{"data":{"framework":"Three-leg hybrid: Leg 1 = hierarchical two-block weighted geometric composite (Structural Fragility S x Dynamics/Trigger D) with VIX multiplier, non-compensatory red-flag override, and a seeded Monte Carlo whose MEDIAN is the headline; Leg 2 = Faber 10-month trend trigger; Leg 3 = fast volatility alarm. The legs are NOT averaged. Action bands: < 45 hold; 45-60 trim; >= 60 or override -> de-risk.","disclaimer":"**bubblegauge is a research instrument, not investment advice.** The headline is a 0–100 regime heuristic produced by structured expert judgment; it is **uncalibrated and is not a probability**. The reference class of comparable US equity manias is roughly four events {1929, 2000, 2007, 2021}, so no honest probability calibration is possible. The current episode may be rational general-purpose-technology repricing rather than a bubble. Nothing here is a recommendation to buy, sell, or hold any security. Any de-risking rule may destroy value net of costs. Use at your own risk.","indicators":{"s1":{"name":"Valuation Extremity","weight":0.33,"grounding":"literature-grounded","references":["Campbell, J. Y. & Shiller, R. J. (1988). 'Stock Prices, Earnings, and Expected Dividends.' Journal of Finance 43(3): 661-676. doi:10.1111/j.1540-6261.1988.tb04598.x","Siegel, J. J. (2016). 'The Shiller CAPE Ratio: A New Look.' Financial Analysts Journal 72(3): 41-50. doi:10.2469/faj.v72.n3.1"],"caveats":["Siegel (2016) shows that post-1990 GAAP changes — especially FAS 142 goodwill impairment and mark-to-market accounting — depress reported earnings and bias CAPE upward relative to its own history, so cross-era comparisons overstate current extremity. CAPE is a long-horizon expected-return gauge, NOT a timing tool (near-zero one-year predictive power); Asness/AQR's 'sin a little' caution applies — a high CAPE can persist for years."]},"s2":{"name":"Concentration","weight":0.27,"grounding":"literature-adjacent","references":["RBC Wealth Management, 'The Great Narrowing' (Jan 2026)","S&P Dow Jones Indices concentration data","JPMAM Guide to the Markets (cross-check)"],"caveats":["The lo/hi anchors are judgmental (chosen to bracket the 2000-peak ~27% and a plausible extreme ~41-44%), not estimated from a labeled crash dataset; concentration has weak standalone timing power. It also feeds red-flag #4 in combination with breadth.","Some data vendors report '36.4%' as the information-technology sector weight — the S2 input is the sum of the top-10 individual holding weights, read from the holdings XLSX, not a sector table."]},"s3":{"name":"Semiconductor GSY Run-up","weight":0.2,"grounding":"literature-grounded","references":["Greenwood, R., Shleifer, A. & You, Y. (2019). 'Bubbles for Fama.' Journal of Financial Economics 131(1): 20-43. doi:10.1016/j.jfineco.2018.09.002"],"caveats":["The 53%/80% crash frequencies are Fama-French-49 industry-level base rates, NOT calibrated to this specific AI episode. The whole-market Magnificent-7 basket does not currently meet the run-up threshold (~0 pp net two-year run-up), so the indicator is deliberately scoped to semiconductors only; applying the GSY threshold to the broad index would understate the signal because the mega-caps are already the market."]},"s4":{"name":"GSADF Explosiveness","weight":0.07,"grounding":"contested","references":["Phillips, P. C. B., Shi, S. & Yu, J. (2015). 'Testing for Multiple Bubbles: Historical Episodes of Exuberance and Collapse in the S&P 500.' International Economic Review 56(4): 1043-1078. doi:10.1111/iere.12132","Homm, U. & Breitung, J. (2012). 'Testing for Speculative Bubbles in Stock Markets: A Comparison of Alternative Methods.' Journal of Financial Econometrics 10(1): 198-231. doi:10.1093/jjfinec/nbr009","Vasilopoulos, K., Pavlidis, E. & Martinez-Garcia, E. (2022). 'exuber: Recursive Right-Tailed Unit Root Testing with R.' Journal of Statistical Software 103(10): 1-26. doi:10.18637/jss.v103.i10","Wasserstein, R. L. & Lazar, N. A. (2016). 