Working notes
from the desk.
Methodology, backtests, and risk philosophy from the Koryu research desk. New material every few days. No paywall, no email gate, no preface about why you should subscribe. Just the work.
50 notes
One of these you own. The other owns a piece of you.
The asset versus the position: spot ownership against margined, funded, liquidatable perpetuals, why perp volume dominates the tape, and the three costs off the screenshot.
The ladder is real. The timetable never was.
The rotation folklore is half true: the liquidity-ladder mechanism behind it, where the script breaks, the 2024-2025 dominance era, and the measurable alternative to narrative.
The chart shows demand. The schedule shows supply.
Vesting cliffs, linear unlocks, continuous emissions: how scheduled supply reaches the market, what unlock events reliably do and don't, and the calendar's defensive value.
Same daily behavior, different year. Check the calendar.
The calendar hiding inside volatility, Sharpe, and drawdown figures, three comparison traps between crypto and equity statistics, and the conventions this site pins.
Compute where the cliff is. Then stand well back.
The growth-optimal bet size and its violence: the formula, 50/50 odds of halving at full Kelly, zero growth at double Kelly, and why fractional is the only sane dose.
Trends digest their levels. Ranges recycle them.
Ranges recycle the levels trends digest: the meat-grinder arithmetic, why chop-sensitivity is an exit-design property, and what one regime gate changed in our own test.
We found a way to make the short work. Our own validation gate told us to slow down.
The short side of our candidate strategy had a drawdown problem. Here is how patience and profit-taking fixed it, why our own validation gate still refused to bless the exact recipe, and what measurement-first discipline does with a promising result it cannot fully prove.
Survival first. Compounding is for the survivors.
The probability the game ends: ruin as an absorbing state, streak arithmetic at 1/2/5/10% risk, why losing streaks are scheduled, and crypto's three accelerants.
The curve looks magnificent. That was the selection criterion.
Memory versus understanding, the five amplifiers that make crypto the easiest market to fool yourself in, the trial-count arithmetic, and the plateau principle.
Up 50, down 50 is down 25.
The gap between average and compounded returns at the volatility where it becomes the main event: the half-sigma-squared rule, a fair coin losing 69% a year, leveraged decay.
Everyone can see the average. Almost no one can follow it.
The moving-average family on the reference asset: the 2022 dodge, the whipsaw invoice nobody advertises, why the effect survives being public, and its honest limits.
Risk is losing money, not making it irregularly.
Sharpe with a fairer denominator: downside deviation only. Why right-skewed momentum profiles deserve it, and how the Sharpe/Sortino gap detects skew.
Tune on one era. Grade on one it never saw.
Why full-window optimization takes the exam with the answer key, the holdout and walk-forward design space, what validation costs, and the split this site pre-registered.
Return per unit of risk, flattered by the calendar.
Return per unit of risk, and the 20% flattery crypto's 365-day calendar adds to every annualized Sharpe: the definition, the failure modes, and the deflation we apply.
Outside the noise, inside the disaster: crypto redraws both lines.
Why equity stop geometry gets destroyed on crypto, the liquidation-cascade mechanism behind the wicks, the layered design space, and the no-gap gift that offsets the tax.
We went looking for an edge in picking coins. The data kept pointing at the regime.
We ran a sustained effort to find a coin-picking edge in crypto. On data that includes the coins that went to zero, the answer kept returning the same: the only decision that reliably pays is whether to be in the market at all.
How far does it actually move? One number answers it.
Average True Range from zero: Wilder's formula, why 24/7 trading shrinks true range toward high-minus-low, percentage ATR, and its three jobs: stops, sizing, regime.
Every ranking hides its biggest call in the guest list.
The monitor and research universes, every inclusion rule with its reason, the edge-case ledger for rebrands and delistings, and why publishing selection criteria resists gaming.
The short now respects one more thing: priced-in panic.
Version 2.3 of our candidate strategy adds one clause: the short book stands down when Bitcoin implied volatility is in its capitulation zone. Where the rule came from, the threshold we refused to tune, the five-gate test with shuffled-time placebos, and why it is running in shadow before anything else.
Three majors up, eighty alts bleeding: breadth is how you know.
The share of the board actually participating: the standard measurements, why breadth matters doubly in a one-factor market, and our dial's published breadth input.
Three out of four fail. The mechanics explain which.
The order-book mechanics of a breakout, who is selling into it, the 25% base rate from our own five-year test, the four ways breakouts die, and confirmation without a closing bell.
Fix the loss first. The position is just division.
Risk-first sizing with the worked example: why volatility shrinks size, the portfolio-heat trap when correlations hit one, and the survival arithmetic behind small fractions.
Price needs capital to move. Volume just needs a printer.
Why reported volume lies more here than anywhere, the landmark evidence, four detection heuristics anyone can apply, and the mitigations our pipeline uses with limits stated.
Equities have industries. Tokens have stories.
Equities have industries; tokens have stories. The working taxonomy, what sector membership actually predicts, the diversification illusion, and the data-quality fine print.
