The Framework in Math

A formula is only as useful as your sense of how much to trust it, so three conventions come first.

Authority. Where the framework’s written rules and the running engine disagree, this page quotes the engine, because the engine is what scores your portfolio. Where the two have forked, the page says so rather than picking a side without telling you.

Classification. Not every number carries the same weight. Each quantity below is one of four kinds, and the kind matters more than the value.

Coded, or written down. Some of what the framework asserts runs as code against your portfolio; some of it is a written rule a person follows. This page keeps the two apart wherever mixing them could mislead. A threshold is called live only when a running consumer changes state because of it. Logic that exists but receives no input is called coded but not yet fed, and the page names what is missing. Descriptions of what the dashboard does are as of September 2026.

Everything else on this page starts from one number per holding. The Convexity Integrity Score is a weighted sum of four components, each scored 0 to 100, producing a 0 to 100 result. At the reference weights:

CIS = (C × 0.40) + (R × 0.25) + (M × 0.25) + (E × 0.10)

The four-component structure is doctrine. The weights are parameters: the engine holds them in one place, and the active thesis shifts them within a fixed envelope, described below.

Six distinct objects travel under the name. Comparing one with another and calling the difference a bug is an easy way to misread the system, so they are separated here first.

The 40/25/25/10 split above is the reference weighting. When a thesis is active, its profile sets the weights, which is why one position can score differently under two theses without either score being wrong.

One further adjustment is built in rather than chosen. When the macro read is low or medium confidence, the macro weight drops to 60 or 80 percent of its value, and the difference moves evenly to convexity and risk. A thinly evidenced macro view does not get to carry full weight.

The heaviest component at the reference weights, because asymmetric upside is the framework’s objective. It is built from four capped sub-scores that sum to 100: headroom 35, optionality 25, catalyst density 20, scarcity 20.

Headroom (0–35) asks how much larger the addressable market is than the company. Let H be that ratio, capped at 30×:

headroom = 35 × ln(1 + H) / ln(31), where H = min(TAM ÷ market cap, 30)

Logarithmic, so the first multiple of headroom counts for far more than the twentieth: 30× scores 35, 10× scores 24.4, 5× scores 18.3, 3× scores 14.1, and parity, an addressable market exactly the size of the company, still scores 7.1; the curve reaches zero only when the ratio does. The shape is doctrine; the 30× cap and the log base are parameters.

Optionality (0–25) is size-convexity (0 to 18) plus sector convexity (0 to 7), and the thesis themes can add up to 4 of the sector points. Size-convexity is a continuous curve over market capitalization that peaks between three and forty billion dollars and falls away on both sides, which is the arithmetic form of the claim that size limits how far a company can travel. It is 14 below $0.5 billion and ramps to 16 at $2 billion; climbs from 16 to 18 between $2 billion and $3 billion; holds at 18 through $40 billion; falls to 10 by $100 billion and to 4 by $200 billion; then tails off exponentially toward 2. An unknown capitalization scores the midpoint, 9, rather than a guess. Four override flags (pricing power, a margin inflection, an embedded call option, regulatory leverage) can lift a company above $40 billion back toward 18; they are coded but not yet fed, because nothing in the software sets them today.

Funds score through the same size-convexity curve as equities, on the weighted-average market cap of what they hold rather than their own assets: a broad index fund lands in the curve’s low, mega-cap zone, while a niche fund of smaller names can reach its peak. No data provider supplies that weighted average today, so the figure is usually an estimate (a language-model figure or a category average), and the score records which one it used.

Catalyst density (0–20) prices identifiable, dated reasons for a re-rating. Each catalyst earns:

points = impact × probability × independence × confidence × time decay

Impact is tiered: transformational 7, major 5, moderate 3.5, minor 2. Independence runs from 0.5 to 1.0, so a catalyst that repeats another is discounted by up to half. Time decay is hyperbolic, 1 ÷ (1 + months out ÷ 12): a catalyst today counts fully, one at six months two-thirds, one at a year half, one at two years a third. Only the top four count, and catalysts spanning different impact tiers earn up to 3 more points.

That formula runs on a researched catalyst list when one exists. Without one, the engine scores a sector template list, blended toward a neutral 8 at 70 percent weight, or failing that a thematic prior. A catalyst score with no researched catalysts behind it is a sector estimate, and the score’s breakdown records which source it came from.

Scarcity (0–20) prices what cannot be replicated. Each moat signal earns type weight × strength × durability × confidence, where type weight runs from protocol scarcity at 5.0 down through regulatory license, supply constraint, network effects, patents, capital intensity and data, to switching costs at 3.0; durability multiplies by 1.0 for permanent down to 0.45 for short-lived. The top five signals count, plus a bonus for breadth of moat type. Without researched signals, scarcity falls back the same way catalysts do.

Survivability under stress, which is a different thing from volatility. Higher is less fragile. Four sub-scores sum to 100: balance sheet 30, business model 30, factor correlation 20, tail risk 20. There is no sector bonus.

The thesis enters risk through one bounded channel:

risk penalty factor = clamp(2.0 − thesis volatility multiplier, 0.5, 1.5)

A thesis with more tolerance for volatility softens the risk penalties; one with less sharpens them. The factor applies only to factor correlation and tail risk. Balance sheet and business model are untouched by the thesis, on purpose.