'The ASA Statement on p-Values: Context, Process, and Purpose.' The American Statistician 70(2): 129-133. doi:10.1080/00031305.2016.1154108","Chen, Chen & Huang (2026). arXiv:2604.25826 (flag: verify at build time)","Basele, Phillips & Shi (2025). Cowles Foundation Discussion Paper d2430 (flag: verify at build time)"],"caveats":["The CONTESTED flag is currently permanent because of Chen-Chen-Huang (2026): under hump-shaped GPT fundamentals the test spuriously rejects 93-100% of the time. We expose the binary decision plus the p-value (per the ASA p-value statement, Wasserstein & Lazar 2016) rather than pretending to a graded posterior we cannot honestly calibrate."]},"s5":{"name":"Credit-Sentiment Fragility (t - 2 yr)","weight":0.13,"grounding":"literature-grounded","references":["Lopez-Salido, D., Stein, J. C. & Zakrajsek, E. (2017). 'Credit-Market Sentiment and the Business Cycle.' Quarterly Journal of Economics 132(3): 1373-1426. doi:10.1093/qje/qjx014"],"caveats":["t-2yr STRUCTURAL horizon, NOT same-month timing.","FRED truncated BAMLH0A0HYM2 to a rolling 3-year window in April 2026, so the service must persist its own history table (hy_oas_history), seeded with the 3 available years on first boot and appended daily; the percentile is only as good as accrued history — this limitation is documented in the API payload."]},"d1":{"name":"Breadth","weight":0.35,"grounding":"judgmental","references":["No published AUC/skill statistic exists for this specific mapping; the weight and anchors are expert-judgmental. General late-cycle breadth-divergence literature (market-technician breadth studies) is cited as background only."],"caveats":["Both the weight and the linear map are JUDGMENTAL. Breadth is also used in red-flag #4 with the <50%-while-index-within-2%-of-ATH condition."]},"d2":{"name":"Margin-Debt Rollover","weight":0.13,"grounding":"judgmental","references":["FINRA Rule 4521 margin statistics","Advisor Perspectives/dshort margin-debt analysis","CXO Advisory margin-debt studies"],"caveats":["CXO Advisory finds ~0.00 correlation between margin-debt changes and next-month returns and a 1-2 month lag versus stocks -> this is confirmation-only, low weight. There is a 3-4 week publication lag (published ~third week of the following month) and no true fallback source — cache and tolerate staleness (MacroMicro mirrors the same series for display only)."]},"d3":{"name":"Hyperscaler FCF Quality","weight":0.32,"grounding":"literature-grounded","references":["BIS (2026), Annual Economic Report (buildout/AI-capex context) (flag: verify at build time)","General buildout-analogy literature (railroad/telecom capital-cycle studies)"],"caveats":["Cloud-segment revenue is best-effort from XBRL segment data; when it is unavailable the total-revenue proxy is deliberately conservative (harder to trip the gate), which biases the indicator toward under-alarming — documented as such."]},"d4":{"name":"LPPLS Confidence","weight":0.2,"grounding":"literature-grounded","references":["Johansen, A., Ledoit, O. & Sornette, D. (2000). 'Crashes as Critical Points.' International Journal of Theoretical and Applied Finance 3(2): 219-255. doi:10.1142/S0219024900000115","Sornette, D. (2003). Why Stock Markets Crash: Critical Events in Complex Financial Systems. Princeton University Press.","Demirer, R., Demos, G., Gupta, R. & Sornette, D. (2019). 'On the Predictability of Stock Market Bubbles: Evidence from LPPLS Confidence Multi-Scale Indicators.' Quantitative Finance 19(5): 843-858. doi:10.1080/14697688.2018.1524154"],"caveats":["LPPLS has documented false-alarm / too-early behavior; one published evaluation reports ~90% recall but ~29% precision (it fires often in ordinary bull markets). It is also computationally heavy. Treat the sub-score as a noisy corroborator, not a stand-alone alarm."]},"v":{"name":"VIX Term-Structure Multiplier","weight":0.0,"grounding":"lagging-confirmation","references":["CBOE VIX/VIX3M methodology; term-structure regime literature.","Bollerslev, T. & Todorov, V. (2011). 'Tails, Fears, and Risk Premia.' The Journal of Finance 66(6): 2165-2211. doi:10.1111/j.1540-6261.2011.01695.x"],"caveats":["LAGGING CONFIRMATION only — never treated as a leading signal; capped so D cannot exceed 1.0."]}},"leg_references":{"trend":["Faber, M. T. (2007). 