Position Pillars for the AI Transition

This is the portfolio side of the AI Economics framework. It is not a recommendation list, a trade idea, or a sell-side note.

I. From framework to portfolio

The main series ends with seven analytical commitments on the record. Each has a portfolio-relevant implication. The point of this section is not to translate each commitment into a vehicle — that is what §II is for — but to make explicit that the framework is structurally connected and the portfolio expression has to be too.

Rent capture mutes the Euler-g channel. Aggregate output rises faster than the marginal household’s consumption because the AI rent flows to capital rather than wages. A productivity-boom story holds simultaneously with anchored real rates and supported equity multiples. Duration and equity are simultaneously holdable.
The marginal pricer of rates is not the marginal pricer of equities. Vissing-Jørgensen’s 2002 heterogeneity result lets η > 1 hold in the rates regime while η < 1 holds in the equity regime. A rates blow-out does not automatically force an equity reset.
Reinstatement is weak. The Acemoglu reading wins the Chapter IV side-pick: AI substitutes for cognitive labour without rebuilding the labour-demand curve, because the physical-reorganisation channel previous general-purpose technologies ran through does not apply. Cognitive-labour-intensive sectors without pricing power get the wage curve without the productivity offset.
Substrate-lockdown durability holds. The non-rival capital layer — compute, advanced packaging, leading-edge logic, EUV lithography, high-bandwidth memory, frontier-lab incumbents — keeps the rent. The DeepSeek event tested this in January 2025 and the trade survived.
The Baumol window is bounded by manufacturing capacity, not by intelligence. Software AI runs on inference-cost curves that fall fifty-fold per year; robotics runs on manufacturing-capacity curves that scale by orders of magnitude per decade. Bodies-in-rooms work has unusual wage power inside the interval.
Capital share does not flow through to wages. Labour share falls. Consumption growth does not pick up the productivity gain. Broad equity returns concentrate in capital-share-holding firms.
The jagged frontier is real. Dell’Acqua’s in-frontier-vs-out-of-frontier asymmetry means AI is not a universal productivity multiplier. Firms that do not manage the boundary have hidden tail risk in their cost structure and their outputs.

The seven are not parallel. They are nested. Commitments (3), (5), and (6) are about where the gains do not go — labour, consumption growth, the marginal household. Commitments (1), (2), (4), and (7) are about where they do go — capital, substrate, the closed frontier, and inside the firms that know the boundary. The portfolio expression in §II respects the nesting.

II. The pillars

Four pillars. Each pillar’s micro-structure is the same: which framework commitment it expresses, the rationale in one line, the vehicle class, the time horizon, the explicit falsifier, the regime shift that would force a reversal.

II.1 Substrate-lockdown long

Framework prediction. Commitments (1) and (4): rent capture at the non-rival capital layer; substrate-rent durability.

Rationale. The most direct expression of the rent-capture-mutes-Euler-g argument. If the substrate does not diffuse, the rent compounds at the firms that own it.

Vehicle class. Broad semiconductor exposure (SOX-style baskets capture the rent diffused across the stack); concentrated single-issuer overweight on ASML (100 per cent EUV share, year-end-2025 backlog €38.8 billion, fully booked through 2027) and TSMC (sole 2nm foundry from January 2026, CoWoS capacity on track for 127k wafers/month by end-2026 with Nvidia booking over half); concentrated overweight on SK Hynix as the cleanest HBM pure-play (53–62 per cent share against Samsung 35 per cent and Micron 11 per cent); hyperscaler exposure through public-market parents (Microsoft, Alphabet, Amazon, Meta); data-centre real estate (Digital Realty, Equinix). The custom-ASIC layer is a substrate vehicle, not a challenger — Broadcom’s $20-billion-plus FY25 AI semi revenue and +106 per cent year-on-year Q1 FY26 growth all runs on TSMC, CoWoS, and HBM (Research Note 1).

Time horizon. Three to seven years. Resolves as the substrate-rent durability prediction (Overview #5) resolves; longer if the antitrust path stalls.

Falsifier. The Epoch Capabilities Index open-weight-to-closed-frontier lag compressing below six weeks for two consecutive measurement periods. Lag is currently approximately three months on average, down from more than twenty-four months in early 2023 (Research Note 4). Confirmatory secondary: an open-weight model overtaking the highest closed-frontier model’s score on Humanity’s Last Exam by at least five percentage points (current leader Claude Mythos Preview at 64.7 per cent; top open-weight scores trail by more than thirty percentage points).

