README: English rewrite with metric badges, headline results, author info · 2cb6244 · anomaly-atlas

5

contact: contact@spboucher.ai

6

data_source: hfmarketdata.io

7

created: 2026-08-12

8

−status: draft

8

+modified: 2026-08-12

9

+status: reviewed

9

10

---

10

11

12

# anomaly-atlas

12

13

14

+<p>

15

+ <a href="https://www.anomaly-atlas.io"><img alt="Live atlas" src="https://img.shields.io/badge/atlas-anomaly--atlas.io-2456c4"></a>

16

+ <img alt="Experiments" src="https://img.shields.io/badge/micro--experiments-8%2F8%20complete-1e7a4d">

17

+ <img alt="Synthetic gate" src="https://img.shields.io/badge/synthetic%20gate-29%20tests%20passing-1e7a4d">

18

+ <img alt="Findings" src="https://img.shields.io/badge/atlas%20findings-2%20%C3%97%20Level%202-2456c4">

19

+ <img alt="Hypothesis budget" src="https://img.shields.io/badge/hypothesis%20budget-22%20(declared)-8a5a1e">

20

+ <img alt="Data" src="https://img.shields.io/badge/data-hfmarketdata.io%20(sole%20source)-4a3aa7">

21

+ <img alt="Rows analyzed" src="https://img.shields.io/badge/bars%20analyzed-~68M-5d6167">

22

+ <img alt="Platform" src="https://img.shields.io/badge/platform-macOS%20%2F%20Apple%20Silicon-1a1c20">

23

+ <img alt="License" src="https://img.shields.io/badge/license-all%20rights%20reserved-8b1e1e">

24

+</p>

25

+

13

26

**Which statistical regularities in open high-frequency market data are real —

14

−surviving out-of-sample testing, multiple-comparison correction, transaction

15

−costs and robustness checks — and which are artifacts (bid-ask bounce, stale

16

−timestamps, survivorship, look-ahead, microstructure noise)?**

27

+and which are artifacts?** A systematic, pre-registered, fully reproducible

28

+research project that scans 1-minute-to-daily bars (equities, ETFs, futures,

29

+indices, FX, crypto, options chains) for mean-reversion, lead-lag, and

30

+calendar anomalies, then pushes every candidate through a three-layer

31

+validation ladder: **measured artifact nulls → multiple-testing correction →

32

+transaction costs → out-of-sample confirmation**.

17

33

18

−This repository is a living research project (charter: [CLAUDE.md](CLAUDE.md)).

19

−Data source: **[hfmarketdata.io](https://www.hfmarketdata.io) — the sole permitted

20

−source**; every result is reproducible from it alone.

34

+> **Honesty doctrine.** Every candidate anomaly is an artifact until proven

35

+> otherwise. In-sample results are never findings. Negative results are

36

+> first-class. Nothing here is investment advice or a trading system.

21

37

22

−The distinction kept sharp everywhere:

38

+## Headline results (train 2000–2016 → validation 2016–2021)

23

39

24

−```text

25

−statistically detectable in-sample

26

−≠ reproducible out-of-sample

27

−≠ robust to artifacts and specification choices

28

−≠ economically meaningful after costs

29

−```

40

+| Validation layer | Survivors |

41

+|---|---|

42

+| Searched rule universe (2 signs × every scanned cell/pair/class) | 372 |

43

+| Naive \|t\| > 1.96 | 232 (62 %) |

44

+| Benjamini–Hochberg FDR 5 % | 226 (61 %) |

45

+| Hansen SPA (data-snooping correction) | 68 (18 %) — *gross, artifact-laden* |

46

+| EDGE cost model, full half-spread per trade | **0** |

47

+| Out-of-sample (validation split, opened once) | **0** — the negative replicates |

48

+

49

+The SPA survivors carried paper Sharpes of 10–31 — bounce harvesting, not

50

+economics — and a deliberately included known artifact (the SPX→SPY "lead")

51

+passed statistical correction unharmed: **statistical correction corrects

52

+for search, not for mechanism.** Median breakeven cost: the surviving rules

53

+capture ~1 % of one half-spread per trade. Meanwhile the artifacts

54

+themselves replicate out-of-sample perfectly.

