SPB Git

spb/anomaly-atlas Public License

Systematic discovery & rigorous validation of statistical anomalies in open HF market data (hfmarketdata.io) — pre-registered, artifact-null-driven, fully reproducible. Live atlas: www.anomaly-atlas.io

quant-researchmarket-microstructureanomaliesreproducibilityduckdbapple-silicon
Python 61.4% JavaScript 28.7% CSS 8.6% Shell 0.7% Makefile 0.5%

expB: artifact baselines measured (after §8.1 synthetic gate)

- synthetic generators with known ground truth + 11-test gate; the gate
  caught a real VR estimator bug (double division by q) before real data
- stats: reversion (VR/AC1/half-life), leadlag, block bootstrap;
  validation: Roll spread, excess reversion, staleness, LOCF
- expB (pre-specified protocol): 31 tickers Q1 2024 RTH 1min —
  bounce AC1 -0.009/-0.051/-0.232 by staleness tercile, VR30 down to 0.55
  with zero economics, SPY spuriously leads stale names (+0.047 at +1min,
  Spearman +0.43), SPX lags SPY by 1min (+0.065)
- artifact_taxonomy.md: 7 entries T1-T7 with measured magnitudes

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Simon-Pierre Boucher committed 5 h ago (Aug 12, 2026) parent f59e891

Showing 14 changed files with +2,024 and −42

added benchmarks/synthetic/generators.py +129 −0
@@ -0,0 +1,129 @@
1 +# =============================================================================
2 +# Project : anomaly-atlas
3 +# File : benchmarks/synthetic/generators.py
4 +# Purpose : Synthetic series with KNOWN properties — the "test the tests" set
5 +# Author : Simon-Pierre Boucher
6 +# Contact : contact@spboucher.ai
7 +# Data src : hfmarketdata.io (sole data source)
8 +# Created : 2026-08-12
9 +# Modified : 2026-08-12
10 +# Platform : macOS / Apple Silicon (arm64)
11 +# License : All rights reserved (research code)
12 +# =============================================================================
13 +"""Synthetic price/return series with planted, analytically-known properties.
14 +
15 +Charter §8.1: before any detector touches real data it must (a) find NOTHING
16 +in a pure random walk, (b) recover every planted effect, and (c) flag a pure
17 +bid-ask-bounce series as an artifact, not an anomaly. These generators are the
18 +ground truth for that gate (see test_synthetic_gate.py).
19 +
20 +All series are generated from an explicit seed; no global RNG state.
21 +"""
22 +
23 +from __future__ import annotations
24 +
25 +import numpy as np
26 +
27 +
28 +def random_walk(n: int, sigma: float = 0.001, seed: int = 0) -> np.ndarray:
29 + """Pure log-price random walk. Ground truth: VR(q)=1, AC1(returns)=0."""
30 + rng = np.random.default_rng(seed)
31 + return np.cumsum(rng.normal(0.0, sigma, n))
32 +
33 +
34 +def ou_prices(n: int, kappa: float, sigma: float = 0.001, seed: int = 0) -> np.ndarray:
35 + """Mean-reverting (Ornstein-Uhlenbeck) log-price around 0.
36 +
37 + Discrete: p_t = (1 - kappa) * p_{t-1} + eps. Ground truth half-life
38 + = ln(2) / -ln(1 - kappa); return AC1 < 0; VR(q) < 1 for q >= 2.
39 + """
40 + rng = np.random.default_rng(seed)
41 + p = np.empty(n)
42 + p[0] = 0.0
43 + eps = rng.normal(0.0, sigma, n)
44 + for t in range(1, n):
45 + p[t] = (1.0 - kappa) * p[t - 1] + eps[t]
46 + return p
47 +
48 +
49 +def roll_bounce_prices(n: int, spread: float, sigma: float = 0.001, seed: int = 0) -> np.ndarray:
50 + """Roll (1984) model: observed log-price = random-walk mid ± spread/2.
51 +
52 + Ground truth: Cov(r_t, r_{t-1}) = -spread^2/4, implied Roll spread =
53 + 2*sqrt(-cov) = spread, and the WHOLE negative AC1 is artifact.
54 + """
55 + rng = np.random.default_rng(seed)
56 + mid = np.cumsum(rng.normal(0.0, sigma, n))
57 + q = rng.choice([-1.0, 1.0], size=n)
58 + return mid + (spread / 2.0) * q
59 +
60 +
61 +def leadlag_pair(
62 + n: int, beta: float, lag: int, sigma: float = 0.001, seed: int = 0
63 +) -> tuple[np.ndarray, np.ndarray]:
64 + """Return series (x, y) where x truly leads y by `lag` steps.