Regime shift forcing reversal. Substrate-rent break-up via enforcement on the Microsoft–OpenAI, Amazon–Anthropic, or Google–Anthropic partnerships flagged in the FTC’s January 2025 6(b) report. Microsoft and OpenAI restructured their exclusivity in April 2026; Microsoft retains 20 per cent revenue share through 2030 and an IP license through 2032 (non-exclusive). The partnership did not break. The EU AI Act’s enforcement powers activate 2 August 2026; on typical enforcement cadence, the first material fine is not before late 2027. The second regime-shift vector is sustained margin compression at Nvidia or TSMC on customer-concentration risk — Nvidia’s top four customers were 61 per cent of Q3 FY26 revenue, against 36 per cent a year earlier.

II.2 Cognitive-labour-intensive low-pricing-power short

Framework prediction. Commitments (3) and (7): weak reinstatement combined with the jagged frontier. The textured part: bodies-in-rooms work is not the short — that wage curve is up inside the Baumol window. The short is cognitive-labour-intensive sectors lacking institutional pricing power: middle-tier professional services, commoditised content production, paralegal and translation and copywriting service businesses without distribution moats.

Rationale. Reinstatement-weak plus jagged-frontier means these firms get the wage curve from Chapter III without the productivity offset that would justify it. The three exposure indices agree on the band: paralegals, translators and interpreters, customer-service representatives, writers and copywriters, medical transcriptionists (Research Note 2).

Vehicle class. Robert Half (RHI) is the single cleanest public-market expression. 2025 revenue $5.38 billion, down 7 per cent from $5.80 billion in 2024. Q1 2026 revenue $1.35 billion, down 8.4 per cent year-on-year. Core Talent Solutions segment is the source of weakness. Kforce (KFRC) is partially clean — white-collar tech and finance staffing focus, Q1 2026 showed slight stabilisation after multi-year slowdown. ManpowerGroup (MAN) is not a vehicle for this pillar — its industrial-staffing mix has it growing in Q1 2026 and confounds the trade. Beyond staffing, the structural underweight extends to firms whose revenue depends materially on the agreement-set occupations as labour rather than as customer demand.

Time horizon. Two to five years.

Falsifier (operational). Primary: Stanford Digital Economy Lab Canaries in the Coal Mine 22-to-25 AI-exposed-occupation cohort recovering to within five percentage points of the late-2022 baseline for two consecutive quarterly readings by end-2028. The cohort sits at approximately 87 per cent of late-2022 baseline today — the headline 13-percentage-point gap. Recovery to within five percentage points of baseline is a meaningful regime shift, not noise. Secondary: a constructed BLS Occupational Employment and Wage Statistics aggregate of the agreement-set occupations (customer-service representatives 2.8 million jobs in 2024, paralegals 376,200, translators 75,300, writers 135,400, medical transcriptionists in continued decline — combined approximately 3.7 million workers) recovering to within five percentage points of the 2022 baseline on the same horizon.

Regime shift forcing reversal. A macro reinstatement surprise: net positive entry-level US payroll growth in AI-exposed occupations sustained over four consecutive quarters, or an aggregate productivity surprise large enough to break the post-2025 labour-share decline.

II.3 Duration and stock-bond correlation positioning

Framework prediction. Commitments (1) and (2): the Euler-g channel does not fire because consumption growth does not move through the wage channel; η > 1 / EIS < 1 at the margin in the rates pricer, η < 1 / EIS > 1 at the margin in the equity pricer.

Rationale. Real rates stay anchored and equities stay supported by the same rent-capture pattern. The trade is long the equilibrium and convex against the disequilibrium tail — if reinstatement fires, real rates rise on the Euler-g channel and both rates and equities fall together.

Vehicle class. Intermediate-duration government bonds as the equilibrium long — US 5–10 year Treasuries, German 5–10 year Bunds, equivalents in any G7 currency. Long-dated out-of-the-money payer optionality as the disequilibrium hedge — 5y10y or 10y10y payer swaptions at 250–350 basis points OTM, or listed long-tenor SOFR caps at high strikes. Equity exposure that does not rely on bonds for diversification in the tail — managed futures, vol-targeted factor strategies, or pillar-aligned equity (the substrate exposure from II.1, which is supported in both the equilibrium and disequilibrium readings, though for different reasons in each).