30

55

31

−- **Author:** Simon-Pierre Boucher — <contact@spboucher.ai>

32

−- **Primary platform:** Apple Silicon Mac (DuckDB + parquet out-of-core), macOS 14+

33

−- **Status:** bootstrap. Next: Phase 0.5 (Experiment A — data reality check). See `research/LOG.md`.

56

+**First atlas entries (Level 2 — corrected, OOS-confirmed, robust):**

34

57

35

−## Honesty doctrine (charter §2.1)

58

+- **F001** — *Nothing in the searched universe survives the full ladder* (negative finding, the project's headline).

59

+- **F002** — *The SPX→SPY minute-scale "lead" is index content staleness* (Fisher 1966, measured live; survives print synchronization AND SPA).

36

60

37

−Every candidate anomaly is an **artifact until proven otherwise**. In-sample

38

−results are never findings. Nothing here is trading advice, a trading system,

39

−or a claim of arbitrage; past statistical regularity does not imply future

40

−returns. Negative results are first-class published findings.

61

+Full write-up: [**P001 — The Artifact Frontier, Part I**](https://www.anomaly-atlas.io/publications/P001_artifact_frontier_part1)

62

+(every figure regenerates live from committed `results.json`).

41

63

42

−## Layout

64

+## What's in the box

43

65

44

66

| Path | Contents |

45

67

|---|---|

46

−| `research/` | Paper trail: log, data-source profile, state of the art, artifact taxonomy, gaps, ranking, methodology, bibliography |

47

−| `src/anomaly_atlas/` | Library: `data/` (single `hf_client`, cache, universe, calendars, cleaning), `stats/`, `validation/`, `atlas/`, `viz/`, `instrumentation/` |

48

−| `experiments/` | Micro-experiments A–H and candidate prototypes (hypothesis → falsification criterion → analysis) |

49

−| `benchmarks/` | Hardware manifest + synthetic series with known properties ("test the tests") |

50

−| `atlas/` | Confidence-labeled findings (Level 0–3) with provenance, via `tools/new_finding.py` |

51

−| `results/` | Raw runs, reproducible from commit + config + data-manifest + seed + manifest |

52

−| `web/` | The public atlas platform (the charter's `site/`, implemented as in localvm-research) — deployed at www.anomaly-atlas.io |

53

−| `tools/` | `check_headers.py`, `new_experiment.py`, `new_finding.py` |

68

+| `research/` | Pre-registered charter artifacts: [data-source profile](research/data_source_profile.md), [artifact taxonomy T1–T7 with measured magnitudes](research/artifact_taxonomy.md), [52-source verified bibliography](research/bibliography.md), [22-hypothesis budget](research/research_gaps.md), append-only [LOG](research/LOG.md), [publications](research/publications/) |

69

+| `src/anomaly_atlas/` | The library: single cache-first API client (never silently refetches; committed data manifest), gated statistics (VR, AC1, lead-lag, block/stationary bootstrap, BH-FDR, White RC, Hansen SPA, DSR), artifact detectors (Roll, EDGE, staleness, LOCF) |

70

+| `benchmarks/synthetic/` | **The §8.1 gate** — 29 tests on series with known properties; no detector touches real data before passing (it caught 2 real bugs) |

71

+| `experiments/micro/` | expA–expH, each with pre-registered `hypothesis.md` (falsification criterion + artifact nulls) and `analysis.md` |

72

+| `atlas/` | Confidence-labeled findings (Level 0–3) with full provenance, created only via `tools/new_finding.py` |

73

+| `web/` | The public platform (Node/Express, server-rendered SVG figures from results JSON, mobile-first, comments) |

54

74

55

−## Confidence taxonomy

75

+## Methodology in one paragraph

76

+

77

+Universes, time splits (train / validation / **sealed holdout 2022→**), and

78

+the 22-hypothesis budget were frozen in writing before any scan. Every

79

+detector passes a synthetic gate first (random walk → nothing; planted

80

+effects → recovered; pure bounce → flagged artifact). Scans report effects

81

+*net of measured artifact nulls* (variance-consistent bounce null, both-fresh

82

+synchronization, permuted calendar). Survivors face White RC / Hansen SPA

83

+over the full searched universe, then an EDGE-spread cost sweep, then the

84

+validation split — opened exactly once. All data flows through one frozen

85

+cache indexed by a committed manifest; two experiments ran with **zero

86

+network requests**.

