0004 — No sleeve is promoted; the portfolio is the control alone

Status: current only as a dated publication conclusion, not as a research constraint. Candidate status belongs to the generated programme report and current syntheses. The zero-leverage portfolio is a reference recommendation conditional on missing investor and financing inputs; financed constructions remain open to measurement. See decision 0010.

Date: 2026-08-12. Status: accepted. Amended 2026-08-12, 2026-08-16 and 2026-08-17 as Experiments 005, 013, 014 and 015 landed; the non-promotion itself has never changed. Decision 0009 lifts the step 6 and step 7 blocks this record argues against itself about below, and leaves the non-promotion standing. Decision 0010 requires that a sleeve verdict name the funding rule that set its hurdle, and reads the promotion conditions below as gates on promotion and publication only — none of them bars measuring, costing or pricing a candidate. Supersede rather than amend when the first sleeve is promoted.

Context

A research programme with this much apparatus invites the assumption that the apparatus produced a portfolio. It has not.

The per-hypothesis status table that used to sit here is deleted. It was a snapshot that was deliberately not refreshed, and a decision record holding a stale copy of generated facts is a status journal wearing an ADR’s clothes. The ledger is the only thing that knows what was run:

cd research && uv run python -m portfolio_edge.reporting.programme_status

The per-candidate reading is on the design map; the instrument’s limits are in the evidence base.

What matters for this decision is the shape rather than the counts. Every terminal status is rejected, unresolved, or capped at exploratory by the fund-level data contract (decision 0002). Nothing has reached walk-forward-tested, shadow-live or production-eligible, and nothing could have. Where a row is unresolved the window could not detect the effect — which is not the same as the effect being absent, and decision 0009 now requires that distinction to be carried in the verdict rather than in a footnote.

Recording this is not bookkeeping. The failure mode of a research programme with this much apparatus is that a reader assumes the apparatus produced a portfolio.

Decision

No candidate sleeve is promoted. The research portfolio is the cheap broad-market control alone (decision 0003), plus the deterministic cost, tax-location and do-not-trade discipline that the edge decomposition prices at about 89 bp/yr against the investor’s own counterfactual.

Consequently:

• Leverage stays at zero. It was conditioned on an unlevered edge surviving the protocol. None has, so there is nothing to lever. The reasoning is contested and the contest is derived, not rhetorical — see the consequences below, where the hurdle a sleeve must clear is shown to move by a_p - sigma_p**2 with the funding rule. The rule stands until a decision supersedes it; what has changed is that its cost is now a measured number rather than an assumption.
• Rebalancing is retained as risk control and forbidden as return. It held exposure within 0.6 to 3.1 percentage points of target against buy-and-hold’s 14.8, for 0.3 to 1.2 bp/yr. Anyone who wants their declared allocation to remain their actual allocation should rebalance. No rebalancing-bonus feature may be built.
• The forty-eight exploratory US products and twelve ex-US products may be used as implementation proxies in a later experiment and for nothing else — and on the ex-US side only eight of the twelve survive a fair comparator. That list was fifteen US products until 2026-08-17 and the change is a frame correction, not a new result: Experiment 002’s numbers reproduce to zero difference on every fund it audited. The shortlist now includes every systematic value and small-value product on the US shelf — AVUV, AVLV, AVSC, DFAT, DFSV, DFUV, DFLV, DFAS — which the old frame excluded by construction. exploratory still permits testing and nothing else, and none of them has an alpha its own window could measure. The count is 48 on the frozen comparator and 47 to 49 on comparators that can express what these funds deliver, so the number quoted must always name its basis. The single US product whose status does not survive a fair comparator is IWN, replaced at a corner by a small-value index fund at a fifth of its fee; all nine systematic value and small-value products survive every basis tested (Experiment 014). The ex-US list is the one that shrinks. Of its twelve, AVDV, AVIV, DFIV, DISV, IDMO, IVLU and SCZ survive every basis tested, EFV survives all but the one that hands a second EAFE value fund to a fund that is EAFE value, and IMTM, FNDC, SCHC and DFIS do not survive at all: each loses to a cheaper fund in its own cell once the basis carries one — IMTM to IDMO at 0.25% against its own 0.30%, the other three to a developed-ex-US small-value column (Experiment 015). The seven survivors have since been decomposed into every leg on their own panel, and the two small-value ones — AVDV and DISV — carry SMB loadings of +0.671 and +0.431 against −0.11 to −0.29 for the four large-value funds, with AVDV carrying RMW +0.386 besides. Both legs are exposures to premia this repository cannot sign, so the ex-US shelf reproduces the US shelf’s ordering: large value ahead of small value, from independent data (factor products). Nothing about that promotes the large-value funds; two of them carry raw alphas their own windows can measure and both are negative.
• No number from research/ may appear in the shipped application as a finding. Lifted by decision 0007, which replaces the ban with four constraints. Everything else in this record stands in full, including the non-promotion itself.

The conditions that would change this

Each is a measurable target, not a hope. A sleeve is promoted only when its own row’s condition is met and it beats the control on the terms in decision 0003.

The chain matters as much as the rows. What a shareholder receives is premium × delivered loading − cost. Experiment 002 measured the second and third and found the loading delivered and the cost measurable; Experiment 001 could not sign the first for any factor. No product can be promoted while its underlying factor is unsigned, which is why every product row below points back at a factor row.

Experiment 005 has run and did both things at once. Value advanced; profitability and investment were closed on public data. The next step was recorded here as the measurement of the long-only capture fraction. Experiment 007 took it, and the answer removed the term rather than filling it in: a capture fraction is an HML loading measured a second way, so the chain for value is premium × (fund loading − incumbent loading) − cost and every term in it is already measured (Experiment 007). The immediate next step is still not a promotion attempt and still not the purchase. It is the prior-window replication the frozen protocol asks for, which no experiment here has fitted.