And it stops at insolvency. When a company shows a current ratio below 0.8 and debt-to-equity above 3.0, on reliable data (observed, derived from observed inputs, or a stale balance-sheet reading, never a proxy or a language-model estimate), the penalty factor is floored at 0.90. A thesis can soften a volatility penalty. It cannot forgive a balance sheet.

Factor correlation measures correlation to macro factors only. Overlap with the rest of your portfolio is left out on purpose: the construction score handles concentration, and counting it here as well would charge a concentrated book twice.

Regime fit, not forecasting. Three sub-scores sum to 100: regime fit (40 points), carry direction (30) and policy and flow (30), read against the macro thesis of Part 2.

Carry is dividend-neutral: yield is not scored as quality. Without a supplied carry score it sits at a neutral baseline of 20 for equities and preferreds (22 for income funds), plus a +3 for a return-of-capital preferred held as Ballast outside a known tax-advantaged account, and a macro adjustment of at most 2 points. Where the dashboard has beta or drawdown data but no yield, it supplies a stability proxy instead (10 to 27 points, from beta and one-year drawdown), and the bump and the macro adjustment do not apply.

Regime fit and policy are written upstream. Your sector preferences (the over, neutral or underweight tilt you set for each sector) set the base: regime fit 32, 22 or 10; policy 24, 16 or 7. Theme overlap with the thesis adds up to 6 and 4 points, and extracted policy evidence can add up to 5. The live macro regime then moves regime fit by at most 4 and policy by at most 3. A macro read can tilt these sub-scores; your own sector preferences move them much further.

The live macro regime is one of four labels (hawkish, dovish, stagflationary, neutral) set by the direction of three series: the Fed funds rate, CPI inflation and the unemployment rate. At least two must show a confirmed trend, or the label stays neutral. Rates, inflation and unemployment all rising read stagflationary; rising rates, or rising inflation without falling rates, read hawkish; falling rates without rising inflation read dovish; anything else is neutral. A new label must repeat across two consecutive syncs before it replaces the old one, and data more than 48 hours old leaves scores unadjusted. The yield curve is recorded but does not enter the label, and unemployment decides only the stagflationary case; there is no direct measure of output growth. By sector group, the label moves regime fit by at most 4 points, policy by at most 3 and carry by at most 2, a few CIS points at most. A second, portfolio-level reading (risk-on, selective, transition, defensive) combines this label with the VIX, high-yield spreads, tripwire alerts and how broadly your holdings are leading; it shapes the Next Dollar Score, not CIS.

E measures whether reality is starting to validate the thesis. It carries the least weight because execution follows quality: momentum confirms a thesis, and it never gets to dominate one. Two sub-scores of 50 points each: execution quality, read from three-month price momentum with a volatility adjustment, and market acceptance, read from trading turnover. That is the equity route; funds use expense ratio and assets, and Bitcoin and crypto have their own inputs.

Momentum has a strict definition, and the engine checks it: (price now − price three months ago) ÷ price three months ago, stored as a ratio and requiring at least 60 observations. Below that minimum, or when a one-month series stands in for a three-month one, the value is demoted to a proxy and loses weight. A language-model estimate may fill it for display, but the score refuses one (Evidence).

Market acceptance is turnover: thirty-day dollar volume over market value, clamped to a sensible range and scored on a tier ladder. It asks whether the market is actually trading the name, which is a separate question from whether the price went up.

Preferred equity runs an inverted momentum ladder. For an instrument held for its carry, stability is the good outcome, so the flattest tape scores highest.

Private → Bitcoin-class → fund → crypto → equity. First match wins, and nothing falls through to the equity path. Bitcoin-class means the asset itself or a spot wrapper for it, and it is checked before the fund branch so that a spot Bitcoin fund scores through the monetary model rather than as a generic fund. Cash sentinels are never scored on the market path at all.

This is the framework’s central guard against fabrication. The failure it prevents is scoring an instrument against inputs that do not exist for it (a preferred share graded on revenue growth, Bitcoin graded on a balance sheet) and producing a confident number from nothing.

The engine records the neutral score and a thesis delta (the weighted sum minus neutral) with every score. That delta captures what the weights and sector relevance do, but it also carries the macro downweight and, for Bitcoin, the addressable-market penalty in the cascade below, and neither of those is a thesis effect. It cannot show the thesis inputs inside the components, because they sit in both numbers. Subtracting Neutral from the final Adjusted score would also mix in the later evidence stages, so read the recorded delta instead. The framework budgets a thesis delta of 3 to 25 points, and a delta beyond 30 is worth investigating. That is a design budget, not a measured range, and nothing in the dashboard flags it.

An update is capped by how good the evidence behind it is. Low confidence permits a move of ±3 points, medium ±5, high ±8, and evidence derived through a proxy rather than observed directly ±6. A first score has no prior, so nothing clamps it. A material thesis change permits ±15. Because any move larger than 20 points bypasses the clamp as a model disagreement (below), the bands restrain only moves between their limit and 20 points: a low-confidence move of 12 is cut to 3, a move of 21 goes through.

