A consistent research record

Paper 01: preserve its historical scope

Building a Research Process for Energy Markets is a September 21 design paper. Its proposed monthly three-ETF baseline and ERCOT study are not descriptions of the later Tide and Roll implementations. An editorial note now makes that distinction visible at the beginning.

The review checked the accounting examples, units, time-ordering discussion, and distinctions among engineering, forecasting, and trading evidence. EIA’s API documentation, ERCOT’s price portal and December 5, 2025 RTC+B announcement, NOAA’s degree-day explanation, Alpaca’s simulation and order documentation, and the cited forecasting and performance-methodology sources support the associated discussion. The 100 MW quarter-hour example equals 25 MWh; the 30%-of-equity turnover example at five basis points equals 1.5 basis points of equity.

No new empirical result is asserted by this paper. Its stated implementation gaps and planned work remain historical descriptions. This review does not certify that every proposed engineering acceptance test has since passed.

Paper 02: correct the observation unit and cutoff

How an 88% Win Rate Lost Money originally called 42 closing lots “trades” and described the October 11 review date as though it were the observation cutoff. The underlying subtotal covers completed markets through October 9. The corrected edition distinguishes those dates and units.

The +3.61-cent equal-market measure replaces the earlier +4.4-cent closing-lot measure so the comparison uses completed markets consistently. This changes the weighting and interpretation, not the underlying cashflows. The historical 88% title is retained, with its closing-lot definition explained in the abstract and correction notice.

The October 7–8 sizing comparison reproduces −$104.30 actual versus −$20.10 under the retrospective ten-contract cap. The $53 offset-pair correction reproduces the two early-exit markets’ combined +$8.0035, rather than the −$44.9965 settlement-only figure. The published total already incorporates the correction.

The separate historical replay figures agree with the archived research output: 309 later-period decisions, 137 modeled contracts filled across 93 closing-time clusters at a 15-second delay, +1.26 cents per filled contract, and −1.17 cents per equally weighted cluster with a 3.71-cent standard error. The archived implementation check records 291 production decisions. The capacity review records 28 entry reviews, only seven with opposite best-price quantity. These checks verify the cited archived reports; the full raw-tape simulation was not rerun for this publication review. Its fills remain modeled and its uncertainty assumptions remain limitations.

Paper 03: retain the reconciled result

The Price of a Winning Streak, revision 1.1, uses 46 completed markets through October 12, including 49 closing lots and 80 exchange fills. Its public replication program was rerun successfully. The cashflow identities, daily totals, cap sensitivity, and size–outcome decompositions reproduce. The −$39.8168 total also agrees with the separate all-fill strategy attribution.

The early +$72.1662, intraday peak +$82.6662, $128.22 market-close drawdown, and retrospective ten-contract result +$2.8648 refer specifically to V3. They are not whole-account equity or desk-wide figures. The initial 31-market unit mean of +13.12 cents and later 15-market mean of −18.40 cents are descriptive, retrospectively separated outcomes.

The linked references were checked against the NBER, CESifo, author-hosted backtest-overfitting paper, and the Stanford/SSRN records for the Bartlett–O’Hara working paper. They support the methodological context; they do not independently validate this desk’s strategy. The existing PDF, CSV, and replication package are unchanged. A web note links the later desk-wide review.

Paper 04: keep the evidence types separate

Learning to Keep the Edge reports the full reviewed live Kalshi record separately from the paper equity system and Meridian’s modeled research. Its live total is −$41.8293 across 181 markets and 281 fills. V3’s −$39.8168 is a component of that total, not an additional loss. The paper’s aggregate data file and repository checks reproduce the displayed strategy and daily-product sums.

For equity, the opening handover, allocated simulated fills, and stored prices reproduce all 124 product-day rows to the nearest cent. Roll’s +$291.69 and Tide’s −$1,477.00 cover 31 sessions through November 6. They are simulated price P&L, not complete net investment returns. Borrow, dividends, financing, live fill feasibility, and other economic effects require separate reconciliation. Proposed challengers and risk rules are labeled as future tests.

How future papers should stay consistent

Each numerical claim should identify the strategy version, live/paper/research mode, observation unit, start and end dates, calculation convention, cost coverage, and source. Review dates and trade dates should be separate. Subsets should be linked to their parent totals. Historical hypotheses should remain dated, with an explicit correction or superseding result when evidence changes.

A versioned claim registry and a publication check now accompany the site source. The check validates selected arithmetic, scope wording, cross-links, and the sitemap. It does not automatically fact-check every sentence or guarantee that future papers can never conflict. New conclusions are legitimate when new evidence warrants them; unexplained contradictions and silently changed denominators are not.

The intended tone is that of an independent trader and researcher showing work: specific about methods, candid about errors, and restrained about what a short sample proves. Claims of institutional-scale profitability, audited returns, or a demonstrated enduring edge would currently exceed the evidence.

Revision record and limits

November 7, 2026: added the historical-scope note to Paper 01; corrected Paper 02 to version 1.1; reran Paper 03’s public replication without changing its data or PDF; published Paper 04 and this review; updated the site’s discovery links. AI tools assisted the review, calculations, drafting, and editing. These are author-controlled, unreviewed working papers, and this review is not an independent financial audit.