FNC-1: The Novacene Composite

Executive Signal Summary

I. What FNC-1 Is and Is Not

FNC-1 is a measurement instrument. It produces a weekly reading of where capital is being deployed across the four substrates of the AI transition, with constituents, weights, and calculation method published openly. Paired with the Belief Index overlay, it functions as a divergence detection tool, surfacing the gap between what capital is building and what prediction markets believe.

Three distinctions are worth making explicit, since FNC-1 could be confused with each.

Investment vehicles. The methodology is built for institutional readers tracking the AI transition: investors, board members, policymakers, analysts. Retail allocation is not the use case. There is no FNC-1 ETF and no plan to create one. The instrument exists to produce a weekly reading comparable across time rather than to generate returns.

AI ETFs. Existing AI-themed indices (the Roundhill Generative AI ETF, the Global X Robotics & AI ETF, the Bessemer AI Index) capture exposure to companies that benefit from AI. FNC-1 captures something structurally different: a substrate-weighted reading of where the transition is consuming capital. The selection criteria are about substrate position rather than thematic exposure. A company benefits from AI; a substrate constitutes the transition.

The SOX. The Philadelphia Semiconductor Index tracks pure compute exposure. FNC-1 weights compute at 30%; the remaining 70% sits in three other substrates that move on different cycles. Over the inaugural twelve-month window, FNC-1 returned 71.4% while the SOX returned 147.3%. The 76-point gap is the substrate diversification doing its work. If that gap closes to zero, the methodology has failed (see Falsification Triggers, Section VII).

II. The Four Substrates

The AI transition is a coordinated reallocation of capital across four interlocking substrates. Each substrate is a layer at which AI-transition capital is currently visible in public equity markets at meaningful scale.

Why these four and not five or three

These are the substrates where AI-transition capital deployment is visible in public equity markets at meaningful scale today. Each has at least two public constituents above $50B market capitalization with sufficient liquidity for repeatable methodology. Climate-tech AI is too dispersed across categories that already have substrate representation. Autonomous vehicles are a single-technology bet rather than a substrate; Tesla's exposure is captured indirectly through its compute and energy positioning.

What is deliberately excluded: the broad consumer technology stack (Apple, Amazon retail, Meta consumer revenue). Those companies benefit from AI as deployers. They sit downstream of the transition rather than constitute it.

Why substrates and not first-principles categories

A reasonable critique of the four-substrate architecture is that it imports Anthropocene categories (sectors, public equities, market caps) into a framework that claims to read a Novacene transition. The deeper organizing principles, free energy minimization, autopoietic alignment, the principle of least action, are the lens through which the transition is actually moving. Why not weight the index around those?

Because an instrument has to be calculable. Compute, Energy, Frontier, and Biological are load-bearing scaffolding: each has at least two public constituents above $50B market capitalization, sufficient liquidity for repeatable methodology, and a falsifiable inclusion rule. "Cognitive minimization" or "epistemic alignment" do not have constituents. They are the right lens for reading what the instrument shows, but they are not the architecture of the instrument itself. The substrates are how capital becomes visible. The principles are how the capital movement gets interpreted.

This division of labor is structural to FP1's stack. The instruments measure the world. The correspondents (Vera, Manticus, Darśan) reason about what the measurements mean, under named scientific frameworks. The Manifesto and the Parables carry the larger Anthropocene-to-Novacene narrative. Collapsing all three into a single first-principles index would lose the falsifiability that makes the methodology paper credible at all.

Watched Substrates

Two substrates are named explicitly as candidates for inclusion in v1.0 or a subsequent version. Each is significant enough to track but does not currently meet the public-market threshold for substrate-level weighting. The criteria for inclusion are stated in advance to commit the methodology to a falsifiable rule rather than ad hoc additions.

Physical AI (humanoid robotics and embodied intelligence). The substrate where compute meets the physical world. Capital deployment is real and accelerating: Figure, 1X, Physical Intelligence, Skild, Apptronik, Tesla Optimus, plus the Chinese humanoid push (Unitree, UBTECH, and others). The substrate is currently dominated by private actors. Public-market representation is thin: Tesla's Optimus exposure is captured inside a larger automotive and energy business, Symbotic and Intuitive Surgical are adjacent rather than core, and pure-play public humanoid companies do not yet exist at scale. Physical AI is also definitionally entangled with the existing substrates: its compute is in Compute, its foundation models are in Frontier. The substrate-distinct slice is actuators, sensors, and integration. Inclusion criterion: when at least three public companies with combined market capitalization above $100B represent the substrate's distinct slice, Physical AI is folded in at a stated weight (provisionally 10%), with the existing four substrates rebalanced. Tier 2 may also introduce a Physical AI sub-index drawing on private-market valuations as a watch instrument before full inclusion.