Three cases skip the delta clamp: a change in the scoring model’s version, which re-bases rather than adjusts (for crypto, a change in the model that produced the prior score); a private position, scored on a separate structural path with no clamp; and the model disagreement above, which lands in full. So moderate re-ratings, up to 20 points, earn their way across successive updates, while a larger jump passes through and is logged as a model divergence for you to inspect. Machinery built to resist mood should not also be able to suppress a real correction.

Every metric enters carrying a provenance, and provenance carries a weight. A metric is blended toward a neutral baseline in proportion to how much it is trusted:

contribution = baseline + (raw value − baseline) × admission weight

Two further gates sit above the individual metrics. Coverage weighs the core fields at 0.70 and the enhancing fields at 0.30; below 70 percent the score is capped at 75, below 50 percent at 68, below 30 percent at 60. Language-model estimates count as present here; the admission weight is where they are discounted. Proxy suppression caps the score at 85 when more than 40 percent of the sub-scores rest on proxies or heuristics.

Both encode the same rule: incomplete evidence limits how good a score is allowed to look, and never inflates one. A missing field can cap how high the score may go, and it lowers the score’s confidence. It is never scored as if the missing value were bad: an absent margin is not a zero margin.

Two holdings with the same inputs get the same score; a tie is a tie, and what separates thin-data holdings is coverage and confidence. Until engine version 3.2 the engine also added a fixed offset computed from the ticker symbol, up to 20 percent of a sub-score’s maximum, which carried no information about the holding. The documentation audit found it, and it was removed on October 3, 2026; the Evidence page records what it did.

Each stage is held within 0 to 100, and the result is carried as an unrounded float: a position scores 78.3, not 75 or 80. Only the wrapper-adjusted number shown beside a holding is rounded.

Scores stay floats until display, where they fall into four bands. The bands are half-open and matched from the top: 70.0 lands in Strong, 69.999 in Moderate.

In the dashboard the same bands carry numeric labels (70+, 60–69, 50–59, ≤49) and conviction tiers (core, standard, starter, avoid); the Next Dollar Score’s strong, moderate and weak ladder cuts at different points, so read the two separately.

The framework also carries a calibration target for how a healthy universe should distribute: roughly 3 to 5 percent above 88, 15 to 20 percent from 80 to 87, 30 to 40 percent from 70 to 79, 25 to 30 percent from 60 to 69, and 10 to 15 percent below 60. It is illustrative, and no code applies it. Its use is diagnostic, and the rule attached to it is worth stating in full: if the universe cannot reach 88, audit the implementation before questioning the philosophy.

The canonical score stops at the single holding. The overlays in the table above are where context enters, and the book as a whole is judged by the construction score, next.

Twelve good positions can still make a bad portfolio. Where CIS judges one holding, the Framework Integrity Score judges the book as assembled. FIS starts the assembled portfolio at 100 and deducts capped, named penalties in five buckets: allocation (25), governance (15), dead capital (15), concentration (15) and complexity (10, hard cap).

FIS = max(0, 100 − Σ min(bucket penalty, bucket cap))

The caps sum to 80, so a valid FIS cannot fall below 20. That floor is derived, not declared: it falls out of the caps. It is also only a bound. Governance and dead capital are value-weighted (below), so for any book of fewer than 200 positions each of those buckets stays under 10 points, and FIS stays above 30 even with every other bucket full. A zero on a screen therefore means a broken input.

An invalid portfolio (no positions, no capital, malformed input) returns null, never 100. An empty portfolio is not a perfect one.

Two of the five buckets scale their penalties by how much of the portfolio the offending position represents:

severity = clamp(position value ÷ portfolio total, 0.002, 0.12)

Governance and dead-capital charges are multiplied by the position’s share of the portfolio, counted between 0.2 and 12 percent: a 6-point charge on a 5 percent position costs 0.3 points, and no single position can cost more than 0.72 in governance or 0.84 in dead capital. Allocation, concentration and complexity charges are flat. The floor means a rounding-error position still registers; the ceiling means one enormous position cannot fill a bucket by itself. This is what value-weighted means wherever the framework uses the term.

Allocation checks two things. The first is the dollar split across Roth, taxable and pre-tax, measured against the FIS wrapper targets of 45, 35 and 20 percent and charged only once total drift passes 10 points (worth at most about 6 points). Bitcoin held under the dashboard’s own Bitcoin wrapper counts in the total but has no target, so it adds to the measured drift. The second is the count of positions in each score band against a target spread: up to 5 points per band, and nothing for a drift of 2 points or less. The bucket does not check whether each position sits in the wrapper Part 4 assigns it; that placement stays your check.

Governance applies one charge per position, first match wins, each multiplied by the position’s severity: a score below 60 carries 6 base points (50 to 59 is charged the same as below 50); a position scoring 60 to 69 that weighs more than 8 percent of the portfolio, 6; a position scoring 60 to 69, 4; a momentum breakdown, 4 (specified, but nothing in the engine sets it yet). A 5 percent position scoring 55 therefore costs 0.3 points. This bucket reads the canonical CIS, not the wrapper-adjusted one that the dashboard’s governance checks use.