Time horizon. Three to ten years on the equilibrium thesis; the optionality leg cycles shorter.

Falsifier (joint condition). US 10-year TIPS yield trades above 3.0 per cent on a six-month trailing average and US labour share (BLS PRS85006173) rises to 55.1 per cent or higher on a quarterly print, both within twenty-four months of the May 2026 baseline — 1.98 per cent real yield and 54.1 per cent labour share, the lowest reading since the BLS series began in 1947 (Research Note 3). Either leg alone is insufficient. Real-yield-only rise can come from term-premium or inflation re-acceleration without the Euler-g mechanism firing. Labour-share rise on its own can come from a policy intervention that does not require g to move. Both together is the disequilibrium signature.

Regime shift forcing reversal. Yield-curve control or financial repression by the Fed or ECB. Neither central bank is signalling this in May 2026 — the Fed ended quantitative tightening 1 December 2025; balance sheet at $6.58 trillion; no FOMC discussion of long-end caps; market pricing no further Fed cuts in 2026. The ECB deposit facility holds at 2.00 per cent. Yield-curve control remains a tail scenario, not the base case. The second regime-shift vector is capital-share redistribution at sovereign-wealth-fund scale executed faster than the rent-capture argument assumes — Norway has the template; no other jurisdiction has moved.

II.4 Open-weight vs closed-weight asymmetry

Framework prediction. Commitments (4) and (7): substrate-rent durability combined with the jagged frontier. Capability commoditises faster than the substrate.

Rationale. The pillar is the symmetric short to II.1 — the explicit refusal to overweight application-layer exposure whose moat is the very thing the framework predicts will diffuse.

Vehicle class — narrower than it first looks. Most large-cap enterprise SaaS firms have non-model moats and are not in the short universe: Palantir (proprietary government and operational data; +70 per cent year-on-year revenue in Q1 FY26), Salesforce (customer-data flywheel plus Agentforce $800 million ARR), ServiceNow (workflow integration; Now Assist tracking toward $1 billion in annual contract value), Snowflake (data infrastructure), Workday (deeply embedded payroll-and-benefits-and-finance integrations), and the broader infrastructure-and-workflow incumbents whose defensibility runs on data, distribution, or compliance surface rather than on closed-model exclusivity. The genuinely closed-model-exclusivity-dependent application-layer firms are mostly private as of mid-2026 (Research Note 4). The pillar therefore expresses as a positioning rule rather than a short basket: an explicit underweight on pure-play model-licensing or model-exclusivity-dependent exposure as new public offerings price into the portfolio, and a labelled sub-allocation within II.1 — not double-counted, see §III.

Time horizon. Same three to seven years as II.1.

Falsifier. Same as II.1: the Epoch Capabilities Index lag below six weeks for two consecutive measurement periods, or an open-weight overtake on Humanity’s Last Exam by at least five percentage points against the closed-frontier leader.

Regime shift forcing reversal. A closed-frontier breakthrough the open-weight ecosystem cannot replicate within twelve months. A true multi-hour autonomous-agency capability tied to proprietary post-training data is the canonical example.

III. Sizing under correlation and reversibility

Two rules carry the section.

Rule 1 — correlation aggregates. Pillars II.1 and II.4 are two readings of the same underlying bet. Aggregate exposure across them; do not double-size. Pillar II.3 has anti-correlation with II.2 in the disequilibrium tail: if rates blow out on a reinstatement surprise, the short on cognitive-services-staffing names is on the wrong side of the same move — a Brynjolfsson-side surprise closes II.2’s thesis at the same time it fires II.3’s tail. Size pillars such that no single regime shift forces a portfolio-wide unwind.

Rule 2 — reversibility, not conviction, sets pillar size. Pillar size is a function of how fast the position can be exited when the prediction inverts. A position with a measurable falsifier reported on a quarterly clock (II.1 and II.4 share the Epoch lag; II.2 has the Canaries cohort) is more reversible than one whose falsifier is annual (II.3’s joint condition) or longer. Size accordingly.