65 +
66 + y_t = beta * x_{t-lag} + noise. Ground truth: cross-corr peaks at `lag`
67 + with corr ≈ beta*sd(x)/sd(y); zero at all other lags.
68 + """
69 + rng = np.random.default_rng(seed)
70 + x = rng.normal(0.0, sigma, n)
71 + noise = rng.normal(0.0, sigma, n)
72 + y = noise.copy()
73 + y[lag:] += beta * x[: n - lag]
74 + return x, y
75 +
76 +
77 +def seasonal_returns(
78 + n: int,
79 + period: int,
80 + hot_phase: int,
81 + amplitude: float,
82 + sigma: float = 0.001,
83 + seed: int = 0,
84 +) -> np.ndarray:
85 + """Returns with a planted calendar effect: mean = amplitude on one phase.
86 +
87 + Ground truth: mean(returns | t % period == hot_phase) = amplitude,
88 + all other phases 0.
89 + """
90 + rng = np.random.default_rng(seed)
91 + r = rng.normal(0.0, sigma, n)
92 + r[np.arange(n) % period == hot_phase] += amplitude
93 + return r
94 +
95 +
96 +def stale_observe(
97 + prices: np.ndarray, p_observe: float, seed: int = 0
98 +) -> tuple[np.ndarray, np.ndarray]:
99 + """Simulate an illiquid ticker: each price prints with prob p_observe,
100 + otherwise the last print is carried forward (LOCF).
101 +
102 + Returns (locf_prices, observed_mask). Ground truth: LOCF returns of a
103 + random walk gain SPURIOUS positive lag-1 autocorrelation, and a fully
104 + observed correlated series appears to LEAD the stale one.
105 + """
106 + rng = np.random.default_rng(seed)
107 + observed = rng.random(len(prices)) < p_observe
108 + observed[0] = True
109 + locf = prices.copy()
110 + for t in range(1, len(prices)):
111 + if not observed[t]:
112 + locf[t] = locf[t - 1]
113 + return locf, observed
114 +
115 +
116 +def correlated_pair(
117 + n: int, rho: float, sigma: float = 0.001, seed: int = 0
118 +) -> tuple[np.ndarray, np.ndarray]:
119 + """Two random-walk log-prices with contemporaneously correlated innovations.
120 +
121 + Ground truth: corr(r_x, r_y) = rho at lag 0, zero at every nonzero lag —
122 + any measured lead-lag after LOCF is pure artifact.
123 + """
124 + rng = np.random.default_rng(seed)
125 + z1 = rng.normal(0.0, sigma, n)
126 + z2 = rng.normal(0.0, sigma, n)
127 + rx = z1
128 + ry = rho * z1 + np.sqrt(1.0 - rho**2) * z2
129 + return np.cumsum(rx), np.cumsum(ry)
added benchmarks/synthetic/test_synthetic_gate.py +176 −0
@@ -0,0 +1,176 @@
1 +# =============================================================================
2 +# Project : anomaly-atlas
3 +# File : benchmarks/synthetic/test_synthetic_gate.py
4 +# Purpose : §8.1 gate — detectors must pass synthetic ground truth first
5 +# Author : Simon-Pierre Boucher
6 +# Contact : contact@spboucher.ai
7 +# Data src : hfmarketdata.io (sole data source)
8 +# Created : 2026-08-12
9 +# Modified : 2026-08-12
10 +# Platform : macOS / Apple Silicon (arm64)
11 +# License : All rights reserved (research code)
12 +# =============================================================================
13 +"""The mandatory gate of charter §8.1, as executable tests:
14 +
15 + 1. pure random walk -> NO anomaly may be detected
16 + 2. planted mean-reversion -> must be recovered (incl. half-life)
17 + 3. planted lead-lag -> must be recovered at the right lag
18 + 4. planted calendar effect -> must be recovered on the right phase
19 + 5. pure bid-ask bounce -> must be flagged as ARTIFACT, not anomaly
20 + 6. staleness (LOCF) -> must manufacture the documented artifacts
21 +
22 +A detector that fails any of these is broken and must not touch real data.
23 +Multi-seed checks use fixed seed lists — fully deterministic.
