PILOT DEMONSTRATION — THREE DATA LAYERS.Candidates, scores, verdicts, tensions and gaps are the platform's own export (KC Cross-Sector Convergence, 22 Jul 2026, 435 simulation responses). Analyst names and cited calls are public record, verified. Corpus statistics and friction numbers remain simulated pending pilot data. Street layer reused from the verified 22 July sweep.
CONTEXXT
Comparative Verdict · Producer: Kepler Cheuvreux

Cross-sector convergence: the AI-infrastructure bake-off, re-read against a different bench

Leading candidate — "When the Hyperscaler Walks Away: PPA Termination Risk in European Utilities"
Platform export ingested — 22 July 2026 · 12 candidates · 3 simulation runs · 435 responses
Street layer — verified 22 July, reused · Posture — mixed (conquest 0.5 / leverage 0.5)
Read this first · What you are looking at

A pilot-grade Comparative Verdict: from one research brief to a sequenced publication decision

What this is. This page is the output of an end-to-end run of the CONTEXXT platform for Kepler Cheuvreux, fused with two independently built grounding layers. It shows what the platform delivers at the moment a Head of Research needs it most: after the ideas exist, before any analyst time is committed. Nothing here is a mock-up of a hypothetical workflow — the candidates, scores, verdicts, tensions and evidence gaps were generated by the platform against a fully specified brief, authored for this demonstration as a realistic Kepler Cheuvreux scenario, and the analyst names attached to them are Kepler's real bench, verified against the public record.

What it ultimately does. It converts a thematic idea into owned ground. Every research house holds monopolies it under-exploits — Kepler's are its independence (no banking conflicts, so it can say what bank-affiliated houses cannot), its №1-ranked corporate access, and a 12-office European network no aggregator can replicate. The verdict engine reads each candidate piece for which of those monopolies it leverages and which new monopoly it builds — here, the "AI Infrastructure Stack": a standing cross-sector franchise whose every instalment generates a proprietary dataset or scoring framework that compounds the firm's information advantage. The end state is not a report; it is a monopoly plan — which piece publishes first, carried by which analysts, gated on which evidence, sequenced so each instalment deepens a moat competitors can read but cannot replicate.

The scenario, as briefed. Kepler Cheuvreux brings one thematic concern to the platform: cross-sector convergence — how power, semiconductors, real estate and capital markets are being reshaped by the AI infrastructure buildout. The brief's parameters as entered: EMEA geography, 12-month horizon, time-sensitive, consensus-challenging stance, and a mixed strategic posture — 0.5 leverage (exploit existing franchises in European utilities, industrials and semi capital-goods) and 0.5 conquest (build the cross-desk synthesis product no European house currently produces). House constraints the platform had to respect: Kepler's analytical, European-first tone; no speculation on unannounced M&A; no naming of undisclosed PPA counterparties; primary sources from 2024 onward.

What the platform did before this page existed Platform export

0 · Producer profile — before any project, the Research Orgs catalogue defines who is publishing: Kepler's four monopolies, brand reputation and authority signals (see the Producer profile tab) — the weighting layer every later score is computed through. 1 · Brief intake — the thematic concern, posture weights, constraints and horizon, structured as above. 2 · Universe construction — the Subject Universe (who the pieces are about: ~250 listed EMEA corporates, ~40 bellwethers, ~€800bn of AI-linked capex commitments through 2030) and the Reader Universe (who they are for: EMEA allocators making the largest cross-sector capex call of the decade, with current consensus and conviction thresholds mapped). 3 · Candidate generation — 12 candidate pieces, each scored on remarkability decomposed into novelty, controversy, utility and timeliness. 4 · Question elicitation — the engaging questions each candidate raises, drawn onto candidate cards and question sets. 5 · Persona generation — synthetic reader personas built from the Reader Universe. 6 · Simulation — three runs, 435 persona responses, each recording answer, reasoning, confidence, emotion and counterfactual. 7 · Synthesis — the Comparative Verdict itself: winner and ranking, P(success) decomposed into uptake, franchise, credibility and distribution, narrative-shift magnitudes, open tensions, evidence gaps, and a five-step investigation plan.

What this page adds. Two layers the platform run did not contain, built and verified separately: the bench layer — Kepler's real analysts and their cited public calls, verified July 2026, mapped onto each instalment — and the street layer — a 134-stock, 1,250-analyst sweep of every covering analyst on the affected universe, with 27 load-bearing claims source-checked. Provenance is marked throughout: Platform export Verified public record Simulated pending pilot

The platform run itself

The full interactive research preview this page was built from is shareable directly from the platform.

Open the shared research preview →
In plain terms One brief went in. The platform built the audience, generated twelve ideas, stress-tested them against 435 simulated readers, and ruled on what to publish and in what order. This page grounds that ruling in reality — real analysts, verified calls, the real street — so the decision it shows is one a Head of Research could act on tomorrow.
What's clickable on this page: the eight tabs below navigate the demonstration (each section has its own shareable URL); the coral button above opens the live CONTEXXT research preview in a new context; and the underlined source links in the footer and Producer profile open the public-record evidence behind each verified claim. Everything else — score bars, provenance tags, cards and tables — is a static exhibit: this page is a rendered deliverable, not the platform itself. The interactive product lives behind the coral button.
The bake-off — the platform's 12 candidates Platform export

Twelve candidates, one split decision, one sequencing verdict

All scores below are the platform's, as computed. The two verdict deliverables disagreed instructively: Brief A crowned the Grid Bottleneck on remarkability and differentiation (confidence 0.74); the fuller synthesis crowned PPA Termination Risk on composite (0.81) and reader activation. The synthesis resolves the split with sequencing, not a coin toss — see Playable lines.