87

+

88

+## Reproduce

89

+

90

+```bash

91

+make setup # venv + deps (macOS / Apple Silicon)

92

+make test # 29 synthetic-gate + unit tests

93

+make headers # author-header compliance

94

+python experiments/micro/expA_data_reality/benchmark.py # then B..H in order

95

+```

56

96

57

−`Level 0` in-sample only (never published) → `Level 1` corrected & OOS →

58

−`Level 2` robust → `Level 3` cost-real & confirmed on the once-touched holdout.

97

+Every result JSON embeds the hardware manifest, client instrumentation, and

98

+attribution; every finding cites its commits and the SHA-256 of the data

99

+manifest.

59

100

60

−## Reproducibility

101

+## Author

61

102

62

−Every finding: commit hash + config + data-manifest index + seed + hardware

63

−manifest (`benchmarks/hardware_manifest.py`). No result from an uncommitted tree.

103

+**Simon-Pierre Boucher** — <contact@spboucher.ai>

104

+Data source: [hfmarketdata.io](https://www.hfmarketdata.io) (sole source) ·

105

+Live atlas: [www.anomaly-atlas.io](https://www.anomaly-atlas.io)

64

106

65

−*Research on statistical properties of market data — not investment advice.*

107

+*Research on statistical properties of market data. Not investment advice,

108

+not a trading system; past statistical regularity does not imply future

109

66

110

		`@@ -5,61 +5,105 @@ author: Simon-Pierre Boucher`
5	5	contact: contact@spboucher.ai
6	6	data_source: hfmarketdata.io
7	7	created: 2026-08-12
8		−status: draft
	8	+modified: 2026-08-12
	9	+status: reviewed
9	10	---
10	11
11	12	# anomaly-atlas
12	13
	14	+<p>
	15	+ <a href="https://www.anomaly-atlas.io"><img alt="Live atlas" src="https://img.shields.io/badge/atlas-anomaly--atlas.io-2456c4"></a>
	16	+ <img alt="Experiments" src="https://img.shields.io/badge/micro--experiments-8%2F8%20complete-1e7a4d">
	17	+ <img alt="Synthetic gate" src="https://img.shields.io/badge/synthetic%20gate-29%20tests%20passing-1e7a4d">
	18	+ <img alt="Findings" src="https://img.shields.io/badge/atlas%20findings-2%20%C3%97%20Level%202-2456c4">
	19	+ <img alt="Hypothesis budget" src="https://img.shields.io/badge/hypothesis%20budget-22%20(declared)-8a5a1e">
	20	+ <img alt="Data" src="https://img.shields.io/badge/data-hfmarketdata.io%20(sole%20source)-4a3aa7">
	21	+ <img alt="Rows analyzed" src="https://img.shields.io/badge/bars%20analyzed-~68M-5d6167">
	22	+ <img alt="Platform" src="https://img.shields.io/badge/platform-macOS%20%2F%20Apple%20Silicon-1a1c20">
	23	+ <img alt="License" src="https://img.shields.io/badge/license-all%20rights%20reserved-8b1e1e">
	24	+</p>
	25	+
13	26	**Which statistical regularities in open high-frequency market data are real —
14		−surviving out-of-sample testing, multiple-comparison correction, transaction
15		−costs and robustness checks — and which are artifacts (bid-ask bounce, stale
16		−timestamps, survivorship, look-ahead, microstructure noise)?**
	27	+and which are artifacts?** A systematic, pre-registered, fully reproducible
	28	+research project that scans 1-minute-to-daily bars (equities, ETFs, futures,
	29	+indices, FX, crypto, options chains) for mean-reversion, lead-lag, and
	30	+calendar anomalies, then pushes every candidate through a three-layer
	31	+validation ladder: **measured artifact nulls → multiple-testing correction →
	32	+transaction costs → out-of-sample confirmation**.
17	33
18		−This repository is a living research project (charter: [CLAUDE.md](CLAUDE.md)).
19		−Data source: **[hfmarketdata.io](https://www.hfmarketdata.io) — the sole permitted