Dead capital charges 5 base points, value-weighted, for a position with no documented thesis. A score older than 90 days is stale: the dead-capital bucket charges it 2 base points, value-weighted, and the dashboard’s diagnostics warn once a score is more than 60 days old and fail past 90. A position more than 50 percent above its cost basis whose score is older than 60 days is flagged for a refresh. Both charges can fire on the same position, and a missing timestamp is treated as unknown rather than stale, so it draws nothing.

Concentration charges 8 points for each position above 15 percent, matching the default cap in Part 5, so a position inside its 15 to 18 percent override band is still billed. It charges 6 if the top three exceed 40 percent and 4 if the top five exceed 60, looser than Part 5’s caps of 35 and 50. Bitcoin and its spot wrappers are excluded from all three measures by archetype, because the backbone is not a concentration failure. Equities with Bitcoin exposure are ordinary concentration.

Complexity charges 1 point for each position with an unclassified distribution, and 0.5 for each position scoring below 70 that is carried without documented rationale, beyond an allowance of three. By the FIS rules a documented thesis, hold rationale or remediation plan suppresses the low-score charge.

These are the framework’s FIS rules as the engine computes them. Some inputs are not yet fed: nothing sets the momentum-breakdown flag, and the thesis, distribution-type, hold-rationale, remediation-plan and override records you keep on a holding are not read by the score, which in the current build bills every position as undocumented and unclassified. By the same rules, a concentration breach covered by a documented override is marked as acknowledged and charged in full. The score does not yet read overrides, so the acknowledgment does not appear; the points are the same either way.

FIS reaches outside itself in one place, forward projections, and it does two things there. One projection path scales expected realization by construction quality:

multiplier = 0.50 + (FIS ÷ 100) × 0.70

A FIS of 50 yields 0.85, a FIS of 100 yields 1.20. Another projection path scales modeled volatility the other way, from 1.20 at FIS 0 to 0.80 at FIS 100. The claim being modeled is narrow: a poorly built portfolio is assumed to capture less of its own assets’ upside, and to swing more on the way. FIS never feeds back into CIS, and never into itself. The projection models themselves are among the surfaces this page holds back (Scope).

Neither score is a position size. Turning a score into a weight takes a posture and a band table, which come next.

The translation takes three things, a posture, a band table and a budget, and every number it produces is a ceiling. The arithmetic below is what the dashboard’s portfolio builder applies to the model portfolios it generates. On the book you actually hold, nothing resizes anything: the bands are the guidance you size against.

Within a band the ceiling moves linearly with the score:

justified weight = min weight + band progress × (max weight − min weight), capped at 15 percent, where band progress = (score − the band’s lowest score) ÷ the band’s width

So a Torque position at 70 justifies 8 percent, at 85 justifies 11.5 percent, at 100 justifies 15. A Hype position at 50 justifies 2 percent, at 100 justifies 5.

The builder then scales every ceiling by how complete the build’s data is. A data-confidence score (machine-data coverage weighted 70 percent, analysis coverage 30) sets a multiplier: 1.0 at 80 or above, 0.85 to 0.99 from 60 to 79, 0.65 to 0.84 from 40 to 59, and 0.40 to 0.64 below 40. A Torque position at 85 with a multiplier of 0.65 is capped at 7.5 percent instead of 11.5. A Torque position at 60 with a multiplier of 0.40 computes to 1.6 percent, and it is dropped, for the reason in the next paragraph.

The band’s lower bound is not a floor. It is the low endpoint of the interpolation. In the main pass a computed weight below 2 percent is discarded, never raised. Three fallbacks then fill gaps. A wrapper less than 30 percent deployed takes candidates scoring 55 or more at Ballast ceilings. A wrapper that ends with fewer than 10 positions (8 on the fund route) is filled from admitted candidates scoring 50 or more, at no less than 1 percent each, keeping any fill of 0.5 percent or more. And a wrapper that would otherwise hold nothing takes up to three candidates at no less than 2 percent. So the builder will decline a weak name, and it will also fill a thin wrapper to a minimum count.

Capital and the posture budget size below the ceiling. The ceiling says how much a score justifies; it never says how much to buy.

Concentration appears in three places, at three sets of thresholds, and each has a different consumer. The doctrine is Part 5’s: a single position capped at 15 percent by default, with an 18% absolute maximum under a documented override, the top three at 35 percent and the top five at 50. Bitcoin sits outside all three layers below, and STRC sits outside the builder’s caps. Both of the builder’s exceptions fork from that doctrine, which lets a single position past 15 percent only under a documented override, never past 18, and stops a single Ballast position at 10 percent. They describe what the software does, not the framework’s limits.

You do not tell the builder to hold eighteen names. It works out how many the available conviction supports, starting from the density of the admitted candidate pool:

High: average score at least 75, with at least six names at 70 or above. Moderate: average at least 67, with at least ten at 65 or above. Low: average at least 60, with at least eight at 60 or above. Scarce: anything else.

