Published research usually asks you to trust the numbers. Here you don't have to.
Below are the headline results from four of our papers. Each is either re-run live, in your
browser session, through the same code that produced it — or shown with its published source
and an explicit status saying exactly what cannot yet be recomputed.
contacting the live engine…
The rule on this page. Every claim is either regenerated live through the same code
that produced it, or shown with its published source and an explicit status — nothing here
is simulated. Where the compute engine does not yet expose a method, we say so plainly and link
the paper rather than invent a live number. Three of the four results below recompute live, on page load or on
demand: the Scale-Ordered Contagion profile on real G20 data, the contagion-channel
attribution (run through the paper's own published contagionchannels
package), and the NAMH efficiency and information-flow primitives (run through the paper's own
namh estimation code) — so the published and live numbers are identical by
construction. Two surfaces remain honestly pending: the systemic-risk index (MCPFM), whose method
is not yet registered, and the NAMH surrogate-gated network, which is empty under the paper's own
BH-FDR gate — each shown from its paper with an explicit status, never a fabricated live number.
regenerated live by the engine published result · live regeneration pending
USA → India: tail spillover by time-scale live regeneration
SOCH · arXiv:2606.04113 · directed wavelet-quantile transfer entropy at the 5% tail, across 4 wavelet scales (d1≈2–4 days → d4≈16–32 days). Method: soch_profile · the published sochcontagion package, run live.
Published — Bhandari & Parida (2026)
0.0391
4-scale agg
0.0426
5-scale agg (paper)
d4
peak scale
The paper reports a 5-scale profile (0.0155, 0.0425, 0.0491, 0.0494, 0.0567) across d1–d5, aggregate 0.0426. The live engine decomposes to 4 scales (d1–d4), aggregate 0.0391. The whole difference is the d5 (32–64 day) band — the per-scale d1–d4 values match exactly.
Regenerated live
running on the live engine…
—
4-scale agg
—
peak scale
Try another pair:→
SOCH-A · Ordering
The profile peaks at a coarser scale when the slower market is involved.
✓ p = 0.042
SOCH-B · Shape symmetry
The profile's shape is direction-symmetric even when its level is not.
✓ 28 / 28 pairs (p>0.05)
SOCH-C · Level asymmetry
The overall level is directionally asymmetric (connectivity × shock).
p = 0.105 (n.s.)
The USA→India panel above is the live, single-pair proof. The full 28-pair shape-symmetry battery (block-bootstrap) is the paper's heavier computation, available in the package; it is not run on page load.
FORMAL · Lean 4 + mathlib (toolchain v4.30.0): a machine-checked layer now covers the paper's spectral mathematics — 11 of 12 declarations Lean-accepted. prop:spectrum is formally closed: ‖H(ω)‖² = S(ω) is a Lean theorem, unconditional. The peak's stationary-point half is fully proved: the first-order condition reduces exactly to 3x² + (αs²+αr²)x − αs²αr² = 0 in x = ω²; the symmetric peak sits at ω* = α/√3; and there is exactly one positive frequency satisfying the FOC (Q strictly monotone on [0,∞)), lying in [αmin/√3, αmin]. What remains stated, not proved: lem:peak's analytic half — that ω·S(ω) attains its maximum and the maximiser satisfies the FOC — honest sorry-count 1 of 12 declarations (0 = formally closed). Stated ≠ proved; the rest is a continuing research line, not a claim.
Systemic Risk Index: crisis early-warning live regeneration pending
MCPFM · arXiv:2507.08065 · the equilibrium-derived Systemic Risk Index (SRI) discriminating crisis from non-crisis days. Validation: ten major U.S. equities, 2010–2024, COVID-19 crash as natural experiment.
Published — Bhandari (2026)
SRI0.915
VIX0.947
0.915
SRI AUC
1 day
SRI lead time
0.947
VIX AUC
The SRI achieves AUC = 0.915 (excellent crisis discrimination). It sits modestly below the VIX benchmark (AUC 0.947, gap −3.4%) but warns earlier — 1-day lead vs the VIX's 2 days, with peak detection on 10 Mar 2020, one day before the WHO pandemic declaration.
Live regeneration
No systemic-risk-index method is registered in the compute engine yet, so this result is not recomputed here. It is shown above from the paper. We will not fabricate a live number — when the SRI method is registered, this panel will recompute the AUC on the live engine, exactly as the SOCH profile does above.
—
live AUC
Tier G
method status
◷ Live regeneration available once the systemic-risk method is registered (Tier G). Until then, the figure above is the published result, not a live value.
Companion theory: Nash-equilibrium existence (Theorem 7.2) and the trust-update law drive the SRI. Paper: arXiv:2507.08065.
Contagion channel attribution live regeneration
contagion-channels · arXiv:2604.26546 · each significant directional contagion link attributed to one of five mutually-exclusive transmission channels via IV/2SLS + LASSO + local projections + heteroskedasticity-based ID. 18 G20 markets, eight crisis sub-periods, 2006–2026. Method: channel_attribution — the engine runs the paper's own published contagionchannels package, reproducing Table 5 to 0.000 pp.