Defense AI. Capital deployment is visible but the public-market signal is partial. Anduril is private; Palantir's defense exposure is meaningful but mixed with commercial; Lockheed Martin and RTX have AI-relevant programs embedded inside larger defense businesses. Inclusion criterion: when more than 5% of S&P 500 market capitalization is attributable to AI-relevant defense activity, or when at least one pure-play AI defense company exceeds $50B market capitalization, Defense AI is folded in at a stated weight (provisionally 10%), with the existing substrates rebalanced.

Both watched substrates are reviewed quarterly. When a substrate's inclusion criteria are met, the rebalancing is documented in a subsequent NCB and disclosed before it is applied to the live instrument. Until either substrate is folded in, its capital deployment is reported in the substrate-adjacent commentary in State of the Transition but not weighted in the FNC-1.

III. Constituent Architecture (v0.1)

Version 0.1 of the FNC-1 is a seven-stock proxy basket. The proxy is the actual instrument used for the inaugural readings, sufficient on its own for repeatable methodology and substrate-level signal. Future versions will expand the basket toward fuller coverage of each substrate.

Each constituent is the cleanest publicly-traded representative of its substrate position, with sufficient market capitalization, liquidity, and disclosure to support repeatable methodology. The proxy is optimized for substrate representation rather than return.

Selection notes: Microsoft is assigned to Frontier rather than Compute despite substantial Azure infrastructure, because the Frontier exposure (OpenAI partnership) is what the index is designed to surface. Alphabet is assigned to Frontier rather than Compute despite TPU manufacturing, on the same logic: the Frontier signal is the one the methodology is built to capture. Constituents that span substrates are assigned to the substrate where their exposure is most differentiated.

Version 1.0 expansion: the production FNC-1 will likely expand to twenty-to-thirty names, with additional weight given to TSMC and Broadcom (Compute), Vistra and BWX Technologies (Energy), Meta and possibly select private-market proxies as they go public (Frontier), and several biological-intelligence names beyond Recursion. The expansion methodology will be specified in a future NCB.

IV. Weighting Scheme

The substrate weights reflect FP1's view of each substrate's structural importance to the AI transition, selected directly rather than derived from market capitalization.

Compute (30%). High enough to give compute its central role in scaling-law dynamics, but capped to prevent the index from becoming a SOX restatement. Compute is the substrate where most of the visible capital is currently moving; the cap is a deliberate constraint against single-substrate concentration.

Frontier (35%). The highest weight, because frontier labs sit between compute (their input) and applications (their output). They are the bottleneck through which transition value flows. The weight will be reviewed if the largest frontier labs go public and the substrate's market-cap exposure changes materially.

Energy (20%). Large enough to capture the binding constraint of the late 2020s, but smaller than compute and frontier because energy infrastructure operates on slower capital cycles. Capacity additions take years; the substrate's contribution to the transition signal lags.

Biological (15%). A present allocation that signals long-term substrate importance while reflecting the slower revenue ramp. The weight may increase to 20% if the substrate's revenue base expands materially in 2026-2027, with a corresponding reduction in Frontier.

Within each substrate, individual stocks are equal-weighted. This is a deliberate methodology choice: market-cap weighting within substrates would over-concentrate in the largest names (NVIDIA, Microsoft) and reduce the diversification that defines the basket's design. Equal-within-substrate weighting maintains the substrate-level signal as the primary axis.

The weights are reviewed annually under a stated process. Between reviews, they remain stable regardless of substrate performance.

V. Index Calculation

The FNC-1 Proxy Basket is calculated as follows:

1. At the start of the calculation window, each constituent's price is normalized to 1.0 (price divided by start-of-window price).
2. Each normalized series is multiplied by its target weight.
3. The weighted normalized series are summed across constituents to produce a single time series.
4. This series is multiplied by 100 to produce the index level.

The window is rebased to 100 at the start of each new annual measurement period (currently April 21 each year). Within a window, all readings are comparable to one another and to reference benchmarks rebased to the same start.

Rebalancing. The production methodology specifies quarterly rebalancing. On the first trading day of each quarter, the substrate weights and individual weights are re-applied to current constituent prices. Between rebalancing events, the basket drifts naturally with constituent prices.