'A Quantitative Approach to Tactical Asset Allocation.' The Journal of Wealth Management 9(4): 69-79. doi:10.3905/jwm.2007.674809 (updates 2009, 2013).","Zakamulin, V. (2014). 'The Real-Life Performance of Market Timing with Moving Average and Time-Series Momentum Rules.' Journal of Asset Management 15(4): 261-278. doi:10.1057/jam.2014.25","Moskowitz, T. J., Ooi, Y. H. & Pedersen, L. H. (2012). 'Time Series Momentum.' Journal of Financial Economics 104(2): 228-250.","Huang, D., Li, J., Wang, L. & Zhou, G. (2020). 'Time Series Momentum: Is It There?' Journal of Financial Economics 135(3): 774-794."],"fast_alarm":["Bollerslev, T. & Todorov, V. (2011). 'Tails, Fears, and Risk Premia.' The Journal of Finance 66(6): 2165-2211. doi:10.1111/j.1540-6261.2011.01695.x"],"action_bands":["Alessi, L. & Detken, C. (2011). 'Quasi Real Time Early Warning Indicators for Costly Asset Price Boom/Bust Cycles: A Role for Global Liquidity.' European Journal of Political Economy 27(3): 520-533. doi:10.1016/j.ejpoleco.2011.01.003","Estrada, J. (2008). 'Black Swans and Market Timing: How Not to Generate Alpha.' The Journal of Investing 17(3): 20-34.","Estrada, J. (2009). 'Black Swans, Market Timing and the Dow.' Applied Economics Letters 16(11): 1117-1121. doi:10.1080/13504850701335517 (DJIA 1900-2006; missing the 10 best days => ~65% less terminal wealth).","Cederburg, S., O'Doherty, M. S., Wang, F. & Yan, X. (2020). 'On the Performance of Volatility-Managed Portfolios.' Journal of Financial Economics 138(1): 95-117. doi:10.1016/j.jfineco.2020.04.015"]},"falsification_criteria":["Score < 30 through a > 30% S&P drawdown beginning within 3 months -> construct falsified.","Score > 60 sustained through 24 months of > 10% annualized gains without a > 15% drawdown -> falsified.","Override fires and no > 20% drawdown within 12 months -> override falsified."],"falsification_outcomes":[],"changelog":[{"version":"v1","score":"33","notes":"linear-additive aggregation (fully compensatory); stale concentration 40.8%; HY-OAS sign inverted; LPPLS neutral placeholder."},{"version":"v2","score":"28","notes":"data fixes (concentration, HY-OAS sign, LPPLS); still fully compensatory."},{"version":"v3","score":"~40, IQR 34-47","notes":"two-block geometric aggregation + non-compensatory override + Monte Carlo median. The v2->v3 rise is the aggregation fix (partial compensability now punishes imbalance), NOT market deterioration."},{"version":"v3.0.1","score":"unchanged methodology","notes":"first-live-run bugfixes: (1) Stooq pipeline hardened — typed unavailable/CAPTCHA/limit errors, SQLite series + breadth caches with SLA reuse, >=2s pacing, retry-once-after-60s, partial breadth coverage published with note instead of dropping; (2) FINRA parser sorts by date (file is newest-first; a naive read produced a -22% YoY across the 1997 series start) + >90d cached-reading guard and 60d rollover-assertability guard; (3) GSADF data-missing now floors at the contested/stale 0.25, not the tested-not-explosive 0.05; R/exuber self-check surfaced in /readyz; (4) judgment call failures are machine-detectable (text null + error_class) with SDK param fallback; (5) LPPLS requires >=500 closes, bounded workers, 10-min hard timeout, and drop notes carry the concrete cause; (6) timezone-aware computed_at, S5 history-depth note, SKEW raw-value logging, renormalization regression test."}],"unverified_citations":["Chen, Chen & Huang (2026). arXiv:2604.25826 (verify at build time)","Basele, Phillips & Shi (2025). Cowles Foundation Discussion Paper d2430 (verify at build time)","Bank for International Settlements (2026). Annual Economic Report (verify at build time)"]},"meta":{"computed_at":null,"service_version":"3.1.0","data_freshness":{},"disclaimer":"Research, not advice.","epistemic_caveats":["NOT-A-PROBABILITY: 0-100 regime heuristic = structured expert judgment; uncalibrated.","n≈4 CALIBRATION IMPOSSIBILITY: reference class {1929,2000,2007,2021}.","REFERENCE-CLASS CAVEAT: may be rational GPT repricing (Chen-Chen-Huang 2026).","NOMINAL≠EFFECTIVE WEIGHTS: see annual PSS sensitivity script.","Service never returns 500 on upstream failure: fallback or drop+renormalize."]}}