20		−source**; every result is reproducible from it alone.
	34	+> Honesty doctrine. Every candidate anomaly is an artifact until proven
	35	+> otherwise. In-sample results are never findings. Negative results are
	36	+> first-class. Nothing here is investment advice or a trading system.
21	37
22		−The distinction kept sharp everywhere:
	38	+## Headline results (train 2000–2016 → validation 2016–2021)
23	39
24		−```text
25		−statistically detectable in-sample
26		−≠ reproducible out-of-sample
27		−≠ robust to artifacts and specification choices
28		−≠ economically meaningful after costs
29		−```
	40	+\| Validation layer \| Survivors \|
	41	+\|---\|---\|
	42	+\| Searched rule universe (2 signs × every scanned cell/pair/class) \| 372 \|
	43	+\| Naive \\|t\\| > 1.96 \| 232 (62 %) \|
	44	+\| Benjamini–Hochberg FDR 5 % \| 226 (61 %) \|
	45	+\| Hansen SPA (data-snooping correction) \| 68 (18 %) — gross, artifact-laden \|
	46	+\| EDGE cost model, full half-spread per trade \| 0 \|
	47	+\| Out-of-sample (validation split, opened once) \| 0 — the negative replicates \|
	48	+
	49	+The SPA survivors carried paper Sharpes of 10–31 — bounce harvesting, not
	50	+economics — and a deliberately included known artifact (the SPX→SPY "lead")
	51	+passed statistical correction unharmed: **statistical correction corrects
	52	+for search, not for mechanism.** Median breakeven cost: the surviving rules
	53	+capture ~1 % of one half-spread per trade. Meanwhile the artifacts
	54	+themselves replicate out-of-sample perfectly.
30	55
31		−- Author: Simon-Pierre Boucher — <contact@spboucher.ai>
32		−- Primary platform: Apple Silicon Mac (DuckDB + parquet out-of-core), macOS 14+
33		−- Status: bootstrap. Next: Phase 0.5 (Experiment A — data reality check). See `research/LOG.md`.
	56	+First atlas entries (Level 2 — corrected, OOS-confirmed, robust):
34	57
35		−## Honesty doctrine (charter §2.1)
	58	+- F001 — Nothing in the searched universe survives the full ladder (negative finding, the project's headline).
	59	+- F002 — The SPX→SPY minute-scale "lead" is index content staleness (Fisher 1966, measured live; survives print synchronization AND SPA).
36	60
37		−Every candidate anomaly is an artifact until proven otherwise. In-sample
38		−results are never findings. Nothing here is trading advice, a trading system,
39		−or a claim of arbitrage; past statistical regularity does not imply future
40		−returns. Negative results are first-class published findings.
	61	+Full write-up: [P001 — The Artifact Frontier, Part I](https://www.anomaly-atlas.io/publications/P001_artifact_frontier_part1)
	62	+(every figure regenerates live from committed `results.json`).
41	63
42		−## Layout
	64	+## What's in the box
43	65
44	66	\| Path \| Contents \|
45	67	\|---\|---\|
46		−\| `research/` \| Paper trail: log, data-source profile, state of the art, artifact taxonomy, gaps, ranking, methodology, bibliography \|
47		−\| `src/anomaly_atlas/` \| Library: `data/` (single `hf_client`, cache, universe, calendars, cleaning), `stats/`, `validation/`, `atlas/`, `viz/`, `instrumentation/` \|
48		−\| `experiments/` \| Micro-experiments A–H and candidate prototypes (hypothesis → falsification criterion → analysis) \|
49		−\| `benchmarks/` \| Hardware manifest + synthetic series with known properties ("test the tests") \|
50		−\| `atlas/` \| Confidence-labeled findings (Level 0–3) with provenance, via `tools/new_finding.py` \|
51		−\| `results/` \| Raw runs, reproducible from commit + config + data-manifest + seed + manifest \|