Version 0.1 applies weekly rebalancing for simplicity in the chart-building pipeline. The production methodology will move to quarterly rebalancing in v1.0. The difference is small over a twelve-month window and does not materially change the substrate-level signal.

VI. Belief Index Overlay

The Belief Index complements the FNC-1. It reads what prediction-market participants currently believe about acceleration in the AI transition, normalized to a 0-100 scale where higher numbers mean stronger belief in continued acceleration.

The version 0.1 overlay draws from four Polymarket markets:

For each market, the "Yes" probability is the panel reading. Acceleration markets use the probability directly. Deceleration markets are inverted: panel reading equals one minus the probability. This keeps the index directionally consistent.

The Belief Index is the panel-weighted average of the four readings, multiplied by 100.

Why these four markets. Each addresses a distinct dimension of the transition. The AGI-by-2027 market reads capability ceiling. The GPT-6-by-2026 market reads capability cadence. The OpenAI-IPO-by-2026 market reads capital-market integration. The bubble-burst market reads deceleration risk. The selection is intentionally narrow. A panel of four markets allows movement in any one to be visible. Larger panels would smooth too much and reduce the overlay's value as a directional signal.

Why Polymarket only. Version 0.1 reads only Polymarket because it offers the largest active liquidity for AI-transition markets and a public API with no authentication required. This is a known limitation. A regulated-venue cross-check is on the near roadmap.

Future expansion. The panel will add Kalshi readings in a future version. Kalshi is regulated as a US Designated Contract Market, which provides a different demographic of participants than Polymarket and a regulatory cross-check on Polymarket's signal. Adding two Kalshi markets (a cross-venue AGI market and a longer-horizon AGI-by-2030 market) gives the panel six components and a regulated-venue null test against the Polymarket reading.

What the Belief Index reads, and what it does not

The Belief Index is one signal among several. It is not principled Bayesian prediction. It does not derive forecasts from underlying models of how the AI transition unfolds. It reads what financially-staked participants on a small set of liquid markets currently believe about a small set of dated questions. The participants skew toward international, crypto-native, and institutional traders with stakes large enough to discipline their priors but not large enough to move the markets they trade in.

That is a narrower epistemic claim than "what the market thinks." It is a meaningfully different signal from a Bayesian forecast built on first principles, and FP1 treats it as such. The point of pairing the Belief Index with FNC-1 is not to elevate prediction-market sentiment to the status of structural analysis. It is to juxtapose sentiment with capital deployment and surface the gap between them. The divergence is the read; the components feeding into it are kept honest about their epistemic limits.

In future versions, the panel may be supplemented or partly displaced by signals with stronger first-principles grounding: capacity utilization, capital-cycle indicators, energy-cost trajectories, model-capability benchmarks, and where appropriate, structured forecasts from the correspondent stack. The Belief Index remains valuable as a fast-moving sentiment overlay even as the panel diversifies. It will not be asked to carry weight it cannot bear.

VII. Falsification Triggers

The methodology is invalidated, and the instrument suspended for re-specification, if any of the following occur. Each trigger is concrete and falsifiable.

When a falsification trigger fires, the instrument is suspended for re-specification. Re-specification is documented in a subsequent NCB.

VIII. Limitations

Four limitations of v0.1 are explicit:

Public-market visibility only. The FNC-1 captures publicly-traded substrate exposure. Privately-held frontier labs (Anthropic, OpenAI, xAI, Mistral) remain outside the Frontier substrate weighting until they go public. The OpenAI-IPO-by-2026 component of the Belief Index is partly a forecast of when this gap closes. Until it does, the Frontier substrate is read primarily through Microsoft's OpenAI partnership and Alphabet's owned-and-operated frontier work.

US-centric geography. Six of seven proxy constituents are US-listed. ASML is the exception. The index does not yet capture the substrate exposure of TSMC (Taiwan), Samsung Electronics (Korea), or non-US energy infrastructure providers. v1.0 will expand geographic coverage. Until then, the FNC-1 reads the AI transition primarily through US capital markets.

Substrate boundaries blur. Microsoft is a frontier-lab investor (OpenAI), a compute provider (Azure), and an enterprise software vendor. Alphabet is a frontier lab (Google DeepMind), a compute manufacturer (TPU), and an advertising business. The methodology assigns each constituent to a single substrate based on the exposure the index is designed to surface, but the assignments are judgment calls that will be revisited annually.

Belief Index is venue-concentrated. v0.1 reads only Polymarket. Until Kalshi is added, the Belief Index inherits the demographics, jurisdiction, and trading dynamics of a single venue. Polymarket is restricted from US persons, which means its participants skew toward international and institutional crypto-native traders. Calling the reading "what markets believe" would overstate the panel's representativeness; it reflects what one specific market believes, with that demographic skew baked in.