Worked example. A portfolio sizes Pillar II.1 at 30 per cent of equity risk budget and Pillar II.4 at 15 per cent. The correlated-exposure rule treats II.4 as a labelled sub-allocation within II.1, not as an additional pool. Aggregate exposure remains 30 per cent of equity risk budget, with fifteen of those thirty points labelled as the II.4 trade specifically — the pure-model-exclusivity underweight and the new-issuance positioning rule. The label matters for unwinding. If Overview #5 falsifies (II.4’s primary trigger via Epoch lag), the fifteen labelled points exit; the remaining fifteen hold on substrate-rent durability that survives open-weight diffusion. If II.1’s broader thesis falsifies via the antitrust path or sustained customer-concentration margin compression, the full thirty exit. Two failure modes, two unwind paths, one aggregate sizing constraint.

Current pricing observation, not framework dependency. The MOVE index — bond-market implied volatility — closed 13 May 2026 at 69.63. The normal range is 55–130; below 60 is calm. Long-dated payer optionality is at the low end of its post-pandemic price range (Research Note 3). The implication for II.3 sizing is tactical: the disequilibrium-hedge leg can be sized heavier than the steady-state mix would suggest, because the carry cost is currently modest. This is timing, not framework. If MOVE re-rates to its 2022–2023 regime — 100 or above — the steady-state mix returns. Note it and proceed.

The general principle: pillar size is not a function of conviction in the framework. It is a function of how reversible the position is when the framework’s prediction inverts.

IV. Holding the position through the cycle — what 18 months to 5 years looks like

Framework-consistent positions can be uncomfortable for long stretches even when correct. The discomfort is not a bug. Name it honestly and design the position to survive it.

Drawdown sources, by pillar.

II.1 carries DeepSeek-shaped tail-event risk. The 27 January 2025 single-session minus 17 per cent on Nvidia is the precedent — $589 billion of market capitalisation lost in one trading day, the largest single-day loss in US market history. The substrate-rent thesis recovered the loss within a quarter, hit a new all-time high in July 2025, and crossed $5 trillion market cap on 29 October 2025. The trade survives such events only if it is sized to.

II.2 carries squeeze risk. A public reinstatement narrative — a quarterly BLS or Canaries surprise that runs against the Acemoglu reading — can move the cognitive-services-staffing names sharply against the short before the structural thesis is falsified. The cleanest single-firm vehicle (Robert Half) is down materially through 2025 but is the kind of name that bounces hard on any positive surprise.

II.3 carries multi-year negative carry on the long-dated payer optionality. The leg bleeds even when the equilibrium reading holds. MOVE at 69.63 makes the bleed modest in May 2026 (§III) but does not eliminate it. The duration leg pays in the equilibrium; the optionality is insurance with running cost.

II.4 carries the same drawdown profile as II.1, by construction.

Cross-pillar correlation. Stated explicitly in §III. II.1 and II.4 move together. II.3 and II.2 are anti-correlated in the disequilibrium tail. II.3 alone is closest to a diversifier inside the position set, and is the leg that carries the portfolio through II.1 drawdowns.

What ends the position before the framework resolves. The most-watched substrate name has eighteen months of one-sided insider selling on the public record. Nvidia’s Form-4 filings show a 15-to-0 sell-to-buy ratio over the trailing eighteen months — zero open-market purchases by insiders, $3.3 billion in executive sales in 2024, and more than $450 million in additional sales from executives and directors between December 2025 and March 2026. EVP Ajay Puri (head of global sales — the executive with arguably the best real-time visibility into enterprise AI spending) sold $148 million across three months. Director Mark Stevens disposed of more than $100 million on a similar clock. The visible signal looks framework-contradicting (Research Note 6).

The honest read is that the easy interpretation of this data is in fact framework-contradicting, and the cross-sectional test that distinguishes the easy read from the right read has not yet been run. The structural alternative explanations — pre-scheduled 10b5-1 plan executions at elevated stock prices, standard wealth-diversification mechanics at high single-name concentration — are plausible but unverified at the cross-section. The cross-section is what makes the signal informative. §VI Q5 names the methodology and the script-away framing.

A private allocator reading quarterly statements during a Pillar II.1 drawdown will see this Form-4 pattern and ask about it. The discipline is structural sizing, not arguing with the visible signal. The framework can be right and the position can still get unwound for reasons external to it. Career risk on a sell-side desk where the substrate chart looks wrong for three quarters running. Capital-allocation pressure on an institutional allocator whose LP base does not share the framework prior. Behavioural pressure on the same private allocator who reads the same quarterly statements.

