X-Quant

Two lives

X-Quant was Kacper Saks's solo quantitative research program, and it had two lives. v1 (2024–2026) grew over two years into a ≈274K-line trading platform built around a single strategy. It was rejected on 2 August 2026 and decommissioned on 9 August 2026. v2 started from zero on the day v1 was decommissioned, under a written research constitution drawn from v1's post-mortem. It then spent seven weeks measuring whether any strategy tradeable at retail scale had a real edge. None did, and nothing in v2 was ever recorded as trading on paper or live.

What v1 taught

The v2 documents do not record a single verdict on v1. What they record are the lessons they attribute to it: an overfitting statistic computed in-sample, an open-ended search for features, a cost assumption that was never checked, and eleven engineering failure patterns. The through-line, in Kacper's own assessment: v1 had validation machinery, but not a validation process that could stop its author from believing his own numbers.

• PBO was computed in-sample; v2 allows PBO only on the out-of-sample matrix
• Free-form feature discovery, which v2 calls “an overfitting machine”; ML may only combine up to three pre-registered signals
• v1's paper-fill cost estimate was never verified
• Eleven engineering failure patterns, each now paired with a guard rule, e.g. gates that reported “clean” without measuring anything

The v2 research constitution

v2 was governed by a written constitution of eight articles; a result produced in breach of any of them is void, however good it looks. The aim was to kill bad ideas cheaply. The rules required hypotheses, thresholds and data to be frozen before the first look, and each test to be run once. Data was pinned by hashed manifests, with total-return prices and point-in-time constituents. Build, repair and verify ran in separate sessions, so no session judged its own repair, and decisions were argued out with an adversarial AI partner (Claude) whose objections stay in the register.

• Permanent attempts ledger: every configuration counts toward later tests' deflated Sharpe ratio
• PBO only out of sample; folds covering 2008, 2011, 2020 and 2022
• Fail-closed gates: zero means “not measured”, never “success”
• Append-only decision register: errors corrected by a new entry, never an edit
• Sealed 2018–2025 holdout for the equity discovery programmes; none ever qualified to open it
• No strategy without a falsifiable hypothesis: who pays, and why it isn't arbitraged away

Results on record

Every line of work was set up to die cheaply, and each one did. On 23 September 2026 the decision register recorded, in its own words (translated from Polish): “After 7 weeks and 8 programmes: 0 candidates.” The NQ futures track that followed added none; its results come from build sessions, and none was independently verified. The register's lesson from the dead programmes: cost and capacity kill earlier than the absence of a signal. Where a line showed a signal before costs, the costs a small account actually pays erased it.

• Three pre-registered US-equity event hypotheses (post-earnings drift, insider-purchase clusters, 8-K drift): all dead
• Discovery scans of 633 event and 166 indicator configurations: 0 candidates
• 57 weak signals: not a candidate when combined; re-costed at 5k, 25k and 100k USD, none survived
• Replication of Carver's trend-following system on ETFs, three versions: closed, no measurable edge at retail scale
• NQ futures (third-party rules, ensembles, ML): no tradeable candidate; one formal pass treated as void
• Final micro-futures run on 29 Sep 2026, never frozen or committed: no result under v2's rules

What it produced

The 23 September 2026 assessment that closed the ETF and equity phase names the real output: not a strategy, but the instrument — an engine, a register, a calibration method and a set of executable gates that can close a research question with evidence in about four sessions. The discipline was tested on the program's own record: a separate verify session found a wrong number in the register, it was corrected by a new entry, and the standard was tightened so that numbers enter the register only from numeric artifacts. Kacper's takeaways:

• A pass you can explain away is not a pass
• In-sample is not a result
• Numbers come from artifacts, not from prose
• Saying no is a deliverable

Status and successor

X-Quant has had no commits or register entries since 29 September 2026; neither version's repository is public. On the evening of that last commit, Kacper started its successor, Alpha Lab, a private, local research workstation. It carries over the method, not the code: Alpha Lab rebuilt its engine and integrity checks on X-Quant's patterns, checks its engine against X-Quant's golden results and reads X-Quant's raw data read-only. Its trial ledger starts from an empty genesis block, so none of the 2,976 attempts came with it. As of 11 October 2026, Alpha Lab has completed build phases F0–F3 and has no research results on real returns.