Density sets both the target count and how much capital deploys, each by formula. Denser conviction concentrates into fewer names: High targets 14 down to 10 positions and deploys fully; Moderate targets 18 down to 13 and deploys 95 percent; Low targets 16 to 22 and deploys 75 to 90 percent; Scarce targets 20 to 25 and deploys 60 to 75 percent. Counts stay between 10 and 25 names (8 and 15 on the fund route), but never exceed the number of candidates scoring 60 or more.

The undeployed remainder is held as cash, on purpose. On every build, position weights plus cash sum to exactly 1.0, so unallocated capital shows up as an explicit cash position instead of being spread over whatever ranked next.

A build is one moment. Governance is what watches the book you hold afterward.

Governance watches a defined set of conditions and speaks up when they cluster. What it does when it speaks up is the part worth being exact about, and these six terms are not interchangeable.

Start with what is not measured. The dashboard does not yet measure the three momentum dimensions of Part 5 (distance from the 52-week high, relative strength and breadth). Its Momentum Death tripwire reports unavailable rather than guessing. A coded screen of price against the 200-day and 50-day moving averages with an RSI band of 30 to 70 receives no data, so it reports every position as unavailable, and it is not treated as a stand-in for the three dimensions.

What the dashboard does watch is stress across your holdings as a group. Its cohort confluence tripwire tracks ten registered signals, read from your non-cash positions (a fixed eight-name reference list stands in when you hold fewer than three) and from market-wide liquidity and volatility data. Eight have evaluators of their own, two of which fall back to rougher approximations when better data is missing; the other two slots, rates-volatility stress and market breadth, are filled by approximations built from free data. A signal read by approximation counts for 0.6 of a signal in the weighted score. Two signals within 48 hours raise level one, flag only; three raise level two, hedging only; four, or three including an intraday reversal, raise level three, which displays a pre-set playbook to trim the Torque cohort by 25 percent. The dashboard shows the playbook; it does not queue, propose or execute it.

Four of the signals, as examples: an equal-weighted cohort gain of 12 percent or more over three sessions, or of 7 percent in one session with at least five names up more than 4 percent; four or more names at a two-day RSI of 98 or higher; three or more names trading 8 percent or more above the prior close and then closing at least 5 percent below the session high, on two and a half times their 20-day median volume; and a three-session decline of 4 percent or more with at least 60 percent of names at five-day lows.

Escalation is not a raw count. The thresholds (two, three and four signals) scale with how many signals the system can read at all, directly or by a labeled approximation. Levels two and three need both the count and the weighted score; level one can be reached on count alone. A signal counts for 48 hours after it fires. A signal that cannot be evaluated on a run cannot fire, and it is reported unavailable rather than quiet. Most such signals still count toward the thresholds, but a few (the dealer-positioning index without a fresh reading, the leadership flip without data, an approximation short of data) drop out of the count, which can lower the thresholds.

A core set of macro conditions is watched alongside the cohort, each with a watch, caution and critical step: liquidity contraction (net liquidity, the Fed balance sheet less the Treasury account and reverse repo), a volatility regime shift at a VIX of 20, 25 and 35, credit stress at 1.0, 1.5 and 2.5 standard deviations of high-yield spreads, yield-curve inversion, a dollar spike of 2, 3 and 5 percent over ten days, bond-volatility acceleration (defined, but reported unavailable until a genuine MOVE feed exists), and a compound condition that needs both a VIX above 20 and a sharp five-day rise in it.

Earnings proximity is checked position by position. The dashboard’s Framework Rule Register flags a position above 3 percent of the book once its report is five calendar days away or less, and the positions table’s status column starts marking the report six calendar days out. Because both count calendar days, the flag can open a trading day or two after the doctrine’s T-5. It trims nothing; the trim is yours.

The weekly loop itself is set out in Part 6. In the dashboard you run it from the Weekly Review, which recommends actions in a fixed precedence: a finding from the dashboard’s three-level check first, then an earnings window above the 3 percent cap, then FIS below 70 (fix the top penalty), then a CIS move of 10 points or more (resize); no trigger means hold. The review cannot yet raise the earnings step, because it receives no earnings dates; the Framework Rule Register flags that case instead.

The three-level check also runs with every portfolio recompute. Level one flags any position whose wrapper-adjusted CIS is below 75, level two a sector where more than half the scored positions sit below 80, and level three a FIS below 60 (on a recompute, also three or more of its broader rule checks failing at once). Their deadlines are 7 days, 14 days and 24 hours. None is an exit signal and none trades. Level one sits inside the Strong band, so a holding whose wrapper-adjusted score is between 70 and 75 draws a recommendation every time you run the review, even though on CIS 70 is a band boundary, not an action trigger.

The dashboard records every trade and never blocks one. When a tripwire breaches, the Add Position screen recommends pausing additions and still records the trade. No governance path in the system can modify a position, write a trade, alter an allocation weight, trigger a rebalance, or change how CIS or FIS is computed.

The complete set of things governance writes is: evaluation proofs, transition events, acknowledgments and other issue-lifecycle records, watch-list selections, change proposals and their changelog, override records, single-day-move flags, the advisor notices it raises, and the Weekly Review’s own record (the review, its decision log, the recommendations it made and the trigger queue it consumed). A Weekly Review may also ask the scoring pipeline to refresh up to ten stale scores, which rewrites those scores the normal way. Nothing else.

