01 / IMPLEMENTATION & CONTRIBUTION
What the work involves
Local implementations connect LPPLS fitting, meta-labeling, validation, backtests, and two research cockpits. The A-share service supports dated research and historical reconciliation; the global monitor fits trailing windows on indices and ETFs.
Technical depth
Nested trailing-window LPPLS fits; purged out-of-fold meta-models and calibration; PIT entry/label clocks; frozen histories; incremental data refresh with rollback and regression guards; separating timing, stock selection, sizing and hedging; serving prebuilt research artifacts through FastAPI without recomputing the engine on each request.
The project family
A_Share/openassetpricing/metalabelA_Share/openassetpricing/global_bubbleA_Share/openassetpricing/frame_attackA_Share/openassetpricing/tail_runwayA_Share/openassetpricing/pm_systemA_Share/openassetpricing/report/cbsf_factor02 / RESULTS
What came out of it
Inspectable dashboards expose market states, signals, simulated books, performance, and data freshness. Research records preserve rejected mechanisms and checks that stopped unsupported conclusions from advancing.
03 / SUPPORTING EVIDENCE
Follow the source
Implementation notes, project records, and supporting artifacts.
Source context & project scope
This pass inspected documentation and source only; it did not launch or validate current services, scheduled refresh, accounts or model results. Local documents use live for a current cockpit or paper workflow, which must not be presented as verified real-money execution. Historical return/Sortino/capacity figures are not promoted. Cost-basis histograms are vendor-imputed price/turnover proxies, not direct observations of investor intentions.
June 2026 preregistration separates meta-label selection from timing/sizing and specifies PIT availability, entry and evaluation gates. Goals are not outcomes.
SOURCE · 2026-09-17Current FastAPI source includes startup reconciliation plus date, regime, signal, candidate, simulated-book, performance and status endpoints.
SOURCE · 2026-09-17Current source includes snapshots, restoration, staged updates, frozen-window regression guards, dry verification and explicit status handling. Code presence does not establish that scheduling is active.
SOURCE · 2026-09-17Current source contains purged OOF fitting, isotonic calibration, ensemble construction and strict-OOS evaluation.
SOURCE · 2026-09-17Separate global monitor for 23 indices/ETFs using trailing nested LPPLS fits; supported by lppls_scan.py and a cockpit API.
SOURCE · 2026-09-17Dated hypothesis scorecard records failed conditional-information tests, a confirmatory gate not run, null exploratory results and killed mechanisms.
SOURCE · 2026-09-17The proposed supply-runway primary hypothesis was rejected; a later candidate remained parked pending stronger forward evidence.
SOURCE · 2026-09-17Mechanism gate distinguishes a fitted effect from an explained decomposition and quarantines the unexplained residual.
SOURCE · 2026-09-17Cost-basis factor study explicitly notes that vendor chips are imputed from price/volume, so the experiment tests predictive transforms rather than disposition-physics claims.
SOURCE · 2026-09-17