The honest read. The framework-consistent position is right on a multi-year clock and uncomfortable on a quarterly one. Discipline is not conviction; it is structural. Size such that no single pillar’s worst quarter forces a portfolio-wide unwind, and pre-commit the unwinding logic in §V before the data is in.

V. Unwinding logic — what 2028 / 2032 falsifies

Held against the five Overview predictions. For each, two pieces: the resolution case at the prediction window, and the signal timeline — the data path that indicates falsification is likely before the window closes. The portfolio runs on the signal timeline, not the resolution.

Overview #1 — humanoid shipments under one million per year through 2030. Resolution: verified annual deliveries at end-2029 / early-2030. Signal timeline: combined verified-delivery rate across Tesla Optimus, Figure, 1X, Unitree, Agility crossing twenty-five thousand units per quarter for two consecutive quarters by end-2028, against a current ~1,500–2,500 per quarter (Unitree ~90 per cent of the run rate); or sustained fleet-scale unit-price decline of more than 40 per cent year-on-year for two consecutive years (Research Note 5). Pillar consequence: II.1 partly reverses (substrate-rent durability was wrong on the robotics side — the manufacturing-capacity gate turns out openable). II.2 closes (Baumol-window wage power on bodies-in-rooms work compresses).

Overview #2 — US public-sector real unit-labour-cost not falling year-on-year through 2028. Resolution: BLS Employment Cost Index for state and local government workers vs private-sector ECI through end-2027 reads. Signal timeline: state-and-local ECI growth below private-sector ECI by at least twenty basis points (year-on-year basis) for two consecutive quarterly readings. Currently state-and-local is at +3.5 per cent year-on-year against private +3.4 per cent — a 10-basis-point gap in the wrong direction for the falsifier. The inversion plus the ≥20bp threshold filters out noise. Pillar consequence: II.3 tightens but does not close. The public-sector fiscal-pressure leg of the inflation thesis weakens; the duration position holds on the Euler-g-channel-muting logic.

Overview #3 — Danish wage share falls at least three percentage points by 2032. Resolution: Eurostat compensation-of-employees series for Denmark at end-2032 versus the 51.3 per cent baseline (2024–25 reading). Signal timeline: trajectory check. The series trending to 48.8–49.3 per cent of GDP by 2030 (-0.4 to -0.5 percentage points per year of structural compression from the 2025 baseline) signals on-track. The series holding at or above 51 per cent through 2030 signals the prediction is unlikely to land by 2032. Pillar consequence: no direct pillar-level unwind. Denmark is one case in a wider thesis; II.1 and II.2 hold on US data. The signal feeds into §VI Q3.

Overview #4 — Danish 10y statsobligation positive spread to Bund by 2028. Resolution: rolling four-of-four-quarters positive spread in 2028. Signal timeline: rolling 12-month average DK–DE 10-year spread compressing toward zero from the May 2026 baseline of −18 basis points (DK 2.93 per cent, DE 3.11 per cent). A directional move of at least +10 basis points by end-2027 signals on-track. A widening to ≤−30 basis points through 2027 signals unlikely. Pillar consequence: no direct pillar consequence (Danish rates are not a pillar vehicle). The signal feeds into §VI Q3.

Overview #5 — open-weight lag at or above three months through 2028. Resolution: Epoch Capabilities Index trailing-12-month average through 2028. Signal timeline: Epoch lag compressing below six weeks for two consecutive measurement periods. Current headline ~3 months on average, with the caveat that newer open-weight frontier releases (gpt-oss-120b, MiniMax-M2) lack ECI scores; the actual underlying lag is likely below the headline. Secondary signal: open-weight overtake on Humanity’s Last Exam by at least five percentage points against the closed-frontier leader. Pillar consequence: II.1 and II.4 both close. II.3 holds on the rates leg but loses the equity-rate-correlation support — if substrate diffuses, the rent capture stops compounding and the Euler-g channel is no longer cleanly muted.

The hardest unwind. If all five falsify simultaneously, the framework is wrong and the entire portfolio reverses. Joint probability is low. The more likely path is two or three falsifying together, forcing partial reversal. The above is written to handle the partial cases; the all-five case forces a full rewrite of the framework, not a reposition.