That list is short by design, so every emergency condition in the framework ends with a person, not the code. Of the twelve emergency conditions in the framework’s governance rules, five are detected live and shown with a pre-set response: a construction score below 60, a CIS drop of 15 points or more in 30 days, a Hype position 25 percent or more below cost, average correlation above 0.70 when price history allows, and a top five above 65 percent excluding Bitcoin. Two more are coded but report unavailable until their inputs exist: all three momentum dimensions turning negative, and a portfolio 20 percent below its peak. None executes its response. The other five are written instructions for a human: a broken position thesis, a regime change, credit spreads above 500 basis points, a broken macro thesis, and a market fall of more than 10 percent in a week. The Parts carry these conditions only in outline, as the tripwires of step 6 in Part 1 and the protocols of Part 5; the full list, with every threshold, is only here.

The correlation condition is the Correlation Spike tripwire. It fires when the average pairwise correlation of all your non-cash holdings, not Torque alone, rises above 0.70, and it reports unavailable rather than calm unless every holding has at least 60 daily returns of history.

One holding sits outside most of this machinery. Bitcoin is left out of every concentration layer and every Next Dollar addition, and it is scored on a route of its own, which comes next.

Because Bitcoin has a scoring model of its own, the first question the system answers is what counts as Bitcoin. There are four categories, and they are not interchangeable.

Market capitalization resolves down a chain: network data, then a quoted figure, and only when both are missing, the last known price × 21,000,000. That last rung uses the protocol cap, not the roughly 20.1 million coins in circulation (as of September 2026), so it reads about 4.5 percent high. A stale fundamental is never substituted for a missing one: the chain prefers null to wrong.

Bitcoin runs through the same logarithmic headroom curve as every equity; only the addressable market differs. Absent a research profile, the baseline is a conservative monetary total of roughly 11.5 trillion dollars, well below the estimate the doctrine works with in Part 3:

headroom = 35 × ln(1 + H) / ln(31), where H = min(TAM ÷ market cap, 30)

On September 29, 2026 Bitcoin’s market capitalization was about $1.68 trillion, a headroom ratio of about 6.8×, which scores roughly 21.0 out of 35: a strong reading, not a maximal one. At $3 trillion the ratio falls to about 3.8× and the score to roughly 16.1, because a rising price uses up the headroom it is measured against. The small pool is chosen on purpose: the default score should not depend on the most optimistic version of the thesis being right. When a research profile supplies its own addressable market, headroom uses that figure, and convexity stops paying the 2-point baseline penalty in the CIS cascade.

Network convexity decays as adoption progresses:

network convexity = 18 × max(0, 1 − penetration^0.6), where penetration = market cap ÷ $11.5 trillion

Penetration always divides by the $11.5 trillion pool, even when a research profile supplies the headroom figure. At $1.68 trillion network convexity is about 12.3 of 18. A fixed monetary-convexity score (5 to 7 points, and 7 for Bitcoin) completes the 25-point optionality sub-score.

Separately, the dashboard displays a scenario that builds an addressable market from three pools (40 percent of above-ground gold, 5 percent of global real estate, and 5 percent of emerging-market broad money, taken as 40 percent of the world’s) and divides by the 21 million cap to imply a price. That panel is a display scenario, not the scoring input. It is labeled as one, and its only live input is the gold price.

The power-law overlay on the price chart is modeling, not scoring: a trend of 2.88 × (days ÷ 1000)^5.82 and a floor of 1.2828 × (days ÷ 1000)^5.928, where days count from the genesis block on January 3, 2009. On September 29, 2026 (day 6,478) the trend read about $152,000 and the floor about $82,900, so Bitcoin, at about $83,600, sat just above the floor and about 45 percent below trend. Despite its name, the floor line does not hold price up: closing prices have fallen below it several times, most recently for most of February to mid-September 2026. The dashboard draws only these two lines, with no euphoria band. The accumulation pace in Part 3, set by its valuation models, is written guidance you apply; no code triggers on either line, and nothing in the software paces purchases.

The projection model is held back (Scope), but one rule of its contribution accounting is published, because it keeps added money from posing as performance. Scheduled contributions are simulated as a daily flow, with no weekly or monthly step at all:

daily contribution = monthly amount × 12 ÷ 365

Contributions land after each day’s return and consume no randomness, so the same simulation run with and without contributions draws identical market paths. That keeps the two headline numbers separate:

contribution effect = wealth growth − market gain

Expected compound return and probability of loss are computed on the market-only path. Money you added is not performance, and it is never allowed to flatter a return figure: a simulation with zero market return and steady contributions shows wealth rising and market gain at exactly zero.

Allocation is measured against the portfolio’s total balance, Bitcoin value ÷ total portfolio balance × 100, counting native holdings only by default. The mode that also counts spot wrappers needs the wrapper list to resolve, and falls back to native-only when it cannot. Bitcoin is excluded from every concentration measure. At setup the dashboard proposes a 15 percent Bitcoin baseline (10 percent under the Conflict Economy and Re-industrialization theses), which you confirm or replace with your own conviction. Its construction score carries a 10 percent Bitcoin posture reference within a 5 to 15 percent band as reference metadata; it costs no points and never triggers a sale. The reserve ranges themselves are doctrine, set out in Part 3.

