Nineteen working papers and preprints on corporate events, portfolio construction, risk measurement, and the boundary of what freely available data can be made to measure. None of them has been peer reviewed, and none of them appears on SSRN — the papers in this series that do are separate documents and are not reproduced here. What follows is made available as it is completed. Several of these papers report findings the programme afterwards withdrew, restated, or settled as null, and those are listed with the same prominence as the confirmed ones.
| Paper | Type · Date |
|---|---|
| A Measured Ladder of Pre-Announcement and In-Process Deal Probabilities, Reconstructed Entirely from Free EDGAR Filings The path a company travels from plausible acquisition candidate to completed transaction, estimated rung by rung from filings the regulator publishes at no charge. The wording of an announcement carries no measurable information at the power available, and every completion figure is stated as a lower bound because completion is detected indirectly. | RESULT August 2026 · 14 pages |
| Deal Completion as a Staged Hazard: Filing-Typed Stage Conditioning, Elapsed Time, and the Limits of a Delisting Proxy A completion probability is usually attached to a transaction. The quantity worth estimating attaches instead to a transaction at a stage of its regulatory process, and elapsed time inside a stage turns out to carry at least as much information as the stage itself. | RESULT August 2026 · 11 pages |
| An Insider-Silence Precursor for Corporate Control Events and the Boundary of Its Transfer Four measurements of insider behaviour around takeovers, and the boundary separating the one that confirmed from the three that did not. The direction inverts the premise that motivated the work: what separates is the absence of buying, not its presence, and it marks that a process exists without predicting that it will finish. | RESULT August 2026 · 11 pages |
| A Look-Ahead Convention in Intraday Price-Structure Backtests and the Size of the Edge It Manufactures A widely taught intraday entry cannot be identified until the move that defines it has finished, yet is ordinarily timestamped before it. Holding bars, instrument, geometry, costs, sizing and exits fixed and varying only that convention measures how much edge the convention invents on its own — more than any genuine effect found elsewhere in the same study. | RESULT August 2026 · 6 pages |
| The Anatomy of a Trend Sleeve's Drawdown: Attribution, Asymmetric Re-entry, and the Load-Bearing Lag A companion piece that takes apart a trend sleeve whose headline property is drawdown control rather than return. Three stacked mechanisms produce the effect, an insurance premium in calm periods pays for it, and the slow re-entry that looks like the obvious thing to fix turns out to be load-bearing — every preregistered attempt to improve on the simple construction lost. | RESULT August 2026 · 12 pages |
| When Bonds Stop Hedging: An Exogenous Inflation State Selects the Equity Hedge Instrument The standard bond hedge for an equity book fails in inflationary states. The mechanism is replicated here on entirely free data, then tested as a rule for choosing the hedge instrument: a version that fires only rarely passes every bar frozen in advance, and the continuous version that fires often does not. The failed variant is reported with the same prominence as the successful one. | REPLICATION August 2026 · 5 pages |
| The Fisher Geometry of Signal Breadth: A Cramér–Rao Ceiling for the Information Ratio The fundamental law of active management is shown to be the special case of an exact estimation-theory identity. That yields a ceiling on attainable performance, a principled rule for when to stop adding signals, and — applied to a live market-neutral book — the conclusion that more breadth requires a genuinely new data source rather than another feature of the ones already in use. | METHODS August 2026 · 10 pages |
| Silent Failure Modes in the Automated Extraction of Merger Terms from Primary Filings, and a Fail-Closed Gate That Detects Them Five ways an automated reader of merger paperwork returns a wrong number without returning an error, each documented against the filing it misread and the repair that followed. Their shared property is that every one produces a value of the expected type and a plausible size. The detection practices assembled against them refuse rather than annotate, and are shown not to nest. | METHODS August 2026 · 13 pages |
| Distinguishing a Firm Attribute from a Moment in Event-Prediction Studies: A Placebo-in-Time Design and a First Application Studies that predict takeovers compare acquired firms against firms that were not, which cannot separate the kind of company that eventually attracts a bidder from the months during which one is actually being approached. The two readings carry opposite operational content. This paper describes an arm that holds the firm fixed and moves only the observation moment, and applies it. | METHODS August 2026 · 12 pages |
