The Target Book: Rebuilding the Ideal Portfolio Every Rebalance

TL;DR. Every rebalance, the engine asks one question: if we were building this portfolio from scratch today, what would it hold? It ranks the eligible universe, fills sector slots, and caps position weights into a target book — then diffs that book against what you actually hold, and trades only the difference: exit what no longer belongs, swap in what's meaningfully better, nudge weights back toward target, buy what's left over. A name that slips a few ranks keeps its seat inside a retention band, and a 60-day minimum hold protects new positions from being swapped out on transient score noise. None of that protects a thesis that's actually broken — an AVOID rating, an external veto, or a 40%-from-entry disaster stop all exit immediately, band or no band. Two gates block new buys before they start: chronic share dilution and low-quality earnings. Everything below is the reasoning behind those numbers.

The buy-and-hold failure mode

You bought X six months ago, rated STRONG — a clear top-of-book name. It was one of the best ideas on the screen that week. Six months later the composite has drifted: earnings quality slipped, the sector re-rated, the discount closed. X is now FAIR — not broken, just no longer standout. Meanwhile Y has entered the universe rated EXCEPTIONAL. Y wasn't available when you bought X. Y is available now.

In a pure buy-and-hold framework, you do nothing. You sit in X until its thesis actually breaks. The opportunity cost of holding FAIR while EXCEPTIONAL is available never shows up on a statement as a loss, but it is one — it's the return you didn't earn.

Every rebalance has to answer the same question, for every position at once: is this still the book we'd build today?

From patching positions to rebuilding the book

The obvious way to answer that is pairwise: find your weakest holding, find the strongest available challenger, swap if the gap is wide enough to be real. It's tempting because it's a small, local decision — one holding, one challenger, one number to compare.

It's also the wrong level to reason at. A pairwise comparison only ever sees two names at a time. It can't see that the sector is already full, that several "weakest holding" candidates all want the same replacement, or that the rest of the book has drifted from its target weights while attention was on the one swap in front of you. Those are portfolio-level facts a two-name comparison is structurally blind to.

Tessera's engine now answers the question differently: it builds the entire portfolio from scratch, every rebalance, as if opening a fresh account today with today's scores and today's regime. Call that the target book. Then it compares the target book to what you actually hold, and trades only the difference. The book is rebuilt completely; the portfolio moves only partially, and only where the diff says to.

Building the target book

Three steps, in order, nothing else in between:

1. Eligibility. A name has to clear a rating floor — STRONG or better, on Tessera's five-tier scale (EXCEPTIONAL, STRONG, FAIR, WEAK, AVOID) — to be considered at all. No separate momentum cap, no sector boost, no bonus for a "validated" thesis. The rating already prices in quality, growth, and the regime's own caps upstream.
2. Rank. Eligible names sort by composite score, highest first.
3. Sector slots, then weights. Walk the ranked list and fill seats, skipping a name once its sector already holds its share of the book (5 of 20 seats per sector, by default). Once the book is full, size each position: higher-scored, lower-volatility names get proportionally more capital, up to a hard per-name cap; anything trimmed by the cap redistributes across the names still under it.

Held positions are not an input to this step, deliberately. The target book doesn't know or care what you currently own; it answers "what's the best 20-name portfolio right now," full stop. What to do about the gap between that answer and reality is a separate decision, covered next.

Tessera's target-book policy (exact thresholds)

The default policy's construction, churn, and exit knobs — the same numbers that get hashed and stamped on every order plan this engine produces.

What the band and grace zone actually do

The swap band and the retention band look similar — both are score gaps — but they answer different questions.

The swap band governs offense: how much better does a challenger have to be before it displaces a name you hold? At 0.10, the gap has to be real, not noise — the composite score moves week to week on ordinary data updates, and a band that fired on every small fluctuation would just be an expensive random walk.

The retention band governs defense: how far does a held name have to fall before it's cut loose for no longer making the fresh cut? At 0.15 below the book's weakest current member, this is deliberately generous. Without it, a name oscillating right at the rank-20/rank-21 boundary would get exited and re-bought every few weeks as its score wobbles a little either side of the cutoff — churn with no signal behind it.

The minimum hold (60 days) is a third, independent brake, and it protects something the bands don't: a genuinely new position, whose score can move on the next data refresh before its thesis has had time to play out. It does not protect a broken thesis — AVOID, an external veto, or the disaster stop still fire on day one if they need to. Minimum hold only blocks relative-rank exits, never thesis-broken ones.