24 +"""
25 +
26 +from __future__ import annotations
27 +
28 +import sys
29 +from pathlib import Path
30 +
31 +import numpy as np
32 +
33 +sys.path.insert(0, str(Path(__file__).resolve().parent))
34 +
35 +from generators import ( # noqa: E402
36 + correlated_pair,
37 + leadlag_pair,
38 + ou_prices,
39 + random_walk,
40 + roll_bounce_prices,
41 + seasonal_returns,
42 + stale_observe,
43 +)
44 +
45 +from anomaly_atlas.stats.bootstrap import moving_block_bootstrap, percentile_ci
46 +from anomaly_atlas.stats.leadlag import lagged_xcorr, leadlag_asymmetry, peak_lag
47 +from anomaly_atlas.stats.reversion import ac1, half_life, variance_ratio
48 +from anomaly_atlas.validation.artifacts import (
49 + excess_reversion,
50 + locf_fill,
51 + roll_spread,
52 + staleness_ratio,
53 +)
54 +
55 +N = 100_000
56 +SEEDS = [1, 2, 3, 4, 5]
57 +
58 +
59 +# ----------------------------------------------------- 1. random walk: nothing
60 +def test_random_walk_triggers_nothing():
61 + for seed in SEEDS:
62 + r = np.diff(random_walk(N, seed=seed))
63 + assert abs(ac1(r)) < 0.02
64 + assert abs(variance_ratio(r, 5) - 1.0) < 0.05
65 + assert abs(variance_ratio(r, 30) - 1.0) < 0.12
66 + assert half_life(random_walk(N, seed=seed)) > 5_000 # effectively none
67 +
68 +
69 +def test_random_walk_ac1_inside_its_bootstrap_ci():
70 + r = np.diff(random_walk(N, seed=7))
71 + boot = moving_block_bootstrap(r, ac1, block=390, n_boot=200, seed=7)
72 + lo, hi = percentile_ci(boot)
73 + assert lo < 0.0 < hi # zero is inside the CI: no detection
74 +
75 +
76 +def test_independent_walks_show_no_leadlag():
77 + for seed in SEEDS:
78 + x = np.diff(random_walk(N, seed=seed))
79 + y = np.diff(random_walk(N, seed=seed + 100))
80 + xc = lagged_xcorr(x, y, 5)
81 + assert max(abs(v) for v in xc.values()) < 0.02
82 + assert abs(leadlag_asymmetry(xc)) < 0.05
83 +
84 +
85 +# ------------------------------------------- 2. planted reversion is recovered
86 +def test_ou_reversion_recovered_with_half_life():
87 + kappa = 0.02 # true half-life = ln2 / -ln(0.98) ≈ 34.3 bars
88 + true_hl = np.log(2) / -np.log(1 - kappa)
89 + for seed in SEEDS:
90 + p = ou_prices(N, kappa=kappa, seed=seed)
91 + r = np.diff(p)
92 + assert ac1(r) < -0.005
93 + assert variance_ratio(r, 30) < 0.9
94 + assert abs(half_life(p) - true_hl) / true_hl < 0.25
95 +
96 +
97 +# --------------------------------------------- 3. planted lead-lag is recovered
98 +def test_planted_leadlag_recovered_at_correct_lag():
99 + for seed in SEEDS:
100 + x, y = leadlag_pair(N, beta=0.3, lag=2, seed=seed)
101 + xc = lagged_xcorr(x, y, 5)
102 + assert peak_lag(xc) == 2
103 + assert xc[2] > 0.2
104 + assert abs(xc[1]) < 0.02 and abs(xc[3]) < 0.02
105 +
106 +
107 +# --------------------------------------- 4. planted calendar effect is recovered
108 +def test_planted_seasonal_effect_recovered_on_right_phase():
109 + period, hot, amp = 5, 3, 0.0005
110 + for seed in SEEDS:
111 + r = seasonal_returns(N, period, hot, amp, seed=seed)
112 + phase_means = [r[np.arange(N) % period == k].mean() for k in range(period)]
113 + assert np.argmax(phase_means) == hot
114 + assert abs(phase_means[hot] - amp) < amp * 0.2
115 + rest = [m for k, m in enumerate(phase_means) if k != hot]
116 + assert max(abs(m) for m in rest) < amp * 0.2
117 +
118 +
119 +# ------------------------------------- 5. pure bounce is an artifact, not a find
120 +def test_roll_spread_recovers_planted_spread():
121 + spread = 0.002
122 + for seed in SEEDS:
123 + p = roll_bounce_prices(N, spread=spread, seed=seed)
124 + est = roll_spread(np.diff(p))
125 + assert abs(est - spread) / spread < 0.10
126 +
127 +
128 +def test_pure_bounce_reversion_vanishes_after_artifact_adjustment():
129 + spread = 0.002
130 + for seed in SEEDS:
131 + p = roll_bounce_prices(N, spread=spread, seed=seed)
132 + r = np.diff(p)
133 + assert ac1(r) < -0.2 # naive detector screams "mean reversion!"