Wherever Bitcoin and everything else are held, the ledger has to say what each holding cost and what each sale realized. That is the next chapter.

Wrappers decide what you keep, so it pays to be exact about what the dashboard computes here and what it leaves to you and your tax preparer. Most of what it computes is ledger accounting, and this chapter spends as much time on what it leaves out.

What is computed is accounting: cost basis, realized gain, holding period and the effect of a return of capital. Those are facts about your ledger, not statements about your tax return.

Wrappers are Roth, Taxable, Pre-tax, Bitcoin and Unknown. The dashboard marks an account it does not recognize as Unknown rather than guessing Taxable, because guessing the wrapper is guessing the tax character of everything in it. Not every part of the dashboard meets that bar yet: a few older ones still assume Taxable when the wrapper is missing (the positions table, for example), and a trade record has no Unknown option.

Cost basis is total dollars at the position level, and per-share only inside a lot, where it is recomputed at disposal from what remains instead of being stored at purchase. Since September 25, 2026 a recorded purchase commission is added to the lot’s cost, and a recorded sale commission reduces proceeds and realized gain. A trade with no commission recorded counts as unknown, not as zero: the dashboard withholds the cost basis and unrealized gain of the holding it opened (for a sale, the proceeds and realized gain) and asks you to enter the commission, 0 if there was none. That follows the IRS rule that the basis of stock you buy generally includes costs of purchase such as commissions (Publication 551).

Lot method is set per account, and specific identification picks lots sale by sale. First-in-first-out takes the oldest lot first, last-in-first-out the newest, highest-in-first-out the highest per-share basis (ties go to the older lot), and average cost relieves every open lot in proportion. An account can also mirror Schwab’s Tax Lot Optimizer. With no method set, the ledger uses average cost, and a method it does not implement is refused rather than booked as average cost. The IRS allows average basis for mutual fund shares and treats unidentified shares of stock as the ones you bought first (Publication 551), so check the method against your broker’s before comparing realized gains with a Form 1099-B.

Holding period follows the calendar: a lot is short-term through the one-year anniversary of its acquisition and long-term from the day after, which is the IRS test of more than one year (Publication 550). A lot bought on January 31, 2025 turns long-term on February 1, 2026. Where an acquisition date cannot be established, the term is reported as unknown rather than assumed. One summary card on the Data page still counts elapsed days, so it can differ by a day across February 29. The trade record labels the term advisory; read it as a reading of your ledger, not tax advice.

A distribution is split by its return-of-capital fraction p. If p is 1 the entire gross amount is return of capital, with no companion dividend. Otherwise:

return of capital = gross × p, rounded to the cent and taxable portion = gross − return of capital

The two sum to the gross by construction. The return-of-capital portion then reduces basis per share across open lots, floored at zero so a lot never goes negative; any amount beyond zero basis is recorded as a gain on that lot, which is how the IRS treats it (Publication 550). Reinvestment runs after the reduction, never before.

The accounting stays deliberately flat: a return of capital reduces cost basis and is not taxed when received. Whether a distribution counts as return of capital is set each tax year and reported on Form 1099-DIV; the ledger records the fraction it is given, and the arithmetic stops there. The strategy around it is in Part 4.

Three more absences are worth naming, because people often assume them. There is no wash-sale engine: no substantially-identical matching, no replacement detection, no disallowed-loss carryover. (Where a broker supplies a lot’s realized gain, the ledger uses it as given, and that figure can include the broker’s own wash-sale adjustment.) There is no Roth conversion, rollover or required-distribution feature. And a transfer between accounts carries no economics: it moves at a price of zero, keeps basis and the original acquisition date, and never realizes a gain. A move that may have tax consequences is recorded as a plain transfer, and the dashboard does not classify it.

Every figure in this chapter is only as good as the data under it, which is where the next chapter goes.

Every number on this page rests on inputs of uneven quality. Four different things often travel under the one word “confidence”; here they are separate quantities, tracked separately.

A fifth state sits off the ladder entirely. Unknown is not low. It is the absence of a rank, and it is carried that way instead of being folded into the bottom tier, because “we do not know” and “we know it is bad” are different claims about the same position.

Confidence reaches the score through two channels. The delta clamp limits how far a score may move in one update: 3 points at low confidence, 5 at medium, 6 when the evidence is proxy-derived, 8 at high. And when the macro component rests mostly on heuristics, its weight is cut to 60 percent (low) or 80 percent (medium) of its value, with the difference split evenly between convexity and risk. Beyond those two channels, confidence does not raise or lower the number. Two caps show how that works. An unclassified over-the-counter venue forces confidence to low, and a market capitalization under 50 million dollars caps it at medium. Both lower the confidence tier and leave the computed score alone. Through the clamp, the lower tier then limits how far the score can move on its next update.

An evidence-adjusted view of a score discounts each component by how directly its inputs were observed: nothing for direct company-specific evidence, 5 percent for direct category-level evidence, 15 percent for proxy-derived inputs, 35 percent for heuristics and 50 percent for missing data:

evidence-adjusted score = max(0, CIS − Σ points × weight × discount rate)

The adjustment runs one way. It can only reduce, and no path lets good-looking evidence lift a score above what the components produced.

