Portfolio Construction & Position Management

Part 5 is where the framework stops describing and starts building: which positions to hold, how large, in which accounts, and under what rules they stay or go. Part 6 supplies the two scores all of it depends on.

Start with a distinction the industry blurs. Asset selection answers what to own. Portfolio construction answers how each position should behave inside the whole, and that behavior is what the framework classifies. Every position except Bitcoin carries exactly one of three behavioral postures.

These aggregate ranges are doctrine; the dashboard does not build to them, and it measures posture drift against the targets your operating thesis implies (Part 6).

Why three? Because removing any one breaks the portfolio in a predictable way. Strip out Ballast and every drawdown hands you a forced choice: sell Torque at trough prices, or hold it with no dry powder. Strip out Torque and the portfolio keeps its capital and stops compounding: safe, and going nowhere. Let Hype in without governance and reflexive narrative positions grow until one of them breaks the book. The difference between speculation and gambling is governance.

Bitcoin sits outside all three. It is the convexity backbone from Part 3, scored like every other holding, but the score does not size, gate or sell it. Part 3’s rules govern it: accumulate only, pace set by valuation discipline, held outside the three postures and the concentration limits. Nothing in this part sells it or draws on it to fund anything else, though Ballast income may fund its purchases during a dislocation. The exhibit below runs all four through one market cycle.

Read the postures as jobs. Torque turns a thesis into convex exposure. Ballast is the reserve that lets Torque be held through its own drawdowns. Hype’s job description is mostly a list of things it may not do. Torque’s job comes first, because the rest of the structure exists to carry it.

Torque is where the framework takes its risk on purpose: a bet on a structural shift you believe is real, durable and capital-backed, which institutions are underexposed to because mandates, compliance processes or volatility limits keep them out. Returns are lopsided by design: a minority of positions deliver the multi-baggers while the rest underperform. And Torque can fall 50 to 70 percent during liquidity stress. That is the price of convexity; position sizing and Ballast exist to make it payable.

Entry is earned three ways at once. Miss any one and the position does not qualify.

Passing the gate earns a position its place, not its size. Size is earned in stages, as evidence accumulates: the ladder below records what has been proven and carries no percentages by design. The percentages come from the score, through the band table that follows.

The CIS is the composite score Part 6 builds from four components: convexity, risk, macro alignment, and execution. The posture picks the table and the score picks the band within it, and because the bands are set in advance, no threshold has to be argued twice. Read each band as a ceiling: its top is the most a score in it can justify, and its lower bound is not a floor. The dashboard’s portfolio builder scales that ceiling with the score and lowers it when data confidence is low: at full confidence a Torque 70 justifies at most 8 percent and an 85 about 11.5 percent (formula on the Math page).

Account routing follows expected turnover: the more often a position is likely to be sold, the stronger the case for Roth. The account rules come from Part 4; the table applies them to Torque.

Torque that can fall 50 to 70 percent is holdable only if part of the portfolio refuses to fall with it. That is Ballast’s job, and the reason aggressive Torque sizing is possible at all. Ballast businesses fund their dividends when credit tightens and keep operating without access to capital markets.

What earns Ballast its space is rotation. After Torque runs, trim any position that has grown past its concentration limit or past the size its score justifies, and move the proceeds into Ballast; deploy Ballast into Torque during drawdowns that momentum and the thesis still support, and into a position 20 percent or more below cost only after the confirming signal the freeze requires. That loop is the framework’s primary way of buying low without raising outside cash or selling Bitcoin. Everyone plans to buy low; Ballast is the part of the plan that still has buying power when the low arrives.

Qualification is a checklist, not an impression: a position must pass at least three of the five tests below. Yield is deliberately not one of them: income can follow from qualifying, but it never qualifies a position on its own.

Ballast scores lower than Torque of equivalent quality, by construction. The Convexity component carries 30 to 45 percent of the CIS composite depending on the thesis profile (40 percent at the reference weights), and capital-preserving positions lack the TAM headroom behind high Convexity sub-scores. The same four bands apply, with smaller ranges scaled to a reserve’s job.