134 + # ...but the excess over the bounce null (true spread supplied) is ~0
135 + assert abs(excess_reversion(r, spread)) < 0.03
136 +
137 +
138 +def test_true_reversion_survives_artifact_adjustment():
139 + # OU + bounce: after removing the bounce share, reversion must REMAIN
140 + kappa, spread = 0.05, 0.001
141 + for seed in SEEDS:
142 + mid = ou_prices(N, kappa=kappa, seed=seed)
143 + rng = np.random.default_rng(seed + 999)
144 + p = mid + (spread / 2.0) * rng.choice([-1.0, 1.0], size=N)
145 + r = np.diff(p)
146 + assert excess_reversion(r, spread) < -0.01
147 +
148 +
149 +# ------------------------------------------------ 6. staleness manufactures lies
150 +def test_locf_creates_spurious_positive_autocorrelation():
151 + for seed in SEEDS:
152 + p = random_walk(N, seed=seed)
153 + locf, mask = stale_observe(p, p_observe=0.3, seed=seed)
154 + r = np.diff(locf)
155 + assert ac1(np.diff(p)) < 0.02 # underlying: nothing
156 + # LOCF returns of a pure walk: AC1 pushed NEGATIVE at lag 1 grid steps
157 + # is not the failure mode; the artifact is CROSS-serial (next test) and
158 + # a big mass of zero returns. Document the zero-mass here:
159 + assert (r == 0).mean() > 0.5
160 + assert staleness_ratio(mask) > 0.6
161 +
162 +
163 +def test_locf_makes_fresh_series_appear_to_lead_stale_one():
164 + for seed in SEEDS:
165 + px, py = correlated_pair(N, rho=0.7, seed=seed)
166 + # underlying returns: correlation only at lag 0
167 + xc_true = lagged_xcorr(np.diff(px), np.diff(py), 3)
168 + assert abs(xc_true[1]) < 0.02
169 + # y observed sparsely, LOCF-joined on the grid: x now "leads" y
170 + mask = np.random.default_rng(seed).random(N) < 0.3
171 + mask[0] = True
172 + y_locf = locf_fill(py, mask)
173 + xc = lagged_xcorr(np.diff(px), np.diff(y_locf), 3)
174 + assert xc[1] > 0.10 # spurious lead of the fresh series
175 + assert leadlag_asymmetry(xc) > 0.1
176 + assert peak_lag({k: v for k, v in xc.items() if k != 0}) == 1
modified data_manifest/index.jsonl +47 −0
@@ -52,3 +52,50 @@
52 52 {"author": "Simon-Pierre Boucher", "cache_key": "1f72e7b9a118cc59ee59baafdfd5ad65", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/MSFT", "fetched_utc": "2026-08-12T05:44:40Z", "first": "2025-10-01", "last": "2025-10-03", "params": {"adjustment": "adj_splitdiv", "end": "2025-10-05", "start": "2025-10-01", "timeframe": "1day"}, "rows": 3, "sha256": "54892bb0c9a6bda2b3e56338602d65e3f281d6ba1807028ac60a66352ea04edf"}
53 53 {"author": "Simon-Pierre Boucher", "cache_key": "98078fa812ec7189fd243149762722aa", "data_source": "hfmarketdata.io", "endpoint": "/v1/options/expirations/SPY", "fetched_utc": "2026-08-12T05:44:40Z", "first": null, "last": null, "params": {"trade_date": "2026-06-15"}, "rows": null, "sha256": "84ddfee82f215136f022f1b375895b964894f516d732dfb8f85baa3ab5b83fbc"}
54 54 {"author": "Simon-Pierre Boucher", "cache_key": "b2ce81ddb1aa764e164205ca08aec02d", "data_source": "hfmarketdata.io", "endpoint": "/v1/options/chain/SPY", "fetched_utc": "2026-08-12T05:44:40Z", "first": "2026-06-15", "last": "2026-06-15", "params": {"limit": 2, "trade_date": "2026-06-15"}, "rows": 2, "sha256": "cc63767f8abc43b71f5618e19bfd92b725e9b377d7aa9d828e05bd133d426d62"}
55 +{"author": "Simon-Pierre Boucher", "cache_key": "57193251505095d115d9821258d84799", "data_source": "hfmarketdata.io", "endpoint": "/v1/stock/tickers", "fetched_utc": "2026-08-12T05:56:02Z", "first": null, "last": null, "params": {"adjustment": "adj_split", "limit": 10000, "timeframe": "1min"}, "rows": null, "sha256": "032cb0a1f5518faf65f87e86397aa1d65c746c0fed24bae4fe4e025276c274ef"}
