Journal

One change to how the options position is funded. An options position that expires on the same day a new one would be bought still counts as money tied up when the fund checks whether the parked share can pay for the new one. Its proceeds are not used to pay for a purchase on the same day, because in the account the fund uses they are not reliably available that soon. The practical effect is that, when Bitcoin already takes up most of the fund, a new options purchase is skipped a little more often. In the historical record this skips three purchases in early 2023 and lowers the simulated return with options slightly; the figures without options do not change. The rule for when to buy, the sizes and the cap are as they were. It moves no position today and changes nothing the fund holds.

Reporting change; no rule changes. Two things. First, the fund now publishes two reference dials beside the one it follows: the dial as it would read under the rule in force before the last change, and the dial with only the four-year-average exception and no read-ahead. They are yardsticks: nothing is bought or sold on them. The read-ahead will be judged against the second one, the simpler rule it has to beat, and the yearly review reports both. Second, the wording. The read-ahead is a momentum adjustment on the fund's liquidity reading, chosen because it improved the historical record. The observation that Bitcoin tends to follow liquidity with a lag of a few months is what prompted it; the adjustment is not calculated from that lag, and earlier wording that suggested so has been corrected. Every rule, band, cap and figure the fund runs on is exactly as it was and nothing trades differently.

Two bounded changes to how the dial reads liquidity. First, when global liquidity has improved over the last three months, part of that improvement is read ahead. This is a momentum adjustment chosen from the historical record; the observation that Bitcoin tends to follow liquidity with a lag of a few months is what prompted it, not a formula it follows from. A deterioration is read as before, so the dial is never below what the old rule would say. Second, while Bitcoin's price closes below its four-year average, the liquidity haircut is not applied: at those levels the record shows the haircut costs the recovery more than it saves, and it comes back the day price closes above the average. Nothing else moves: the valuation and rotation factors, the executed dial's memory, the floor, the bands, the parking rule, the options position and every override are as they were. The rule was re-backtested over the full history before adoption and judged on return per unit of drawdown; it improves the record on every window and passes the after-tax check. It is not free of cost: in the past it would have held a material Bitcoin position through the November 2022 exchange collapse, which cost about twelve percent of the fund's value in a fortnight before the 2023 recovery more than repaid it. It makes no trade today: at the current close the new and old readings are the same number.

The fund now accounts for when Federal Reserve balance-sheet information becomes available. Historical simulations also use earlier information for the valuation reference and include ETF distributions consistently. Actual past decisions remain the fund's record. The allocation effect was checked against the production state; future rebalances follow the existing execution rules.

The fund retains selling patience while liquidity supports buyers, reducing that patience as valuation becomes extreme. Lower volatility alone no longer ends the delay sooner. The valuation formula and the options rules are unchanged.

Framework change: a small options position joins the strategy. When the strategy calls for it, the fund buys convexity — bought options only, which gain on a large move in Bitcoin and can lose no more than they cost. When and how much is set by rule, not by judgement; the rule itself is not published. The position is paid for from the parked share, capped at a fixed share of the fund, held to expiry, and counted at cost in the fund's values until it settles. It never changes the Bitcoin share: the dial, the bands, the floor, the parking rule and every override are as they were. The rule was re-backtested over the full history before adoption. Its backtest is priced on a different venue from the one it trades on and rests on a few large payouts, so live results will confirm or revise it. The signal was off at adoption, so no position follows from the adoption itself. Filed with NAV below its 2025-Q4 high-water mark and audited as such at the annual review.

Framework change in the valuation factor. The two cheapness readings the dial averages — how far the market sits above its cost basis, and price against the power-law floor — were calibrated in an era when Bitcoin moved far more violently than it does today. In a calmer regime the same readings reach the "expensive" end of their scales less often, so the dial has been holding more Bitcoin near recent tops than the rules intended. From today both readings are stretched by a single factor derived from Bitcoin's own trailing one- and two-year realised volatility: the factor is exactly one whenever volatility is at or above the calibration era's level, so the rule does nothing in a wild market, and it grows as the market calms. The same stretched reading also governs where the executed dial's memory of a cut fades. Nothing else moves — the liquidity and rotation factors, the executed dial's timing, the floor, the bands, the parking rule, the options overlay and every override are as they were. The rule was re-backtested over the full history before adoption; it improves the post-2022 record and lowers the modern-era drawdown. It is NOT inert: at adoption the executed target sits about four points of NAV below the weight the fund holds, so the first close after deployment will sell about four percent of NAV. That trade is the decided consequence of the change and will be journaled as an ordinary rebalance. Filed as a mid-drawdown change — NAV below its 2025-Q4 high-water mark, R* 0.50 at decision — and audited as such at the annual review.

Documentation change; no rule changes. The written framework the fund follows has been reorganised: the page now states only the current rules, while the history of how each rule was adopted, the record of retired rules, the sequence of published backtest figures and the running list of open questions with their evidence moved, unchanged, to two companion pages. Every rule, band, cap, budget, feed and figure the fund runs on is exactly as it was; nothing trades differently and no number of record moves. The vintage is bumped only so the site's framework badge keeps matching the page it cites.

Reporting and disclosure change; no rule changes. The fund's NAV of record carries any open options position at the premium paid until it settles. From today the site also shows a companion series beside it that values open positions at their intrinsic worth — what they would pay if settled at the day's Bitcoin close, with no allowance for time value — together with its own drawdown. It is a reporting line only: every rule, band, cap and budget keeps reading the at-cost NAV, and nothing trades off the companion. The monthly pass additionally prints the fund's exposure to a single sharp Bitcoin fall — the held weight times a reference two-day and one-month shock — and the share of Bitcoin's fall the fund actually took in the latest fast crash, as information lines with no threshold behind them. The approach page now states plainly that a crypto-only crash is an accepted risk the rules buy into by design, covered only conditionally by the options overlay.

