-at-horizon-end rule — each trade sized when it opens from settled equity only, its result booked at its horizon end; per-trade receipts, book digests and engine commit on every card) is live in the Evidence Console. Where the two disagree, the console is the answer. Side by side, current versus retired: LHCS / Curve +198.4% dynamic (retired panel: +268.82%) · MODEL +179.9% dynamic (retired: +31.82%) · True Edge 615.2 R over 520 graded decisions in 13.3 years — one roughly every nine days, from 664 decision points, on 144 of which the card declined to call · 102.2% CAGR dynamic · −42.1% max drawdown · 59.7% win rate (retired: +118.85%; flat-1x baseline +947.1% / −9.8%). Drawdowns move too, in both directions — LHCS's true drawdown is deeper than the retired book showed (−35.6% vs −11.83%) and True Edge's is shallower (−42.1% vs −69.47%). The 2026-10-05 re-statement itself lowered True Edge's dynamic CAGR from 136% to 102.2% — old and new side by side in Ask the Record. We publish the corrections that hurt us in the same breath as the ones that help.
Notice what this table does that marketing pages do not: it publishes the weakest card, the smallest sample, and the ugliest drawdown in the same type size as the best number. Mixed results publish as mixed. That is the truth contract working.
Historical research replay · No client funds, no custody, no trading authority · Founder's own micro-scale execution tests disclosed · No investment-product claim. First order 0.00005 BTC ≈ $3.95; largest a full three-tier bracket ≈ $11.98; total fees ≈ $0.25. LYFE has not taken any oracle calls.The Blocked Headline
True Edge's sealed book is the biggest number this system owns — 615.2 R over 520 graded decisions in 13.3 years, 9 of its 10 evidence cells graded proven. It is still not the headline claim, because the gates that would license one have not all passed: live leverage waits on its proof contract, broker execution waits on legal and evidentiary gates, and every authority flag in the sealed artifact — trading, capital, deployment, model promotion, broker execution — reads false. On the founding card, the block is even more explicit: LHCS / Curve has no proven cell on its long side, where most of its return comes from (its 2 proven cells are short-side, on 6 and 35 trades), its three-year geometry comparison came back scored, not promoted (−0.10 R paired delta, 2026-08-08, with a named owner and a scheduled re-hearing), and until the 2026-08-31 depth restoration its call source abstained across 2013-2016 behind a data-depth wall rather than guessing — its book now spans 13.25 years, at parity with its siblings.
Why this matters
Anyone can publish their best backtest. A system that can refuse its own best number — name the exact gate, date it, assign it an owner, and schedule its re-hearing in public — is a system whose other numbers mean something. Every refusal on this page is checkable in the Evidence Console: open any card, read the grade on every cell, and hash the book yourself.
The Builder
Aether began as hand-charted market patterns drawn by a concrete foreman — two decades of building things that are not allowed to fail, brought to software. AI agents became the workforce. The thesis, the taste, and the refusal to sign off on unverified work came from the builder.
The Capability Gates
The same pipeline that grades a prediction gates what the system is allowed to become. Each stage is real, staged work with published criteria — none of it activates on confidence alone.
Provenance