Convexity & Framework Integrity Scoring

The first five Parts set out what the framework believes and how a portfolio gets built. Part 6 adds the measurement rules: how the two scores are computed, what each penalty costs, and how often the loop runs. It does not change the construction rules of Part 4 or Part 5.

Rules that nobody measures drift into suggestions, so the framework measures two things. The Convexity Integrity Score (CIS) rates each position on its own. The Framework Integrity Score (FIS) rates how those positions have been put together. Each question belongs to exactly one of the two, and a weekly routine reads both. Position quality comes first.

The Convexity Integrity Score is a 0–100 score for a single asset, and it answers one question.

The canonical CIS looks at the position alone: it scores the same whatever else you hold and however concentrated the portfolio has become. There is one narrow exception, a fixed +3 on the carry sub-score for a return-of-capital preferred held as Ballast outside a known tax-advantaged account (about +0.75 CIS at the reference macro weight). Any portfolio-aware logic belongs to FIS or governance, never to CIS.

The number you see beside a holding in the dashboard is a second version. The framework’s governance rules call for a wrapper-aware CIS, so the dashboard multiplies the canonical score by a factor between 0.97 and 1.04, set by the holding’s account and its posture, and rounds it. This wrapper-adjusted CIS is the one the dashboard’s governance checks and its Next Dollar Score (its score for where new money goes) read. A Torque position scoring 69.5 shows 72 in a Roth and 68 in a taxable account, on opposite sides of the 70 line. FIS reads the canonical score.

The score is built from four components.

The 40/25/25/10 split is the reference weighting. An active thesis profile moves each weight by at most 0.10 (Fourth Turning runs 0.35/0.30/0.25/0.10); with no thesis selected, the dashboard does not pick one for you and uses the reference weights. It also moves weight off macro when the macro read is low or medium confidence, and scales convexity and macro by sector relevance; the Math page gives both rules. The four-component structure itself is doctrine.

Everything CIS leaves out has a home elsewhere.

The composite is carried as a float from end to end: a position scores 78.3, not 75 or 80. Some inputs are read off stepped ladders (momentum and turnover among them), but nothing rounds the stored score; only the wrapper-adjusted number shown beside a holding is rounded. A score that can only land on round tiers hides the small week-over-week movement that weekly tracking is there to catch.

Score updates are confidence-weighted, and each update is clamped to a bounded move: low confidence permits ±3 points, medium ±5, high ±8, and proxy-derived evidence ±6. At medium confidence a +10 update therefore moves the score 5 points. A material thesis change widens the bound to ±15, a first score is not clamped, and a few cases skip the clamp; the Math page lists them.

Why clamp at all? A score fed by the news cycle swings with it, and without a limit one enthusiastic week could rewrite months of accumulated conviction. The clamp makes moderate re-ratings, up to 20 points, earn their way across successive updates. A larger jump is treated as the model disagreeing with its own prior score: it lands in full and is logged as a model divergence for you to inspect, because a brake built to resist mood should not be able to suppress a real correction.

Nobody types in the confidence level. The dashboard derives it from the quality of the evidence behind each component, then caps it at low for an unclassified over-the-counter listing and at medium for a company worth under $50 million. The score says how sure its evidence lets it be, and you decide what to do about it.

Different kinds of asset cannot be scored honestly through one lens, so CIS routes each position to a scoring lane before it applies the rubric. There are six: Bitcoin, crypto, ETF, equity, private-structural and not-scorable. Separately, the dashboard labels each position with one of four archetypes (spot Bitcoin, ETF, levered ETF, or operating equity, the default), used mainly in FIS to keep Bitcoin out of its concentration measures. The archetype is labeling; the route is the lens.

Bitcoin runs on a monetary-TAM path, because it is a scarcity claim on a monetary market rather than a business with a balance sheet. The score does not size, gate or sell it. Part 3’s rules govern it: accumulate only, at a pace set by valuation discipline, outside the three postures and the concentration limits. Funds score through the same size-convexity curve as equities, on the weighted-average market cap of what they hold instead of 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.

The archetype puts no floor under the number. After the weighted sum and the delta clamp come a few named, bounded adjustments and ceilings, among them a coverage cap of 75, 68 or 60 when evidence is thin and an 85 cap when the score leans on proxies. Each is recorded in the score’s breakdown; the Math page walks the order. There is no cash-like archetype either: an asset held for stability is governed as Ballast under Part 5’s posture rules, not scored on a separate rubric here.