One term of the executed dial R* changes. While the liquidity trend is up, a cut in the formula R has been honoured gradually only below a valuation ceiling and taken at once above it. From today that ceiling is a fade rather than a switch: the memory is carried in full below the line the rules already used, shrinks smoothly across a band above it, and is gone at the top of that band, where the valuation factor itself reads zero. Nothing else moves — the entry side, the bands, the floor, the caps, the overlay and every override are as they were, and R* still never sits above the formula's own prior reading. The rule is inert today (valuation far below the band) and identical to the previous definition on every day the band was not reached; it was re-backtested over the full history before adoption, where it changes three past years and leaves every other year to the decimal. Filed as a mid-drawdown change — NAV below its 2025-Q4 high-water mark, R* 0.54 at decision — and audited as such at the annual review.

Framework change in two parts. First, the executed dial R* — the formula R carried with a short memory — now works in both directions: under conditions the rules define, a rise in R is honoured gradually just as a cut already was, so R* can sit below R for a while as well as above it. Nothing raises the fund's Bitcoin exposure above the formula, and every risk override still ranks above R*. Second, at a month-end parking evaluation where Bitcoin and growth equities have both been falling together, the parking leg may not choose the growth-equity destinations — while both sell off they stop diversifying the Bitcoin leg — and picks among the others instead. Both rules were re-backtested over the full history before adoption. The dial rule applies from today and currently holds R* slightly below R, inside the rebalance band, so no trade follows from the adoption. The parking rule first applies at the September month-end; it would have applied at the July month-end, but it was not in force then, so August's parking stands as recorded. Filed as a mid-drawdown change — NAV below its 2025-Q4 high-water mark, R* 0.54 at decision — and audited as such at the annual review.

Framework change: the fund now executes R*, the formula R carried with a short memory. Under conditions the rules define, a cut in R takes effect gradually instead of at once; R* never rises except through R, and every risk override still ranks above it. The formula R itself is unchanged, and the options overlay keeps reading it. The rule was re-backtested over the full history before adoption and is inert at adoption (R* equals R). Filed as a mid-drawdown change — NAV below its 2025-Q4 high-water mark, R 0.64 at decision — and audited as such at the annual review.

s_Fed respec: net liquidity = WALCL − SWPT − TGA − RRP. Swap lines are a stress facility — usage read as easing at squeeze onset, run-off as tightening (Jul-2020: leg 0.92→0.06 vs 0.42 ex-swap into the H2 rally; 2012 both signs). Zero parameters, no-op outside 2012/2020. Re-backtest: mature 40.1/−28.0/1.43, full 57.9/−32.0/1.91.

Execution policy update — tranched execution removed

What changed: The fund previously split any unusually large rebalance — a single move of more than 15 percentage points of NAV — into three tranches spread over roughly three weeks. That rule has been retired. Every rebalance now executes in full at the next daily close, subject as always to the 3-percentage-point tolerance band.

Why: The tranching rule was inherited from a discretionary-investing playbook, where spreading a big, hand-timed trade over weeks guards against picking a bad day. A systematic daily dial doesn't have that problem: targets move gradually, and the rare large jumps come from rule changes or crash-guard events — which already executed immediately. The rule was also never part of the simulated track record: the published simulation has always executed moves in full at the next close. Retiring it makes live execution match the published methodology exactly, and it removes a case where a large de-risking move could have been slowed down at precisely the wrong moment.

Effect on published performance: None. No backtest or model-fund number changes. The risk overrides (acute-crash trim, kill switch) already executed immediately and are unaffected.

Removed the retired override block from fund.yaml — housekeeping, no behavioral change. This is a follow-up from our latest update from 2026-08-09.

No formula constant, weight, threshold, or override parameter changed. The hash moved only because the overrides: block is part of the ratified config surface.

Risk update: one crash guard instead of two, and a more honest backtest

We've overhauled the fund's pre-committed risk rules — the layer that sits above the allocation formula and can only ever reduce exposure — and changed how we publish the simulated record.

New: an automatic crash trim. Equity-market volatility has a distinctive signature in acute sell-offs. When that signature appears — and the drawdown is actually underway — the fund temporarily halves its Bitcoin exposure, pauses all new buying, and parks its reserve in cash. It releases itself as conditions normalize, in a single move each way: crash insurance is worthless if phased in slowly. No judgment calls and no committee — it reacts at the daily close, and every trigger is journaled here automatically.

Retired: the stress freeze. Our previous guard watched bond-market stress and froze buying during systemic episodes. Re-examined against history, everything it did well the new trim does earlier and better — and what remained was as likely to cost as to protect. Rather than run overlapping insurance, we retired it. Its design is preserved in our records, and the specific scenario it uniquely covered is documented and reviewed annually.

The backtest now shows the whole machine. Until now, the simulated record on this site showed the allocation formula alone; the risk rules acted on top only in live operation. That was a conservative publishing convention, but it meant the simulation wasn't quite the fund you're watching. From today, the backtest re-runs the complete ruleset — formula and risk guards together, exactly as deployed — at the fund's actual daily cadence. Simulated drawdowns improve materially as a result, and we want to be plain about why: most of that improvement comes from a single historical episode, the March 2020 crash. Insurance that has paid out once is not a distribution — treat the improvement as an upper bound, not a promise.

Nothing changes in today's positioning: markets are calm by these measures and the guard sits idle. The honest caveat stands, as always: rules like these are calibrated on a limited history of true crashes, and they will occasionally cost a little in false alarms. That's the premium. We publish the misfires, too.