IX. Reflexivity and Epistemic Gain

The standard finance critique of any cited index is reflexivity. If institutional readers begin allocating capital based on FNC-1 readings, capital flows into FNC-1 constituents, and the index measures its own influence rather than the underlying transition. The thermometer becomes the thermostat. Open methodology is a partial mitigation. It is not a complete one.

FP1 treats reflexivity differently from a return-seeking index because the test of the instrument is different. A return-seeking index is judged by whether it tracks an unobserved truth. A measurement instrument built around first principles is judged by something else: whether the act of measuring produces epistemic gain. By that we mean a verifiable narrowing of the gap between what is believed about the AI transition and what is structurally true of it.

Stated more concretely: if FNC-1 readings cause capital to be deployed against substrate-level evidence rather than thematic exposure, that is the instrument working. If they cause capital to be deployed against principled signals (energy-cost trajectories, capability cadence, capacity bottlenecks) rather than narrative, that is the instrument working. The test is not whether the instrument stays passive; it is whether the system it sits inside becomes better calibrated to reality as a result of being measured. The reflexivity that worries finance is the reflexivity of a tool whose purpose is to track. The reflexivity here is an instrument whose purpose is to align.

Epistemic gain is itself measurable, in principle. Candidate metrics include: the divergence between FNC-1 and SOX (does the substrate framing remain descriptive over time?); the leading or lagging relationship between the Belief Index and FNC-1 movements (does sentiment cohere with structural deployment?); the calibration of correspondent-stack predictions made under the methodology (are the named scientific frameworks generating accurate forward calls?); and the degree to which falsification triggers fire at the predicted thresholds (does the methodology break in the ways it predicted it might break?). A future NCB will specify the epistemic-gain audit explicitly.

This is not a defense against reflexivity; it is a reframing of the question reflexivity asks. The traditional reflexivity critique assumes the goal is to be a passive thermometer. The Novacene framing assumes the goal is alignment with first principles. Whether the instrument is succeeding or failing under the second goal is itself an open empirical question, and it is one the methodology will publish honestly as it accumulates evidence.

X. Open Questions

Three questions for future versions:

Is biological intelligence weighted correctly? 15% reflects current market-cap presence of pure-play biological-intelligence names, but the substrate may be undervalued relative to its long-term importance. A weight increase to 20% would reduce Frontier's share to 30%. The decision will depend on how the substrate's revenue base develops in 2026-2027 and whether AlphaFold-derived drug discovery produces clinical readouts that change the public-market picture.

When does a Watched Substrate cross into full inclusion? Physical AI and Defense AI are both named in Section II with stated inclusion criteria. The open question is timing. Physical AI's path runs through public-market readiness of Figure, 1X, Physical Intelligence, or a Tesla Optimus spin-out, none of which has a confirmed schedule. Defense AI's path runs through an Anduril public listing or a Palantir defense-segment carve-out, also without a confirmed schedule. The criteria are stated; the calendar is not. Tier 2 of the methodology will include a watch-list section in the weekly reading that tracks proximity to each criterion.

How are retiring frontier labs handled? If a frontier lab is acquired, fails, or otherwise exits the public market between rebalancing events, the methodology has no current rule for substitution. v1.0 will specify a substitution rule, likely "next-largest pure-play in the same substrate, with substrate weight unchanged."

XI. Roadmap

The methodology will evolve in three tiers.

Tier 1 (current). Seven-stock proxy basket, weekly rebalanced for chart-building. Belief Index from four Polymarket markets. Published weekly in State of the Transition. This is the version specified in the present paper.

Tier 2 (Q3 2026 target). Full constituent expansion to twenty-to-thirty names across the four substrates. Quarterly rebalancing logic. Substrate sub-indices (FNC-1 Compute, FNC-1 Energy, FNC-1 Frontier, FNC-1 Biological) for substrate-level reading. Watched Substrates dashboard tracking Physical AI and Defense AI proximity to inclusion criteria, with a private-market overlay for Physical AI capital deployment. Backtested historical series for retroactive analysis. Kalshi belief markets added to the overlay.

Tier 3 (2027 target). Live hosted version on the FP1 Terminal page. Real-time chart with daily updates. API access for institutional readers. Substrate-level Belief Index breakdowns. Cross-venue belief cross-check fully operational.

Each tier will be documented in a subsequent NCB. The current paper specifies Tier 1 only.