Strategy landscape — capital tiers, competitive structure and out-of-scope models

Strategy landscape — capital tiers, competitive structure and out-of-scope models

Written: 2026-07-16 (session 5) / Nature: analysis document (not a verdict, not an experiment record) Purpose: permanently record structural judgments that had existed only in conversation. A map of why some strategies are possible and others impossible. Every figure here carries an [inference] or [measured] label. This document approves no candidate — approval comes only from preregistration (D-xxx) + execution.

1. Where we stand

1.1 Reconfirming the objective function (Master Plan 0.2)

The largest expected value of this project is not the return on equity of any strategy. On 730 USDT, even finding a 20% annual edge yields 200,000 KRW a year. By contrast, a verified track record with preregistration, ledgers and auditability is an entry ticket to raising capital, collaboration and management, and its value accumulates regardless of capital size.

A sober statement: so far the "revenue model" with the most certain expected value in this project has not been a strategy but the validation machine itself. A record of burning through 10,000 trials with discipline is already an asset on that axis. This is not consolation but grounds for resource allocation — the reason to pick the latter when choosing between strategy search and record quality .

1.2 The capital tier ladder (BTC futures)

"Small capital" is not one number but a range, and different constraints bind at each boundary.

Key insight: our problem is not "small capital is a disadvantage" but "below the floor boundary, design is constrained". Measured evidence — at 730 USDT, legs of rotation K=2 fell below the minimum order and the strategy physically could not run (D-014, E-13 [measured]). The floor where design freedom recovers is roughly 2,000 USDT (the D-015 retrial capital), and for multi-leg market-neutral families, around 10,000 USD is the practical floor [inference].

The only real advantage of small capital: slippage is effectively 0, so strategy returns are realized close to theoretical values (Master Plan premise 1). Not higher expectancy but lower deviation — exactly consistent with this project making the validation system its objective function.

1.3 The precise definition of a "niche"

The ecological position of small capital is capacity, not speed. If a strategy's capacity ceiling is below ~1,000,000 USD, it cannot pay a full-time team's salaries, so there is a prior probability that big money is not in it [inference — back-calculated from team operating costs].

An honest caveat: the real competitors in a niche are not the big players but other small players. The pond is small, so whales do not come, but we compete with individuals and small shops who found the same pond, and the edge decays as participants increase. So the claim can go only as far as "being a niche gives a prior probability worth testing", not "being a niche makes it safe" .

2. Competitive structure — games we cannot win

2.1 Why we do not do high-frequency trading (HFT)

The idea: "If we buy liquidation and whale data, write it in C++ and put the server close, don't we become a high-speed quant?" It is blocked structurally at three levels.

(a) Purchased data is not speed. Most liquidation and whale data sold commercially is historical (for research), or, even if real-time, a feed that hundreds subscribe to at once. A signal sold to many has its edge consumed by the fastest subscriber — the moment we buy and use it, we are at the slow end of that queue. HFT's advantage comes not from buying data but from a line that receives earlier and orders that land earlier.

(b) The entry fee is cheap but the opponents are different. Crypto has no traditional exchange-style colocation, so in practice it is just a matter of putting a server in a cloud region near the matching engine, and entry itself is cheap [inference]. The problem is that full-time teams are already in that room — optimized stacks, VIP maker rebates based on huge volume (fees at or below 0), market-making agreements. Where we pay 5bp as taker, they fight while getting paid .

(c) We have already measured that moat from the other side. E-18's entry-candle stops at 31.5% and E-24's negative markouts in every window — these are exactly the numbers of the side that loses to an opponent with an information advantage. phase6-plan's judgment to put OFI/order books under "will not do" was supported by measurement.

2.2 The slope of speed — when C++ is justified

We cannot win at microseconds, but our candidates are not in that range to begin with:

Principle: engineering must follow the evidence, not run ahead of it. If forward validation measures a minute-scale edge and evidence shows that edge is latency-sensitive, then selectively porting only the hot path to C++/Rust is a justified investment. Starting with C++ now is like gold-plating the scope before firing a shot.

2.3 Splitting accounts does not lift capacity constraints

The idea: "Don't the big players split accounts and separate entry/exit logic to take the same volume?"

The capacity ceiling comes not from per-account limits but from order-book depth and the size of the inefficiency itself. If an inefficiency is worth 50,000 USD per event, even split across 10 accounts, what hits the same thin book in the same 10 seconds is the combined volume — impact arises on the total, not per account, and they compete among themselves, raising their own entry price. Splitting the pie into more plates does not make it bigger.

What account splitting does solve: getting around per-account position limits and API rate limits, minimizing footprints for tracking services, separating sub-strategy operations. Big players really do this.

What it does not solve: the capacity ceiling, information disadvantage, fee tiers (usually aggregated across an account group).