Half the audit came back clean. We published the other half too.
PBO, deflated Sharpe, and a survivorship bound pointed at our own flagship: selection came back clean, the universe did not, and the report became our data constitution.
The backtest says 773x. That is why we don't trust it yet.
Our candidate crypto strategy backtests at 773x with the bear years flipped from losses to gains. Here is the complete evidence: the ensemble gate, the short sleeve, the seed test that killed version one, the lookahead bug we caught in our own code, and the seven reasons it runs in shadow instead of live.
A dozen ideas went in. One seatbelt came out.
We fixed the validation battery first, then ran every idea through it: social attention, alt rotation, mean-reversion, MA exits, vol sizing. One thing survived, a funding-crowding risk filter on trend-gated momentum. The honest record of the losers and the one winner.
A real mechanism, a real correlation, and a sample of four.
A real mechanism, a real correlation, and a sample of four: the supply-shock theory stated fairly, the confounders, the shrinking flow arithmetic, and the post-ETF divergence.
We ran our stock engine on crypto. It lost. Here is the anatomy.
We ran a validated US-equity breakout strategy on five years of crypto daily bars, zero re-tuning, and published the losing backtest: what failed, what carried, and why.
The coins that aren't there are the whole story.
Exchange APIs only remember the living. How many coins actually die, what the graveyard does to crypto backtests, and the point-in-time universe rules we commit to.
Every number on the board, recomputable by you.
The complete public specification of the daily board: universe rules, every column's formula, the 4-hour refresh cadence, and why nothing on it is a proprietary score.
The market has weather. This is our barometer, in full.
One fixed formula for crypto market regime: the BTC trend stack plus top-100 breadth, its three states, the evidence behind it, and its honest failure modes.
What the market is doing, never what you should do.
The design decision behind the whole site, and the difference between analytics and advice: published measurements now, predictive claims only after pre-registered gates.
Screenshots ask for trust. Hashes make it unnecessary.
How cryptographic commitments make a published record tamper-evident: the hash chain, Bitcoin timestamping, what it proves, what it can't, and how to verify a day yourself.
The feed can be edited. That is the whole review.
The direct answer, the tracked-vs-claimed accuracy gap, the referral economics that make dishonesty structural, and what a legitimate provider would have to show.
Ninety-five percent accurate, and broke by winter.
The complete catalogue of accuracy-manufacturing techniques, why win rate is the wrong metric even when honest, and the two-question test that collapses the act.
Seven questions, fifteen minutes, no statistics degree.
A fifteen-minute audit anyone can run: editable history, hidden losses, undefined wins, no bear market, no methodology, referral economics, and the screenshot test.
The tide came in. Everyone learned to swim.
Why bull-phase performance contains almost no information, how crypto amplifies the illusion, and the two tools that recover the signal: regime benchmarks and winters.
Predictions are the pitch. Discipline is the product.
Predictions decay when sold; risk management compounds and can be proven. What this service is for: the mission, the economics, and the four artifacts that make it concrete.
Four guardians hold the directions. The fifth holds the ground.
The origin story in the four-guardians mythology: an equity house of four beasts, the fifth position at the center, and the failed transplant that carried its system here.
A real catalogue, an unfalsifiable oracle.
The largest AI crypto ratings platform, reviewed on verifiability: what the product is, which claims can be checked, the seven-checks scorecard, and who it might still suit.
Ranked by proof, not by promises.
Every category of crypto signal product, ranked by whether its claims can be verified: copy trading, analytics platforms, AI ratings, Telegram groups, and measurement boards.
The records are real. The leaderboard still lies.
Venue-verified records make copy trading the most honest signal category, and the leaderboard format still misleads: survivorship, the capital-arrival curse, and execution drag.
The anomaly is real. The industry mostly isn't.
The anomalies are real; the industry mostly is not. The four structural leaks between a genuine edge and a subscriber's account, and the conditions under which a signal can work.
Ninety percent of the dashboard, for exactly zero dollars.
The free screening stack that covers most of what paid dashboards sell: aggregators, charting screeners, exchange tools, on-chain free tiers, and measurement boards.
A fine thermometer, a terrible oracle.
One division with two distortions: the stablecoin denominator problem, the quality of summed market caps, what dominance readings have accompanied, and the joint read.
The chart shows the market cap. The obligation is the FDV.
Price times what, exactly: the low-float worked example, unlock cliffs and emission drips, the float-ratio habit, and why momentum screens attract exactly this failure class.
A poll where every respondent pays to vote.
The mechanism that anchors perpetuals to spot, the annualized arithmetic that makes 0.01% expensive, and funding as a paid poll of positioning worth watching from spot.
No news behind the candle. Just leverage meeting thin bids.
The forced-selling chain reaction behind crypto's newsless crashes: the mechanics, why this market breeds them, the famous episodes, and how to read a cascade on a chart.
The drawdown is the result. The return is the compensation.
BTC has lost three quarters of its value twice in a decade; alt drawdowns start at ninety percent. The base rates, the recovery curve, and what depth does to decisions.