One concrete reference point is STRC, Strategy Inc.’s variable-rate perpetual preferred stock and the income instrument the framework names as Ballast. STRC qualifies as Ballast under the three-of-five rule on durable cash generation (for a preferred, its distribution record: paid in every period since issuance), low factor correlation (the monthly rate reset is designed to hold its price near its $100 stated amount) and a liquidity-resilient valuation (its $100 stated amount and rate-reset support). It carries no agency credit rating as a new instrument; the framework reads the issuer’s Bitcoin asset coverage as its measure of balance-sheet strength. The issuer resets the rate monthly; it was 12 percent a year, paid semi-monthly, as of September 2026, and your yield depends on the price you paid. Bitcoin drawdowns can compress the issuer’s distributable income and STRC’s market price. Return-of-capital characterization is set each tax year and verified on Form 1099-DIV. The case study at the end of this part notes its 2026 price history, and Part 4 covers how its distributions are taxed.

Placement follows the distribution profile. Taxable is the main home, especially for return-of-capital income and for tax-loss harvesting. Pre-tax suits ordinary-income payers, such as bond funds and REITs whose distributions are not return of capital, where pre-tax space exists. Roth is usually wasted on Ballast, because Roth pays most on the positions that grow the most.

Torque and Ballast govern the capital the portfolio depends on. Hype governs the capital it can afford to lose, chasing asymmetric short-term payoffs where markets misprice assets on sentiment extremes, regulatory shifts or technical dislocations, with nothing structural underneath. Its prices run on attention and reflexive feedback, and they have no floor when the narrative breaks. A Torque position that falls 50 percent can recover, because the regime forces persist. A Hype position that falls 80 percent may be gone for good, and the portfolio never needed it. Many portfolios hold no Hype at all, and are right to.

Whatever its label says, a position is Hype when it shows any of these signs.

Of these rules, the dashboard’s portfolio builder applies the caps, sizing Hype from CIS50 at no more than 5 percent a position and 10 percent in aggregate. On the portfolio you hold, it flags a breach of either cap for review, and its Stop Loss tripwire flags a Hype position 25 percent or more below cost; it records every trade and places no stops. The 15 to 25 percent stops and the 3 to 6 month time limits are yours.

The exhibit below sets the three postures side by side. A CIS score measures quality; the job a position does decides how much capital that quality earns. A CIS70 in Torque sits in the 8 to 15 percent band, and the same CIS70 in Ballast in the 5 to 8 percent band: both strong, at different things.

The portfolio-wide single-position cap, 15 percent by default and 18 percent absolute with a documented override, applies to every posture, but by default only Torque can reach it: Ballast stops at 10 percent unless a documented override holds a position higher (the case study below has one), and Hype stops at 5, with no override. Posture decides what a score is worth, the concentration limits decide how much any one name may matter, and the protocols in the next section keep it that way.

Construction is a moment; management is the practice that follows. Six protocols keep the structure true through the cycle: earnings exposure, momentum, regime stress, additions, thesis integrity, and known events beyond earnings. Each says which part the dashboard watches and which part is yours.

Earnings are binary events. A company can beat estimates and fall because expectations ran higher; it can miss and rally because the miss was smaller than feared. A single report can move a position 10 to 30 percent either way, especially the smaller growth names Torque favors, and conviction about business quality predicts none of it. So the framework sizes around earnings instead of forecasting them.

A position entering its final five trading days before earnings may not exceed 3 percent of portfolio value. The trim happens in one step at T-5, and the cap holds through the report. Take a defense contractor at 10 percent of the portfolio: five trading days before its report it is trimmed to 3 percent, and the proceeds park in cash or Ballast inside the same account until the rebuild. If the thesis is confirmed, the rebuild starts at T+6. Under the weekly add cap of a quarter of the target (Protocol 04), a rebuild takes two to four weeks for positions sized between about 5 and 15 percent (three weeks for a 10 percent position: 5.5, 8, then 10 percent).

The dashboard flags a position above the 3 percent cap in the days before its report. It counts calendar days, not trading days, so the flag can arrive a trading day or two after T-5; start the trim from the earnings calendar instead of waiting for the flag. It trims nothing: the trim is yours, and so is the T-21 to T-6 initiation blackout. Three exemptions apply by doctrine, though the dashboard flags any position above the cap that has a report date on file, exempt or not. Bitcoin has no earnings and answers to Part 3. Broad index ETFs diversify away single-name earnings risk. And Ballast positions at or below 5 percent, where earnings are not thesis-critical, are left alone: a single report rarely threatens a stable name’s classification, so the burden exceeds the benefit.