| The Reachable Boundary of Free Filing Data for Event-Driven Research: A Catalogue of Measurement Limits with Named Defect Classes Filings are free and machine-readable, which is why event-driven research is described as accessible. This is the point at which that accessibility stops, catalogued across one preregistered programme, grouped into named defect classes, and stated wall by wall in terms of what breaching it would require — separating what a paid subscription buys from what only waiting buys. | METHODS August 2026 · 13 pages |
| Disciplined Harvesting of the Variance Risk Premium An unusually narrow claim: the premium earned for selling optionality is decades old and not novel, and most programmes built on it fail at the discipline rather than the thesis. What is described here is the execution machinery — a two-gate architecture that refuses to act without a documented published edge — together with a preregistered validation and the later withdrawal of the certification it produced, after a defect was found in the test engine. | METHODS August 2026 · 11 pages |
| A Self-Consistent Dealer-Flow Field: Nonlinear-Schrödinger Modelling of Spot–Options Feedback Standard option pricing is a one-way map: it assumes that hedging an option does not move the underlying. That assumption is structurally violated where dealers hold large concentrated inventories. This paper formalises the omitted back-reaction, reconstructs the resulting field from a listed option chain, and preregisters the forward test that would decide it. Methods and protocol only; results are reported separately. | METHODS August 2026 · 10 pages |
| Non-Perturbative Extension of Cornish-Fisher Value-at-Risk via Dilute Instanton Gas The standard correction for skew and fat tails in a Value-at-Risk number is a series expansion, and series expansions fail where a loss distribution has two competing modes. This paper supplies the missing term, gives the calibration route from listed option prices, and reports a preregistered settlement in which the estimator was better calibrated than the Gaussian benchmark yet fell short of the significance bar — so it remains a shadow measure and sizing stays on the classical formula. | METHODS August 2026 · 8 pages |
| Combined Crash Detection via Persistent Homology and Instanton Tunneling Two independent crash-detection channels, combined so that the joint term lights up only when both are elevated at once. The headline finding is that the resulting statistic depends heavily on where the firing threshold is set — a textbook multiple-testing hazard, disclosed as such. Two corrections to the paper's own numbers are stated at the outset, and no significance is claimed ahead of the forward quorum. | METHODS August 2026 · 8 pages |
| A Two-Axis Phase Diagram for Market-Regime Classification Regime classification usually commits to a discrete state space over scalar observables. This proposes a continuous two-axis geometric alternative whose axes are mathematically independent. An earlier internal claim of lead time over the standard method is formally retracted, and the one retrospective result reported is negative, which is why no allocation claim accompanies the scheme. | METHODS August 2026 · 8 pages |
| The Valentini H-Function as an Objective for Quantitative Portfolio Management Classical portfolio objectives summarise a return distribution by its first two moments, which misrepresents long-run performance under the multiplicative dynamics that actually govern repeated investment. The alternative proposed here measures the distance between the realised distribution and the one implied by option prices. The simplified form of the idea was settled null and its planned wiring retired; the untested form is what remains. | METHODS August 2026 · 9 pages |
| Matched Random Entry as a Control for Technical Trading Rules: Evidence from 533,275 Mechanised Setups A rule with a stop at an invalidation level and a target at a multiple of the risk produces a characteristic return distribution whether or not the entry carries information, so evaluating it against zero tests the geometry rather than the signal. Evaluated instead against a matched random entry, most of the families tested are indistinguishable from it — and the two management rules that raise the hit rate lower expectancy. | NULL RESULT August 2026 · 11 pages |
| Osmotic-Velocity Scaling at Symmetry-Breaking Transitions in Multi-Asset Markets A regression that is reproducible, stable, and significant under four independent specifications, and still not a measurement of the quantity it names. An instrument audit shows the estimator reduces algebraically to something else entirely, so the paper restates its finding in the only form the measurement supports and withdraws the reading that motivated it. The contribution that survives is methodological and negative. | NULL RESULT August 2026 · 8 pages |
| Why Path Signatures Do Not Predict Crashes as Precursors Four independent reformulations of a proposed crash-precursor signal, all disconfirmed, with the strongest of them carrying the sign opposite to the prediction. The apparent positive separation under curated windows is traced to a training-window artefact. The underlying object stays useful for describing path shape and should not be cited as a precursor in any forward specification. | NULL RESULT August 2026 · 6 pages |