In plain terms The platform generated twelve piece ideas, simulated 435 reader reactions, and ranked them. Two finalists emerged for different reasons: the PPA piece makes readers act fastest; the Grid piece builds the more durable franchise. The verdict: publish the PPA piece first, the Grid piece 6–8 weeks later — and fold, incubate or rework the rest.
№ 1 · Winner (synthesis) · SELECTED

When the Hyperscaler Walks Away: PPA Termination Risk in European Utilities

Composite0.81
Remarkability86
P(success)0.76

Narrative shift 0.72 — highest in set. Verdict: Publish — evidence-strengthen first.

№ 2 · Winner (Brief A) · SELECTED

The Grid Is the GPU Bottleneck: European Power Networks as the Binding Constraint

Composite0.79
Remarkability82
P(success) · highest franchise (0.72)0.77

Verdict: Publish 6–8 weeks post №1 — build TSO queue database first.

№ 3 · FOLD INTO №1

The PPA Mirage: Why Hyperscaler Contracts Won't Save European Utility Valuations

Composite0.70
Remarkability85
P(success)0.65

20–40% haircut range flagged as less forensically credible; cannibalises №1. Verdict: absorb as downside appendix.

№ 4–5 · INCUBATE / REWORK

The Water Wall (0.62 — commission enforcement cases) · Gallium & Chip Sovereignty (0.52 — fix AMG omission)

Water Wall P(success)0.61
Gallium P(success)0.53

Water Wall: theory-to-enforcement bridge missing — KC's Madrid/Amsterdam offices to document basin-level cases. Gallium: AMG Advanced Metallurgy omission is the most-cited credibility gap; SMC team not consulted in drafting.

RankCandidateRemark.NoveltyControv.UtilityStatus
1When the Hyperscaler Walks Away: PPA Termination Risk860.880.900.85selected
2The PPA Mirage850.750.950.90selected → fold
3Grid-Queue Arbitrage: Two-Tier EU Data-Center Market840.800.750.95surfaced
4The Grid Is the GPU Bottleneck820.800.700.95selected
5The Water Wall: Cooling Scarcity820.900.700.85selected → incubate
6The Gallium Chokepoint810.850.800.80draft
7The AI Capex Financing Stack: Who Lends When Hyperscalers Won't Own810.850.650.90draft
8Sovereign Green Bonds Are Mispricing AI-Infra: The €200bn Gap790.850.800.75draft
9Gallium, Hafnium & Chip Sovereignty790.850.700.80selected → rework
10Europe's Cooling Crisis: Immersion Tech as Margin Driver780.900.600.85draft
11Nuclear Incumbents as AI Baseload770.700.750.85draft
12ASML's Customers Are the Wrong Bet: Beyond EUV770.700.800.85draft

Platform export Scores verbatim from the bake-off table (remarkability breakdown: novelty / controversy / utility / timeliness as stored). Composite and P(success) figures from `impact_ranking` and `probability_of_success_per_piece`. The platform flags all quantitative ranges as initial synthetic signals — "the beginning of an investigation, not a research conclusion."

Bench friction × street friction — the KC read Scores model-derived

Same six lines, re-priced for this publisher

Street friction is publisher-independent — it measures the street. Bench friction re-computes per house. The pairing changes which quadrant each line lands in.

In plain terms The street numbers are identical to the prior benchmark (same street). The KC column shows how differently the same lines price against this bench — the utilities-normalisation line flips from "contrarian" to "extends the house call," and the grid line flips from "publish into a void" to "extend a published date."
Narrative linePlayabilityKC bench frictionStreet frictionRead for Kepler
European transmission of the capex cliff
demand end imported; cascade modelled on KC coverage
84Low (14, scenario-framed)8–10%The lead. Anchored by published KC positions at every downstream link.
Grid-capex clock
extends the verified NKT TSO-plateau call (2030–35)
79Low (15) — extends own call90%+ unmodelledWhite space where KC already owns the only published date. Other houses needed a dataset first; KC has a call first. Claim it.
Utility premium unwinding / power-price normalisation
€40/MWh forwards · Becker's near-bubble language (verified)
74Lowest (8) — Becker leads40–50% (segment)At comparable houses this was the most contrarian surviving line; at KC it is the house call. Play as instalment 1's anchor chapter, Becker-led — this is now the platform's confirmed winner.
Cooling / execution-layer margin expansion
Vertiv/Trane/Carrier stats verified; Legrand is KC's own proof point
72Low — extends Mackie<1%Proof layer, as in the prior benchmark — with Legrand's 25%-of-sales data-centre exposure as the European exhibit KC covers first-hand.
Backlog-to-earnings conversion timing
contracted vs sentiment discrimination
66Low-medium5–7%Mackie's ledger chapter: which EU electricals backlog is take-or-pay, which is optimism.
AI chip demand plateau, dated (US)
Northland CY27-decline (verified) · JPM $650B (verified)
43Medium (semicap seats)70–80%Same trap as in the prior benchmark plus the standing problem: wrong jurisdiction for this bench. Absorb the evidence; do not lead.
The cross-house finding: the two-axis grid is the same; the houses occupy opposite corners of it. The prior benchmark's cheapest line (cooling consensus) is KC's proof layer but not its lead; KC's cheapest line (utility premium unwinding, friction 8) was the prior benchmark's most expensive surviving one (40–50% street, against its own utilities franchise). The platform's claim in one picture: the verdict is a function of the bench, and it computes.