Scores fall into four bands. Seventy and above is Strong; 60 up to 70 is Moderate (so a 69.9 is Moderate); 50 up to 60 is Caution; below 50 is Weak. In the dashboard the same bands carry the conviction tiers core, standard, starter and avoid; the Next Dollar Score’s strong, moderate and weak ladder cuts at different points, so read the two separately.

This register is shared: in the dashboard, CIS and FIS read these same four bands. On FIS, 70 is the action line; on CIS, it is the Strong band’s floor, a band boundary, not an action trigger. The second score grades how the positions are put together.

Ten positions that each score well can still add up to a bad portfolio: three of them half the book, nearly all of it in one kind of account, and half the money riding on theses nobody wrote down. The Framework Integrity Score exists to catch that, by measuring the assembly against the framework’s rules.

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). The caps sum to 80, so a valid FIS cannot fall below 20. A score built by addition invites argument about what deserves credit; a score built by subtraction has to name names. Every point below 100 traces to a specific departure from a specific rule, so the score reads as a repair list. Fix the penalty, recover the points; the score only has to be able to see the fix.

Governance and dead-capital charges are multiplied by the position’s share of the portfolio, counted between 0.2 and 12 percent, so a 6-point charge on a 5 percent position costs 0.3 points; the other three buckets charge flat amounts.

The exhibit below walks an example portfolio down from 100, one deduction per bucket, to 83, in the Strong band.

The deductions in the exhibit are illustrative; the table below sets out the rules. The dashboard computes them, but its score does not yet see every input: it does not read the thesis, distribution and other records you keep on a holding, so it bills every position as undocumented and unclassified (the Math page lists the missing inputs). For a well-documented book of ten or more names, the dashboard’s FIS therefore reads up to about 15 points below the score the rules give it: the full 10 in Complexity and up to about 5 in Dead capital.

The floor of 20 is only a bound. Value weighting keeps governance and dead capital each under 10 points for any book of fewer than 200 positions, so a portfolio of that size that breaks every rule still scores above 30.

Two accounting notes. Tax architecture is not a separate bucket, and since the score sees only the aggregate wrapper mix, Part 4’s placement rules stay your check: a book holding Torque that may rotate in taxable and return-of-capital Ballast in a Roth draws no wrapper penalty if its dollar split sits near 45/35/20. Part 4 routes both the other way. Those targets are FIS’s own: Part 4 sets no fixed split, so a book routed by its sequence, with little or no pre-tax, shows some wrapper drift here.

Posture drift is not a score bucket either. The dashboard reports it alongside the score, in percentage points from posture targets derived from your operating thesis (or a fixed default when no thesis is active) rather than the Part 5 ranges; its diagnostics warn at 10 points or more and fail at 20, and its Framework Rule Register grades the same drift at 5, 10 and 15 points.

In the dashboard, FIS below 70 produces a fix-the-top-penalty recommendation when you run the Weekly Review, and below 60 the three-level check (see the weekly workflow below) raises a framework-breach review with a 24-hour deadline. Neither blocks a trade.

The diagnosis comes from reading the two scores together.

CIS never reads FIS. FIS reads CIS only through its four bands: positions below 70 draw governance charges, the count of positions in each band feeds the Allocation bucket, and undocumented positions below 70 count toward Complexity. Under the FIS rules, twenty equal-weight positions scoring 65, each documented, with its distributions classified and the book on its wrapper targets, still lose 14 points (4 governance, 10 band distribution) and score 86. A 95-CIS position does not offset a poorly assembled book, and a well-assembled book does not lift a 65-CIS position.

Crossing the two scores at 70 gives four states, each with its own response, and a repair order where both fall short. A quality holding in the wrong container is a construction problem, not a reason to exit it.

CIS sizes a position within bands that Part 5 defines: posture sets the band table and the score selects the band within it. For Torque and Ballast, 70 or above supports core sizing, 60 to 69 standard sizing, 50 to 59 starter sizing (Ballast calls it marginal), and below 50 is not allocation-worthy; Hype is eligible from CIS 50, at 2 to 5 percent per position under hard caps.