Five kinds of trades, not one

The diff between the target book and the real portfolio produces exactly five kinds of orders, evaluated in this order:

• EXIT — thesis broken. A held name rated AVOID, flagged by an external veto (the monitoring agent's critical red flag, when enabled), or missing a score entirely (delisted, filtered out, a data gap) is exited immediately. No band, no minimum hold, no exception.
• EXIT — fell out of book. A held name absent from the fresh target book, whose score has fallen more than the retention band below the book's tail, exits. Everything inside the grace zone holds.
• SWAP — atomic pairs. A surviving held name and an unheld book entry are paired and traded together, sell and buy as one unit, once the challenger clears the swap band. The two legs move together, so a swap can't quietly leave the book smaller.
• TRIM / ADD_ON — weight alignment. For names staying in the book, actual weight drifted more than the weight-trim band from target gets trimmed or added to. One rule, both directions.
• BUY — leftover seats. Whatever's left in the target book that isn't held and wasn't already paired into a swap gets bought outright, funded by available cash.

Thesis-broken exits fire no matter what; rank-based exits and swaps respect the bands and minimum hold; trims and buys fill in around what's left. A quiet rebalance produces a short list of small trims; a rebalance after a genuine regime shift or a wave of rating changes produces a longer one. List length is just an honest readout of how much actually changed.

Two gates before a name can be bought

Two more checks run before a name is eligible to enter the book as a new buy. Both reject; neither forces an exit.

• Share issuance. A candidate whose net share count has grown more than 35% over the trailing five years — the Daniel-Titman signature of a chronic diluter — doesn't get bought. Existing holders aren't sold on this basis.
• Earnings quality. A candidate whose percent accruals — the trailing gap between reported net income and operating cash flow, divided by net income, in the tradition of Hafzalla, Lundholm, and Van Winkle's accruals research — exceeds 1.0 doesn't get bought either.

Both gates are lenient on missing data (a young listing without five years of share-count history passes through), and both are quality filters, not scoring inputs — a name can fail one and still rank well on the composite; it simply won't be purchased until it clears the gate. A held name that starts drifting on either metric after purchase isn't automatically cut — only a rating drop to AVOID, an external veto, or the disaster stop removes an existing holding.

Regime: cash, not a freeze

The old system froze trading entirely in a crisis. The target book doesn't — it gets more conservative by holding more cash instead.

Regime scales exactly one thing: the deployment target. At NEUTRAL, BULL, and STRONG_BULL, the book targets its full 95% gross exposure. In BEAR, that drops to 75%. In CRISIS, 60%. The mechanism is unchanged in every regime — same eligibility floor, same ranking, same bands — but the invested fraction shrinks mechanically as conditions worsen, and the freed capital sits in cash by construction, not as a separate decision.

Regime already shapes the composite rating upstream — the classifier's own caps tighten what can score well in a downturn — so the deployment scalar is a second, independent lever, not a duplicate. Risk exits don't wait for calm weather: AVOID ratings, vetoes, and the disaster stop fire in every regime, CRISIS included.

Calibration, not vibes

The swap band, the retention band, and the minimum hold are not round numbers picked because they felt right — they came out of a parameter sweep across multiple market windows, the same walk-forward discipline behind every calibration decision on this site. We're explicit about the limits of that exercise: a sweep tells you what worked in the data you had, not what's guaranteed to work next.

What's new, and worth naming on its own: every resolved policy — every knob in the table above, after any per-strategy overrides — gets hashed, and that hash is stamped on every order plan and audit row the plan produces. A trade made months from now under a slightly different band setting is provably traceable to the exact configuration that produced it.

Failure modes (honest caveats)

No portfolio-construction rule is a free edge. Here's where this one can go wrong:

• The retention band and the minimum hold are generous by design, which cuts both ways. A merely mediocre held name — not thesis-broken, just no longer top-tier — can occupy a seat for weeks before the gap to the book's tail crosses 0.15. A fresh position whose score is quietly deteriorating, without crossing into AVOID, gets the same 60-day grace as a fresh winner. Fewer noise-driven exits, but a slower response to real, gradual decay.
• Sector slots can force average names into the book. If a sector is capped at 5 seats and 8 names from that sector rank in the book's top 20, the 3 weakest lose their seats to lower-ranked names in thinner sectors — the price of the diversification the cap is buying.
• The issuance and accruals gates only block the door on the way in. Neither watches a position after it's bought. A held name can drift past either threshold and stay in the book on that basis alone — only the rating, an external veto, or the disaster stop can remove it.
• Tax drag is real and mechanism-agnostic. SWAP and TRIM both realize gains or losses in a taxable account. US short-term gains are taxed as ordinary income — commonly 22–37% federal for individual investors. This system is most efficient in tax-advantaged accounts. This is not investment advice, and tax treatment varies by jurisdiction and situation.

Related reading

• Regime-Aware Investing
• Quality Screening Factors
• Position Sizing and Stop Losses