52		−\| `web/` \| The public atlas platform (the charter's `site/`, implemented as in localvm-research) — deployed at www.anomaly-atlas.io \|
53		−\| `tools/` \| `check_headers.py`, `new_experiment.py`, `new_finding.py` \|
	68	+\| `research/` \| Pre-registered charter artifacts: [data-source profile](research/data_source_profile.md), [artifact taxonomy T1–T7 with measured magnitudes](research/artifact_taxonomy.md), [52-source verified bibliography](research/bibliography.md), [22-hypothesis budget](research/research_gaps.md), append-only [LOG](research/LOG.md), [publications](research/publications/) \|
	69	+\| `src/anomaly_atlas/` \| The library: single cache-first API client (never silently refetches; committed data manifest), gated statistics (VR, AC1, lead-lag, block/stationary bootstrap, BH-FDR, White RC, Hansen SPA, DSR), artifact detectors (Roll, EDGE, staleness, LOCF) \|
	70	+\| `benchmarks/synthetic/` \| The §8.1 gate — 29 tests on series with known properties; no detector touches real data before passing (it caught 2 real bugs) \|
	71	+\| `experiments/micro/` \| expA–expH, each with pre-registered `hypothesis.md` (falsification criterion + artifact nulls) and `analysis.md` \|
	72	+\| `atlas/` \| Confidence-labeled findings (Level 0–3) with full provenance, created only via `tools/new_finding.py` \|
	73	+\| `web/` \| The public platform (Node/Express, server-rendered SVG figures from results JSON, mobile-first, comments) \|
54	74
55		−## Confidence taxonomy
	75	+## Methodology in one paragraph
	76	+
	77	+Universes, time splits (train / validation / sealed holdout 2022→), and
	78	+the 22-hypothesis budget were frozen in writing before any scan. Every
	79	+detector passes a synthetic gate first (random walk → nothing; planted
	80	+effects → recovered; pure bounce → flagged artifact). Scans report effects
	81	+net of measured artifact nulls (variance-consistent bounce null, both-fresh
	82	+synchronization, permuted calendar). Survivors face White RC / Hansen SPA
	83	+over the full searched universe, then an EDGE-spread cost sweep, then the
	84	+validation split — opened exactly once. All data flows through one frozen
	85	+cache indexed by a committed manifest; two experiments ran with **zero
	86	+network requests**.
	87	+
	88	+## Reproduce
	89	+
	90	+```bash
	91	+make setup # venv + deps (macOS / Apple Silicon)
	92	+make test # 29 synthetic-gate + unit tests
	93	+make headers # author-header compliance
	94	+python experiments/micro/expA_data_reality/benchmark.py # then B..H in order
	95	+```
56	96
57		−`Level 0` in-sample only (never published) → `Level 1` corrected & OOS →
58		−`Level 2` robust → `Level 3` cost-real & confirmed on the once-touched holdout.
	97	+Every result JSON embeds the hardware manifest, client instrumentation, and
	98	+attribution; every finding cites its commits and the SHA-256 of the data
	99	+manifest.
59	100
60		−## Reproducibility
	101	+## Author
61	102
62		−Every finding: commit hash + config + data-manifest index + seed + hardware
63		−manifest (`benchmarks/hardware_manifest.py`). No result from an uncommitted tree.
	103	+Simon-Pierre Boucher — <contact@spboucher.ai>
	104	+Data source: [hfmarketdata.io](https://www.hfmarketdata.io) (sole source) ·
	105	+Live atlas: [www.anomaly-atlas.io](https://www.anomaly-atlas.io)
64	106
65		−Research on statistical properties of market data — not investment advice.
	107	+*Research on statistical properties of market data. Not investment advice,
	108	+not a trading system; past statistical regularity does not imply future
	109	+returns. All rights reserved.*
66	110