Published — Bhandari, Parida & Sahu (2026)
5
channels
8
crisis episodes
0.033
edge threshold
Dominant channel per episode (% of attributed links, Table 5). Financial frictions dominate the Pre-Crisis baseline and the European Sovereign Debt Crisis; Trade leads the GFC; the monetary-policy channel leads three later episodes. GFC and COVID are flagged exploratory (Sargan over-identification rejection) and marked * — their shares are descriptive, not identified. These are the paper's point estimates; the engine emits the same point shares (no confidence intervals are invented here).
Regenerated live
running channel attribution on the live engine…
Paper: arXiv:2604.26546 — Bhandari, Parida & Sahu (2026), an arXiv preprint (econ.GN). Because the engine executes the paper's own published contagionchannels code, the live column is identical to the published Table 5 by construction.
NAMH: market efficiency & information flow live regeneration
NAMH · Bhandari & Sahu (2026, in preparation) · the deterministic NAMH primitives — rolling DFA Hurst node-weights φ(H)=1−2|H−0.5| and KSG transfer entropy — recomputed on the 24-market G20-plus-commodities panel (2001–2025). Methods: namh_hurst, namh_te · run through the paper's own namh estimation code on the live engine.
Published — Bhandari & Sahu (2026)
Rolling DFA-1 Hurst per market (window 252, step 252, 20 non-overlapping windows). Japan is the most efficient developed hub (H≈0.484, mean φ≈0.90), the U.S. mildly anti-persistent (H≈0.472), China the least efficient of the four (H≈0.450). φ(H)=1−2|H−0.5| is the NAMH node-weight feeding the network efficiency fixed point.
◷Network hub & centralities — amber: empty under the paper's own gate — pending. The published NAMH network (Japan the eigenvector hub in 13 of 20 windows; the U.S. the out-strength broadcaster) is gated by IAAFT surrogates whose generator is unseeded, and under the paper's own Benjamini–Hochberg FDR gate the surrogate-tested network is empty — 0 of 552 possible edges in every one of the 20 windows. The published centralities use a magnitude gate the deterministic pipeline does not expose, so we show them from the paper and will not fabricate a live hub here — never green, never magnitude-dressed. The seeded end-to-end run is namh_pipeline (async, job-server; reproducible for a fixed seed and core count).
The deterministic NAMH surface (Hurst, raw KSG transfer entropy) reproduces above to floating-point precision because the engine runs the paper's ownnamh code. The effective-TE / surrogate / network surface is the heavier async computation and is honestly reported amber, never simulated. Paper and namh package: Bhandari & Sahu (2026), in preparation.
Frontiers III: the directed network of news attention live regeneration
Frontiers III · Bhandari & Sahu (2026, in preparation) · the estimable directed observation network of Kranton & McAdams (2020)'s market for news, recovered by directed transfer entropy across 15 channel-tagged news-attention topics (daily log-changes, 2018–2026). Method: news_attention_te — the same validated max-norm KSG / Frenzel–Pompe estimator as ksg_te, run live on the news_attention panel.
running news_attention_te (inflation→election) on the live engine…
◷Significance-gated network — amber: pending. The published significant-edge network (density 0.0857, election the dominant attention sink, the monotone densification across the pre-2020 / COVID / post-2022 eras) is gated by IAAFT source surrogates whose generator is unseeded, and under Benjamini–Hochberg FDR the gated network is empty — reported honestly, never magnitude-dressed. The seeded surrogate path is the async ksg_te method on dataset=news_attention. Here we reproduce the deterministic raw transfer entropy, never a fabricated gated edge.
The directed news-attention transfer entropy reproduces to the published precision because the engine runs the same validated KSG core as the working paper — its CPU estimator is bit-exact to the GPU kernel that produced the published TE_matrix (max|Δ| ≤ 7×10⁻¹⁶). Paper: Bhandari & Sahu (2026), in preparation; the live instrument that maintains the news-attention panel is the NEURICX research engine.
Real data · real engine · honest about its limits.
The live SOCH profile and the contagion-channel attribution are computed by the sandboxed compute engine at
shssm-compute…run.app on the bundled G20 daily-returns panel (18 markets, 2006–2026);
channel attribution runs the paper's own published contagionchannels package, so it reproduces Table 5 exactly.
The NAMH Hurst exponents and transfer entropy are recomputed on the paper's 24-market G20-plus-commodities panel (2001–2025) through the paper's own namh code, matching the published values to floating-point precision.
The MCPFM systemic-risk index and the NAMH surrogate-gated network are shown from their papers with an explicit "live regeneration pending" status — not simulated, not a substitute live value.
Papers: SOCH (2606.04113) ·
MCPFM (2507.08065) ·
contagion-channels (2604.26546) ·
NAMH (Bhandari & Sahu 2026, in preparation).
Code: sochcontagion ·
contagionchannels (GPL-3) ·
install from r-universe ·
namh (GPL-3, release pending) ·
explore every registered method in the Workbench.