CIS below 50 removes allocation eligibility: every posture’s band is zero, so the position takes no new capital and is exited or replaced, and reclassifying its posture does not restore eligibility. At 50 and above, exits are governed by posture rules, tripwires and thesis evidence, not by the score band alone. Three signals govern the response to score movement.

Signals only help if someone reads them on a schedule. The framework sets one.

The framework runs on a weekly pass, Sunday evening or Monday before the open: update CIS, calculate FIS, run the governance checks, act or deliberately hold, and log. Event-triggered updates fill the days between passes, and the pass is logged whether or not anything traded.

In the dashboard you run that pass from the Weekly Review page, which marks itself due seven days after the last completed review; nothing runs it on a schedule. It first requests a refresh of up to 10 stale scores (missing, 30 or more days old, or 7 or more days old on a position of at least 2 percent of the book, 1 percent at low confidence), never one you have overridden. It then reads both scores, computing neither itself, runs the governance checks, recommends actions by precedence and logs the review.

In doctrine the CIS station recomputes C, R, M and E for every position; the dashboard refreshes only the stale scores above, and its review reports what changed. The FIS station totals the penalties across the five buckets and ranks the attribution for repair. The governance station then checks earnings proximity (positions above 3 percent; the dashboard counts calendar days, the doctrine five trading days), momentum, tripwires and posture drift.

Momentum here means Part 5’s three dimensions: distance from the 52-week high, strength against the sector and the market, and breadth. The dashboard does not yet measure them, so its Momentum Death tripwire, meant to fire when all three break, reports unavailable rather than guessing; a moving-average and RSI screen is coded but not yet fed, and does not stand in for them. What the dashboard does watch is stress across your holdings as a group, through the cohort confluence tripwire Part 5 describes.

In doctrine, action follows a fixed precedence: a tripwire demands a response 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). Absence of triggers means hold. The Weekly Review ranks its recommendations the same way, except that first place goes to a finding from the dashboard’s three-level check. The review records where each of the framework’s tripwires stands, but those readings raise no recommendation of their own, and it cannot yet raise the earnings step because it receives no earnings dates (the Framework Rule Register flags that case instead).

The three-level check flags a position whose wrapper-adjusted CIS is below 75, a sector where more than half the scored positions sit below 80, and an FIS below 60 (the Math page has the full conditions). None is an exit signal and none trades, but in the Weekly Review each becomes a top-precedence recommendation, so a holding whose wrapper-adjusted score sits between 70 and 75 draws one every time you run it, even though it sits in the Strong band.

Logging closes the pass. The decision log keeps the last 52 reviews, a year at a weekly cadence. It records the positions whose scores changed, the FIS reading, the recommendations made or held back and why, or why none was needed.

Four events interrupt the cadence instead of waiting for it: a position moving more than 5 percent in a single day · material news · an earnings announcement · a CIS update from new research. A tripwire interrupts by definition. The dashboard detects only the first, at 5 percent or more, and lists it on its Governance page as detected, not yet evaluated; acting on it is yours.

Between triggers, frequency limits keep you from confusing activity with diligence.

The best case for the cadence is what happens to a portfolio when it stops.

Implementations rarely fail loudly. They degrade through patterns that send no obvious signal: a score nobody re-ran, an allocation the market moved without a single trade, a tax leak too small to notice in any given month. The exhibit below sketches what those patterns do to a healthy portfolio in the year after reviews stop.

The framework watches five patterns, each with a diagnostic: a threshold that turns a slow, invisible failure into an explicit flag. The Weekly Review runs all five diagnostics when you start it. Four can fire; correlation stacking has no input yet, so a portfolio tripwire watches correlation instead. Staleness, posture drift and large unrefreshed gains are also checked outside the review by the dashboard’s framework diagnostics, and the dashboard thresholds given below for those three are theirs.

Skip the recalculation and the story does the scoring. Each of these failures still rounds to healthy on any single day and shows up only in week-over-week tracking, so framework health has to be read as a trend.

The six Parts form one system. The framework sets the rules; you carry them out, and the weekly pass above is the work. None of it is hard on its own. All of it has to happen, every week, for the measurements to mean anything.

Three implementation paths carry that work.

Pick the one you will keep running for decades. The framework does not care which, only that the loop runs.