The framework asks the market to confirm what you believe, along three dimensions. Absolute momentum is distance from the 52-week high: within 10 percent is healthy, 25 percent or more below is a correction, and 40 percent or more below is severe distress. Relative momentum is performance against the sector and the market, which tells you whether the problem is the company or the tape. Breadth is the share of related names making new highs: broad participation confirms fundamental support, and narrowing leadership warns of exhaustion. How many of the three break decides how much capital a thesis may carry; which one breaks does not matter.

The exit rung is the framework’s hardest discipline, because the tripwire fires on what the market is doing while your conviction is still saying hold. The same momentum filter that forced the exit decides when you may come back.

The dashboard does not yet measure these three dimensions. Its Momentum Death tripwire reports unavailable rather than guessing, and a moving-average and RSI screen is coded but not yet fed any price data. What it does watch is stress across your holdings as a group: a cohort confluence tripwire that escalates as more warning signs coincide, from a flag, to hedging only, to a displayed playbook for trimming the Torque cohort by 25 percent. Running that playbook is yours. The Math page lists the ten signals and how they escalate.

Three companion heuristics, borrowed loosely from Newton’s laws of motion, shape the day-to-day.

Sometimes the problem is not the position but the water it swims in. When cross-asset correlation rises and liquidity drains away, positions that looked independent in calm markets trade as one, and Torque drawdowns outgrow any single-name analysis. The response is a throttle, run as a state machine rather than a judgment call.

The stressed-regime indicators are VIX sustained above 25, credit spreads widening, and Torque positions correlating above 0.7. The dashboard watches all three as signals: its macro tripwires track the VIX and high-yield credit spreads, and its Correlation Spike tripwire fires when the average pairwise correlation of all your non-cash holdings, not Torque alone, rises above 0.70, reporting unavailable rather than calm when price history is too short (the Math page has the thresholds). When a tripwire breaches, the Add Position screen recommends pausing additions and still records the trade. The response is yours, and deliberately boring: pause new Torque adds, hold posture limits strictly, consider reducing gross exposure 10 to 20 percent, and return to standard positioning only after stress indicators normalize for two weeks or more. None of this is market timing: it cuts exposure only when correlation makes position-level diversification unreliable.

Averaging down without rules is how conviction destroys portfolios, so additions are governed too. A position 20 percent or more below its cost basis is frozen: no new capital without a confirming signal, such as a higher low, a reclaimed key level, or new information that supports the thesis. Scaling in happens over weeks, never more than a quarter of the target in one week. Frequency limits pace the rest: a position’s score is recalculated at most every 7 days (except after material news), adds come at least 3 days apart (except in an earnings rebuild), partial trims at least 5 days apart (except under earnings proximity), and an exit is never rate-limited when a tripwire fires.

Pacing works like the concentration caps: the −20 percent freeze and the weekly add cap are practitioner discipline, and the dashboard records every trade you make and never blocks one. Its Weekly Review paces only its own advice: it waits 5 days before repeating a resize recommendation for the same position, never delays an earnings-window trim or a tripwire response, and leaves a score younger than 7 days alone. Nothing in the dashboard spaces your adds, so the 3-day interval is yours too.

Every position carries one to three core assertions written down at entry: “defense budget growth sustains through 2027,” “datacenter capex has three or more years to run,” “management is executing on margin expansion.” When a core assertion breaks, reassessment is immediate and price is irrelevant: the position is downgraded or exited, and a later price recovery does not rehabilitate a broken thesis. Momentum rules handle the price problem; this protocol handles the truth problem.

Regulatory decisions, policy announcements, contract awards and insider lockup expirations all get earnings-style treatment: keep a calendar of known binary events for each position and apply the earnings protocol to each, capping the position ahead of the event or hedging it. Diffuse macro noise, such as Fed meetings and election cycles, triggers nothing unless it bears directly on a thesis. Every protocol in this section governs positions one at a time; the next section widens the lens to the forces those positions share.

The unit of allocation is the force. A sector matters only to the extent it captures a structural force the practitioner has identified (Part 2 sets out how), and the five forces named in Torque’s entry test serve as the examples.

Under the example thesis, the regime-aligned sectors are concrete: defense and aerospace (government customers, multi-year visibility, pricing power, strong balance sheets); AI infrastructure (semiconductors, datacenter buildout and power delivery, favoring picks-and-shovels businesses with order backlogs over speculative plays); energy infrastructure (traditional and nuclear power, grid modernization); and industrials tied to reshoring, chosen selectively, because many industrials are cyclical with no structural tailwind.