Alternatives considered

Promote the trend sleeve as exploratory and build it. Rejected. Its own specification caps a vendor-series evaluation at exploratory, its falsifier fired, and the vendor states no cost basis anywhere in the archived workbook, so every figure is gross of the vendor’s own trading costs by omission.

Promote RMW on the grounds that it is the only factor that did not decay (96% retained). Rejected then as a prioritisation rather than a finding, and now closed outright: pooled across three regions its premium is +2.53 pp/yr against its own measured 2.62 pp/yr detection threshold, 62% of its US premium is the single year 2021, dropping the pooled best year takes it to +1.79, and its volatility carries an unresolved ±5.09% systematic band from Phase 1 (decision 0005).

Promote value now that its premium is signed. Rejected. exploratory is the lowest rung of the ladder and permits an implementation to be tested, nothing more. The premium is gross, long-short and not investable; what a long-only fund delivers of it is a measured loading of +0.537 rather than the whole of it; and its pooled figure is carried by the two non-US regions, with the largest leg in emerging markets where shorting is hardest and dearest. The US leg alone is +1.57 on an interval of [−2.28, +5.54], and on that premium a 20% tilt’s contribution to geometric growth is negative at every weight.

Report the 2000–2019 era, in which annual rebalancing was worth +0.575 pp/yr and cleared the materiality threshold twice over. Rejected, and the rejection rule was frozen in advance precisely so this could not be reported as a finding: it is one twenty-year window inside a thirty-five-year sample, bracketed by two windows of the opposite sign.

Say nothing and leave the absence implicit. Rejected. An unrecorded non-promotion decays into an assumed promotion.

Consequences

The deliverable of this research programme is a design map and a control, not an allocation. An allocation becomes appropriate only after the investor policy is defined — benchmark, horizon, tax status, liabilities, cash flows, drawdown tolerance, liquidity reserve, permitted instruments and objective — which remains open decision 1 in the research framework.

Steps 6 and 7 of the framework’s build order — portfolio combination, then fractional/risk-constrained Kelly and leverage — are blocked, and not on effort. Step 6 combines sleeves and there are none; step 7 sizes an edge and there is none.

Step 6’s block is contested and should be lifted. The reasoning above is circular as applied to the construction tournament, which compares weighting methods — market weight, equal weight, inverse volatility, constrained minimum variance, linear-shrinkage minimum variance, equal risk contribution — on the three regional equity sleeves and the cash leg that already exist. It does not need a promoted sleeve, and it is the only designed experiment that treats a portfolio as a joint object rather than as a base plus a marginal addition. The argument is in search coverage §5, together with the observation that Experiment 010’s portfolio-level closure turns on its 10% reference weight rather than on evidence.

Step 7’s block is circular in the same way, and this is now derived rather than argued. The block reads “it sizes an edge, and there is none”. That is true of leverage applied to a sleeve whose value has already been established. It is false of the prior question, because whether a sleeve has positive marginal value depends on the funding rule, and the funding rule is what step 7 forbids. From studies/overlay_growth.py, which is closed-form algebra pinned by tests needing no market data:

• Funded pro rata — sell the base to buy the sleeve, which is what every marginal experiment here has done — the first sleeve dollar must clear a_p - sigma_p**2 (1 - beta).
• Funded as a financed overlay — sell nothing, finance the notional, which is what a capital-efficient fund does — it must clear rho sigma_p sigma_d.
• The difference is a_p - sigma_p**2 = sigma_p**2 (L_p* - 1), in which every term involving the sleeve cancels. It is +2.44 pp/yr for a 100%-equity base at a 5.0% arithmetic excess and 16% volatility, and +2.08 for the 60/40 base Experiment 004 used.

So the zero-leverage rule is not a conservative simplification that costs a little return. It raises the hurdle every candidate sleeve in this repository has been judged against by more than two percentage points a year — larger than any premium the programme has attempted to measure, and larger than the entire measured edge budget. “There is no edge” and “no edge clears the pro-rata bar” have been used interchangeably, and they are not the same statement.

What follows, and what does not. This is an argument for asking the question, not for an affirmative answer, and three things stand in its way. Two are recorded here rather than left for a later reader to discover:

1. The cost stack, not the correlation, is what binds. Financing spread plus fund fee plus distribution tax character is on the order of 1.4 pp/yr in a sheltered account and 3.5 pp/yr in a taxable one on the placeholders currently available — against a post-publication trend excess return this repository measures at roughly 1.8 pp/yr. The overlay bar is near zero; the cost stack is not.
2. The honest control is leverage-matched, not unlevered. At matched volatility the variance terms cancel and the higher Sharpe ratio wins outright, so an overlay that raises growth over unlevered equity while lowering the portfolio’s Sharpe ratio has bought its gain with beta and must be labelled that way.
3. The instrument is weaker than the effect. A portfolio-level growth difference of the size at stake sits near the detection floor of a 35-year monthly window, so the experiment this unblocks is a sizing question with an interval, not a detection question with a verdict. Freezing it as the latter would repeat the error search coverage §1.2 exists to name.

Both blocks are lifted by decision 0009, and the non-promotion in this record is untouched by that. The step 6 tournament may run, and the funding rule may be measured with overlay funding as a primary arm against a leverage-matched control. What 0009 does not do is authorise leverage in a recommended portfolio; the zero-leverage default and the non-promotion both stand.

This record should be superseded, not amended, when the first sleeve is promoted.