Caution: a structure of "enter in account A / exit in account B" that nets to a 0 position in the same market is meaningless in itself, and the only case where it means something is filling against oneself to distort price and tape (wash trading, spoofing), which is not an edge but market manipulation, in the territory of account freezes and legal risk. This project does not do it.

3. List of out-of-scope and conditional models (unverified — not approval)

Things proposed and reviewed so far but not run because they are outside the project scope (single exchange Binance USDT-M, 730 USDT, strategy research). Nothing listed here is a "model with verified returns". It is only a list of what could become preregistration candidates if capital and infrastructure premises change.

3.1 Cross-exchange funding spread arbitrage [main candidate at the tens-of-thousands-of-USD tier]

• Structure: short the same coin's perpetual on the exchange with high funding / long where it is low → direction-neutral, collect the funding difference.
• Rationale: practitioner consensus holds that cross-exchange funding gaps exist constantly and widen while emerging exchanges court liquidity [unverified — no measurement of our own].
• Why impossible with small capital: margin on both sides + liquidation buffers + rebalancing headroom → practical floor around 10,000 USD [inference].
• The dominant risk is not the strategy but exchange credit risk (the FTX lesson) and transfer delays. From this tier on, custody diversification and withdrawal routines come before strategy design.
• If started: begin with a feasibility scan (collect funding history from 2~3 exchanges → distribution of differences → residual after deducting costs and transfer risk). Half a day with the existing fetch infrastructure [estimate].

3.2 Kimchi premium arbitrage [legal review first — execution on hold]

• Cross-border arbitrage by Korean residents raises real Foreign Exchange Transactions Act issues. A legal review must come before profitability, and this document is not legal advice. It will not be started without confirmation from regulatory and tax professionals

.

3.3 Selling the option volatility premium (Deribit etc.) [tens of thousands of USD + mandatory tail rules]

• Rationale: the volatility risk premium is an axis that exists in the literature [verified — for equities and indexes. Porting to crypto is unverified].
• Structural risk: a return distribution that "earns normally and gives it all back at once in a crash" — failing to manage tail risk wipes out the whole account. Only reviewable combined with our validation culture (preregistration, MDD first).
• Given option minimum units, realistic from the tens-of-thousands-of-USD tier [inference].

3.4 What we already have — the exact status of funding carry

It is not "rejected" but "positive expectancy, on hold for falling short of opportunity cost" (D-012). The val 1.24% annualized is real money, just smaller than a risk-free alternative. The simulator is preserved, so the cost of a retrial ≈ 0, and the retrial condition (30-day rolling annualized 8%+ sustained) is registered.

Capital dependence: returns are linear in capital, so 50,000 USD × 8% a year on regime recovery is 4,000 USD a year — worth a look as market-neutral, low-drawdown income. In other words, this candidate is the only item that could switch on not because the edge grew but because capital grew.

Precise status: not "there is no revenue model at all" but "one small model that is not worth switching on now has been verified".

4. What capital opens / does not open

Principle: capital does not create edge. It only widens the set of testable mechanisms.

4.1 Newly opened at tens of thousands of USD

1. Cross-exchange funding arbitrage (3.1) — the main candidate for this tier.
2. Real-spec TSMOM validation — at 730 it was physically impossible at 28 USD per leg, but at 30,000 it is ~1,100 USD per leg and the threshold vanishes. That is, it can be validated at real spec without the "separate validation capital" workaround.
3. Relaxed conditions for reviving funding carry (3.4).
4. Option volatility premium (3.3) — conditional.

4.2 What capital does not change

• The verdicts on the 1st and 2nd priority candidates (E-20/21 forward validation, liquidation branch) — the capacity ceiling covers this whole tier, so capital is irrelevant to the verdict. Validation decides.
• Dead strategies — pairs, rotation and the harness grammar died of no edge, not of capital. Even with more money, a corpse is a corpse (invariant rule 2).
• HFT and market making — fee and speed disadvantages are not reversed by tens of thousands of USD of capital (2.1).

4.3 Discipline on the order of capital increases (learned from D-014)

Capital increases come after the deployment target is validated, not before. During rotation, a "retrial premised on a capital increase" was run, the strategy died and the increase was cancelled too (D-015 → D-016). The list above is used only as a map of "what could be done if added now", and actual transfers happen only after a survivor emerges from the 1st and 2nd priority forward validations.

5. How to use this document

• When a new strategy idea comes up: first ask with chapter 2 (competitive structure) "is this a winnable game?", check with 1.2 (capital tiers) "is it physically possible at current capital?", and if it passes, write a preregistration draft.
• When a capital increase is discussed: chapter 4, especially the ordering discipline of 4.3.
• When an external explanation (collaboration, raising capital) is needed: 1.1 + chapter 2 provide the grounds for "why this team does not do what does not work"

.

• Update rule: when an [inference] here becomes a measurement, update it immediately and cite the measured evidence (E-xx). When the analysis is promoted to a verdict, move it to decisions.md.