SciRisk

Research & disclosures

The model record behind the SciRisk 100, and what remains open.

Model record

Model

SciRisk ESCRC Model v5.4.2

Calibrated priors

2026.09-5

Dataset

2026.09-23 (beta)

Company data as of

2026-06-30

ESCRC calculated

2026-09-21

Simulation

100,000 paths · 95% · 365d · rho 0.35 · seed 20260821

Documents

• SciRisk 100 methodology — how this index is produced.
• ESCRC Calculator quick start guide (PDF) — the model in practice.

The full ESCRC Methodology Specification and the independent model-validation reviews are maintained with the calculator and are available on request.

Known limitations of this release

• Parameter layer. Per-supplier disruption rates, recovery times and revenue-at-risk fractions are calibrated bucket defaults, not company-specific research. This systematically understates well-researched networks — quantified on the six company pages where a researched network exists for comparison.
• Network completeness. Each company is modelled on a curated subset of its supplier base (9 to 15 suppliers). Unmodelled suppliers carry no risk in the simulation.
• Cross-company supplier overlap. The model represents each company's supply network independently — there is no shared, de-duplicated supplier graph across the index. Of 458 distinct named suppliers across all 100 companies, 207 appear in more than one company's network (Glencore appears in 25 companies' networks, the most of any supplier). Summed figures (aggregate ESCRC, sector totals) therefore count a shared supplier's risk once under every company it appears beneath, not once overall. See the aggregate-ESCRC note wherever a summed figure is shown.
• Deep-tier attenuation is a fixed constant. Suppliers at tier 2 and deeper enter the simulation through a fixed echelon ladder (tier 2 at 0.6, tier 3 at 0.3, deeper at 0.2) rather than through the intervening buffer stock, second-source or restart time that physically determines whether an upstream disruption reaches the focal firm. The engine supports those per-relationship inputs and the ESCRC Calculator exposes them, but this index does not populate them — no such estimate exists in the public disclosure the dataset is curated from. The ladder behaves like an assumption that every deep-tier relationship sits behind roughly a week of buffer: it understates a sole-sourced, unbuffered upstream dependency and overstates a well-buffered one.
• Deep-tier exposure is not evenly distributed. Because the attenuation above is a constant, its effect on a company scales with how much of that company's declared exposure sits at tier 2 or deeper — and that share ranges from 8.9% (HP) to 55.8% (Alibaba Group) across the index, median 24.3%. The assumption is therefore not common-mode: it does not cancel out of a comparison between two companies, and it can move relative rank, not only level.
• Change series. Change is computed against the 2026.09-22 publication (2026-09-19).
• Grade thresholds. Preliminary and distribution-anchored — see the methodology.

Coming in Phase 2

Scenario and what-if analysis on index companies, historical ESCRC and change over time, alerts and watchlists, portfolio monitoring, an API, and PDF reporting. None of these are part of this release.