Methodology & disclosures

Plain-English construction of the two published books. Simulated. Not live AUM. You execute.

What Glacifraga is

Glacifraga is a signal sleeve / white-label product of Thin Ice Digital Ltd. It is not a fund. Thin Ice Digital Ltd does not hold client money. Live client AUM is $0. The numbers on the homepage are a simulated $100k book, not a live track record. You fire the orders in your own account.

Client copy is two books only: Obsidian (conservative) and Aurora (higher risk / higher return). Do not mix constructions.

Two books

Obsidian is the conservative sleeve: equities and commodity ETFs, no crypto. New entries pause when SPY and TLT are both below their own 200-day average. Existing names keep trailing. Full ATR stop and trail.

Aurora is the higher-risk sleeve: the same conservative book plus crypto and additional names that fire on rules. Trail-only ATR. Aurora does not use the Obsidian pause. Crypto is sized in fractions of a coin.

Each book is its own $100k simulation. Adding Aurora’s CAGR to Obsidian’s, or blending stops from one book with names from the other, is not a published product.

Universe

The sleeve was locked in July 2026 at 47 names. Frozen means the live sleeve does not silently add tickers.

Survivorship: the list was chosen as of that freeze, then applied historically from 2016. This is not a 2016-vintage universe that was locked before the decade ran. The honest story is the freeze date versus the backtest window. Names that listed late are silent until tape and an SMA200 exist — they are not back-filled as if they had been tradable in 2016.

Book construction

Headline stats are one $100k shared book, not a sum of independent sleeves. Cash is shared. The max-6 new-entries-per-day cap is shared. When six names signal on the same open, they compete for slots and for cash. That is the product claim.

Next-open fills
1% of book equity at the stop
~10% notional cap per name
Max 6 new names per day
Volume must be 1.4× the 20-day average
ATR stop and trail (Obsidian: full stop + trail; Aurora: trail-only)
Unstated costs are zero: next-open, no slippage, no commission, idle cash earns 0

Why sleeve-sum is not the headline

A Tearsheet-style run that gives every name its own $100k sleeve, then adds the equity paths, does not pay the cash or the six-entry cap. That construction can print a much hotter CAGR (the ~31% neighbourhood). It is research, not the book a client would run. The homepage quotes the shared $100k book. A 2023 Obsidian calendar +31.0% is one year on that shared book, not a licence to market 31% as the product.

Windows

Both published books use 2016-01-01 → 2026-09-10. Research split: in-sample through 2021-12-31, out-of-sample from 2022-01-01. Indicators warm up from 2015-01-01 so SMA200 is live at the 2016 start.

OOS drawdown is peak-to-trough inside the OOS equity segment. Obsidian’s full-window −17.5% trough (13 Mar 2023) is inside the OOS calendar span, but the peak that starts that drawdown is 8 Nov 2021 (in-sample).

Regime rule

On Obsidian only: pause new risk when SPY and TLT are both below their own 200-day average on that session’s close. Open names keep trailing. The book does not flatten on the regime gate alone.

The SPY∧TLT SMA200 pause is applied from the 2016 open of the published sim (23 in-sample pause days before 2022-01-01); it is not switched on at the OOS boundary.

COVID 2020 stays off (0 both-below days / 253 sessions). 2022 turns on for 204 / 251 sessions (~81%) and Obsidian’s calendar year is still −10.7%. That is insurance for a stocks-and-bonds selloff. It is not a free 2022 coupon, and it does not buy a ≤10% Full DD product. Full-path max DD is −17.5% (13 Mar 2023).

VIX only scales signal confidence. It is not this hard pause.

Simulated results

These are simulated paths on Yahoo-adjusted daily bars. They are not live AUM. Past simulated results are not indicative of future results. First live track record is intended to start Q4 2026.

Glossary

Locked universe

The July 2026 lock for Obsidian: 47 names. No silent adds.

Shared book

One $100k account. Cash and the six-entry cap are competed for. Not sleeve-sum.

ITD CAGR

Inception-to-date compound annual growth from the first book bar through that year-end, using 252-day years.

OOS

Out-of-sample: 2022-01-01 through the last published bar (2026-09-10). In-sample ends 2021-12-31.

BARON

Internal name of the live Obsidian worker process. Not a third client product. Homepage copy uses Obsidian.

Quant appendix

	Obsidian	Aurora
Window	2016-01-01 → 2026-09-10	2016-01-01 → 2026-09-10
Full CAGR	+20.2%	+24.6%
Full max DD	−17.5% (13 Mar 2023)	−19.9% (26 Sep 2022)
OOS max DD	−9.3% (8 Sep 2026)	−15.5% (10 Sep 2026)
Terminal on $100k	$711,392	$1,043,127
Trades	896	1036
Regime pause	SPY and TLT both < SMA200	Off — trail-only

Obsidian Sharpe 1.44 · PF 2.01. Aurora OOS Sharpe 1.65 · PF 1.97 · 40 BTC · 21 ETH. Long-run Obsidian claim is high-teens / low-20s CAGR — not 31%, and not a ≤10% Full DD product. Simulated · not live AUM.