VI. Further research — open questions

Six questions, each one a research project in its own right. Section at the end, not bundled into the body.

1. η > 1 robustness across sub-periods. The Bansal-Yaron and Hall-1988 disagreement turns on aggregate calibrations across the post-1980 sample. A sub-period decomposition — 1990s productivity boom, 2000s great moderation, 2010s post-GFC, 2020s post-pandemic — would test whether η is stable or flips across regimes. If η flips, the equilibrium reading in II.3 is more fragile than the headline framework allows. The data exists; the calibration paper has not been published in a form the pillar can rely on.

2. Vissing-Jørgensen heterogeneity in 2026. Who is the marginal regime-pricer in current rates and equity markets? Composition has shifted since 2002. Hedge funds, sovereign wealth funds, retail through passive vehicles, family offices, retail through zero-commission brokerages. If the heterogeneity has narrowed — the marginal equity pricer and the marginal rates pricer are more similar than in 2002 — the η > 1-in-rates / η < 1-in-equities reading needs recalibration. If it has widened, the Pillar II.3 trade is on stronger ground than the framework assumes.

3. Cross-country institutional mapping for jurisdictional divergence. The geographic dimension the pillars deliberately drop belongs here as research. Which OECD economies have labour-market institutions that interact with the Acemoglu reading asymmetrically? Sweden, the Netherlands, Switzerland, Singapore, Korea each have their own version of the flexicurity-analogue question (Chapter VI). The cross-section is the data the framework rests on, and it is uncollected. The §V signals from Overview #3 and #4 (Denmark) feed into this question.

4. The antitrust path for substrate redistribution. What is the policy path under which the rent-capture layer is forced to redistribute? The FTC 6(b) report exists; no enforcement followed under the current US administration. The Microsoft–OpenAI restructuring in April 2026 partially addressed antitrust concerns without breaking the partnership. The EU AI Act’s enforcement powers activate 2 August 2026 — first material fine unlikely before late 2027 on typical EU cadence. Which path triggers II.1’s reversal, and on what clock?

5. Form-4 / 13F observational tests of framework alignment. The Form-4 / 13F observational test runs in three steps. First, identify the CIKs of the substrate-rent-capture firms (Nvidia 1045810, TSMC, ASML, AMD, Broadcom, the wafer-fab equipment specialists, the HBM vendors, the hyperscaler parents, the data-centre REITs) and a matched control set of non-substrate large-caps. Second, scrape Form 4 filings (insider transactions, filed within two business days) and 13F filings (institutional holdings, filed within 45 days of quarter end) for each, separating 10b5-1-plan executions from discretionary insider transactions where the filing tags allow. Third, construct two test metrics: insider net flow at substrate firms minus the same metric at the control set, scaled by market capitalisation; and the change in Herfindahl index of 13F holdings at substrate firms net of passive-index-driven flow, period over period. The framework predicts substrate-firm insider flow neutral-to-positive on a discretionary basis (after stripping 10b5-1) and rising institutional concentration among active managers. The currently visible signal is the opposite at Nvidia — eighteen-month sell-to-buy ratio of 15-to-0, $3.3 billion of executive sales in 2024, and active-manager 13F rotation out of pure substrate exposure. The interpretation is contested: 10b5-1 mechanics and standard wealth-diversification can produce the same data without framework contradiction. The cross-sectional comparison — substrate versus control set — is what makes the test informative, and that test has not been run. Conviction’s scraper infrastructure pulls the EDGAR XBRL feed nightly; the methodology above is a script away.

6. The reinstatement-weakness falsification threshold. At what point of the post-2022 labour-demand data does the Acemoglu reading become falsified rather than defended? Chapter VI’s commitment is testable; the threshold has to be set in advance. This is the question §V’s unwinding logic on Overview #3 and II.2’s Canaries-cohort falsifier ultimately rest on. The operational threshold is not yet specified.

Working document. Position pillars described are framework-consistent positions, not recommendations. No ticker calls. The data underlying each pillar is published as research notes at /essays/research-notes/01-substrate-lockdown through 06-13f-form4-methodology — working files for verification, not a parallel reading path. Framework references: the main series The Economics of the AI Transition, and the published companion essay on substrate-lockdown exterior, No Winning Strategy.