The same thesis names the regime-challenged sectors it avoids: long-duration unprofitable growth that depends on low rates (Hype if at all, never Torque), consumer discretionary without pricing power, and commercial real estate facing secular office decline and a refinancing wall.

Concentration in a force is intentional; concentration in a sector is not. The method is to diversify within the force: own several economically distinct businesses that earn from the same driver through different mechanisms, so one sector’s failure cannot take the thesis down with it. The example illustrates the method; it does not recommend the sectors shown.

A practitioner bullish on AI infrastructure expresses it through semiconductors, datacenter REITs, power infrastructure and cooling technology, rather than through 45 percent in one sector. The limits below hold that shape in place.

Force-level allocation sets the portfolio’s shape; rebalancing decides when the shape may change. The framework rejects mechanical rebalancing, on a calendar or back to fixed percentage bands. Drift past a threshold triggers a review, and the review decides; nothing trades on its own. Mechanical rules ignore what the price move is saying. A position up 50 percent may have appreciated for a reason, and trimming it on a schedule sells winners for winning. A position down 30 percent may be breaking, and adding to it on a schedule catches a falling knife.

So rebalancing follows conviction. On upward drift, recalculate CIS and ask one question: does conviction still support this allocation? If conviction is unchanged, let it run; if it has declined, trim. On downward drift, recalculate CIS and ask a different one: does this decline change the thesis? If conviction is intact and the decline temporary, add, after the confirming signal Protocol 04 requires once a position is 20 percent or more below cost; if conviction has declined, reduce or exit. Between weekly passes, a rebalancing review is called for when a position moves 15 percent or more in a single week, when the earnings protocol triggers, when a tripwire breaches, or when material news changes the thesis. Part 6 sets out the weekly routine and the two other events that call for an early look: a single-day move of more than 5 percent (the dashboard lists any move of 5 percent or more on its Governance page) and a CIS update from new research.

Every rebalancing decision also has a tax side. Part 4 shows why the account a position sits in can matter as much as the position; turnover is where that advantage is kept or given away. Run the arithmetic once. Take $100,000 earning 10 percent a year before tax for 30 years, with each year’s gains realized and taxed at a 25 percent blended rate: the same asset, traded the same way, in two different accounts. Taxed every year, it compounds at 7.5 percent. In a Roth, the full 10 percent survives.

The terminal values are about $875,000 in the account taxed every year and about $1,745,000 in the Roth: a gap of roughly $870,000, so the Roth ends with about twice the wealth from placement alone. The 25 percent rate is illustrative, and the Roth figure assumes qualified withdrawals under current law. Part 4’s Tax Wedge works the case of a gain realized once; this exhibit works gains realized every year, compounding through time.

In Roth, rebalance without friction; that freedom is why rotating Torque lives there, not a reason to rotate. In taxable, prefer new contributions over selling; when a sale is necessary, sell long-term holdings first and coordinate with tax-loss harvesting. Across accounts, treat every rebalancing event as a chance to re-check placement: a position whose expected turnover has changed may belong in a different account, which usually means steering new money there, since cash taken out of a Roth or pre-tax account is a distribution. Every rule in this part holds only if changes to the rules are themselves governed, and that is the last section’s job.

Every framework decays the same way: one small unexamined change at a time. A threshold is nudged, a stop widened, a cap waived just this once, until the framework is a memory wearing its own name. The defense is to classify every change before making it: a doctrine change means abandoning the framework, a parameter change means calibration, and an override means a temporary, documented deviation for stated reasons.

Silent drift is not an override. A stop “temporarily” widened without a note is the signature of a framework being abandoned while its owner still believes in it. Overrides exist so that deviation stays visible, reasoned and reversible.

The portfolio boundary is the household aggregate across all accounts: Roth, pre-tax, taxable and Bitcoin cold storage. Every concentration limit (single position, top three, top five, sector) applies to that aggregate, not to individual accounts, and a position held in several accounts counts as one exposure. An account may therefore be concentrated when the household is not: a Roth IRA can hold 100 percent of its value in one position, provided the household’s exposure stays inside the caps. An account concentrated in one name on purpose is designated a single-name sleeve, in writing; its drift is monitored, and it is trimmed or rebalanced if aggregate exposure breaches a limit. A sleeve is a way of governing concentration, never a way around the limits.

The case study below puts this part’s rules to work in one illustrative portfolio.