Where a language model contributes to a classification, its own stated confidence is one factor among four and never the largest: model self-report 0.30, evidence completeness 0.30, persistence across runs 0.20, agreement with the deterministic classification 0.20. The self-report also passes through a compressing curve with a hard ceiling of 0.85, so a model, on its own testimony alone, can never reach full confidence. A language-model classification must clear a trust-corrected score of 50 before it may be written at all.

A model estimate of a data field is held to the same standard. It never counts as observed coverage in the confidence a score reports (it does count as present for the coverage cap under CIS, where its admission weight discounts it). It is bounds-checked per field and rejected if implausible, and it expires after 30 days. It fills a field only after every provider and derived path has come back empty, and that includes market fields such as three-month momentum, 30-day volatility and trading volume. Those fill the display; for the score, the admission weights refuse a model estimate of momentum, volatility, volume or drawdown (over-the-counter market cap and volume excepted), and the score falls back instead.

Where a research claim becomes a catalyst probability, the mapping is bounded and labeled for what it is: probability = clamp(confidence ÷ 100, 0.20, 0.85). That is a heuristic translation. It has never been fitted to outcomes, and it is not a statistical probability.

The last chapter turns to a score that answers one narrow question: where the next dollar should go.

The position score asks how good a holding is. The Next Dollar Score asks where, given what you already own, the next dollar would do the most work. It is a ranking instrument, kept separate from CIS by design, and it reads the wrapper-adjusted CIS rather than the canonical score.

Six components, at these base weights:

The market environment then scales expression purity and posture alignment, cuts momentum to 60 percent of its weight in a defensive environment, and renormalizes the weights to sum to one. Recompute a score from the base weights alone and you will not match the dashboard.

Size opportunity decays exponentially with what you already hold:

size opportunity = clamp(round(90 × e^{−0.18 × allocation%}), 0, 100)

An untouched name scores 90. At 5 percent it is 37. At 10 percent it is 15, at the 15 percent default cap 6, and it rounds to zero from about 29 percent. The curve is the arithmetic form of a plain idea: the marginal dollar is worth less to a position that already has plenty.

Two components condition on whether the thesis survives, which the engine reads as a CIS of 55 or more. A three-month pullback of 5 to 20 percent in an intact position raises the momentum reading (65 up to 95); the same pullback with a broken score drops it to 35; a fall of 20 percent or more scores 30 either way. An unrealized loss deeper than 30 percent scores 60 when the thesis is intact and 20 when it is broken. Each of these components carries a base weight of 0.10, so the swing is a few points of the final score. Large unrealized gains only lower the reading; they never make a position more attractive to add to.

Seven bounded modifiers then adjust the composite: opportunity cost against the best available peer, a Ballast throttle in risk-on and selective environments, a Ballast boost that replaces it in a defensive one, an expression-purity penalty, a score-trend adjustment bounded to ±8, a readiness penalty capped at 12, and a volatility adjustment between −10 and +6 that reads each name’s trading range against its own baseline and the room left to its invalidation level. A positive trend bonus requires a score of at least 40, so a weak or missing score can never be lifted by its own trend line.

Above 15 percent of the portfolio, the dashboard stops recommending additions. Between 10 and 15 percent it raises a caution and keeps recommending: the 10 percent line is advisory, not a freeze. A score below 40 rules a position out of an add recommendation, and so does a score the engine has marked unreliable. A missing score does not: the position still receives a next-dollar reading, computed with a neutral stand-in of 40, and carries a “cannot fully evaluate” marker. The gate sits below the framework’s allocation floor of 50, so a position scoring 40 to 49 can still show an add reading while the sizing ladder gives it no band. Cash is never recommended for additions, and neither is Bitcoin, which is managed at the wrapper level rather than position by position.

None of this stops a trade in either direction; the dashboard records every trade and never blocks one. On the trim side the score raises cautions only (a position above 15 percent, or a score below 40), and no state rules out a trim.

A position ruled out of additions still receives a full score and keeps its place in the ranking.

Bands are 75 and above for strong, 55 for moderate, 35 for weak and below that avoid, each lower bound inclusive. They order advisory copy and size nothing, and they cut at different points from the CIS bands.

Rotation pairs a bottom-quartile candidate with a top-quartile one, requires a score gap of at least 15, emits at most five pairs, and needs at least four scored positions after cash and Bitcoin are set aside.

Published here: the position score, the construction score, score to size, governance, the Bitcoin backbone, wrappers and basis, evidence and confidence, and the next dollar.

Held back on purpose: the projection models and their parameter estimation, earnings and forward valuation, performance accounting and net asset value, and margin mechanics. Two small pieces of projection are published because they bound what a projection may claim: the FIS scaling in the construction chapter and the contribution-accounting rule in the Bitcoin chapter. The rest are modeling and accounting surfaces where a published formula reads too easily as a forecast or a claim about results, and they will appear here only by a specific decision to publish them.

Within the eight chapters the same rule applies at a finer grain: the page publishes formulas and thresholds, not source-code locations or the logs that verified them.