56 +{"author": "Simon-Pierre Boucher", "cache_key": "6da965ad3c79e7a0875bf8aaaff66203", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/AAPL", "fetched_utc": "2026-08-12T05:56:03Z", "first": "2024-01-02 04:00:00", "last": "2024-03-28 19:59:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 47013, "sha256": "c3798cec4e63e79a0b3757adaa5262b0297d6f21f6e231e6a40c52e0f2a917f9"}
57 +{"author": "Simon-Pierre Boucher", "cache_key": "d26fad0ebd498f478da02af5e937eab6", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/MSFT", "fetched_utc": "2026-08-12T05:56:04Z", "first": "2024-01-02 04:02:00", "last": "2024-03-28 19:59:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 39284, "sha256": "351766261c7200865ca73ee5493a33ec0861d7a49419cff7e457707a8360cda8"}
58 +{"author": "Simon-Pierre Boucher", "cache_key": "b47ded4777110c562a0ea43f8a2429dc", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/NVDA", "fetched_utc": "2026-08-12T05:56:06Z", "first": "2024-01-02 04:00:00", "last": "2024-03-25 16:07:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 50000, "sha256": "941d8e1258b6d4963432600d39b67f06e52f975b8b5b4da6139841ec4be2dfb2"}
59 +{"author": "Simon-Pierre Boucher", "cache_key": "0b343a86c050098a0bc59fd89c6a2e5e", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/NVDA", "fetched_utc": "2026-08-12T05:56:06Z", "first": "2024-03-25 16:07:00", "last": "2024-03-28 19:59:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-03-25 16:07:00", "timeframe": "1min"}, "rows": 2686, "sha256": "ac6970d4679dbfc05d1af5ea93e29f1558571f5d7b1168dd1b56677d9a2efc1b"}
60 +{"author": "Simon-Pierre Boucher", "cache_key": "9e2b82de1e467b9ce2a1e4513b62f8d3", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/AMZN", "fetched_utc": "2026-08-12T05:56:08Z", "first": "2024-01-02 04:00:00", "last": "2024-03-28 19:59:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 44415, "sha256": "cee6532b43e2ed9475b73c89a976d7dcb6cb2685b5cfa05b87096a1a78da4daa"}
61 +{"author": "Simon-Pierre Boucher", "cache_key": "30562c78f93f7091bd62f37ac799b37c", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/GOOGL", "fetched_utc": "2026-08-12T05:56:09Z", "first": "2024-01-02 04:00:00", "last": "2024-03-28 19:59:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 40036, "sha256": "6f41850f3e631287e97059c9328974ae01c666eb02aa3b94e6730f5c53fc31f9"}
62 +{"author": "Simon-Pierre Boucher", "cache_key": "4b7bf27e6c36d034e0ffc67710041bed", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/META", "fetched_utc": "2026-08-12T05:56:10Z", "first": "2024-01-02 04:07:00", "last": "2024-03-28 19:57:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 36679, "sha256": "c5490da0c6d966ecc57a9f1b127ff83c4107e0636c4595616105de1ad471e998"}
63 +{"author": "Simon-Pierre Boucher", "cache_key": "1e20301ea13e362f0d1bd5f4dcbb7a28", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/TSLA", "fetched_utc": "2026-08-12T05:56:11Z", "first": "2024-01-02 04:00:00", "last": "2024-03-20 18:29:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 50000, "sha256": "5004c0c72c6ade5bda1c1ff2f6e7ecc41854befb71c8defebf43fc1e952c0492"}
64 +{"author": "Simon-Pierre Boucher", "cache_key": "8af5eef323872e457d211808f77a73cb", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/TSLA", "fetched_utc": "2026-08-12T05:56:12Z", "first": "2024-03-20 18:29:00", "last": "2024-03-28 19:59:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-03-20 18:29:00", "timeframe": "1min"}, "rows": 5515, "sha256": "350519fbc28951461b623154c50ae14429686d97df0eef354181314afe8d9385"}
65 +{"author": "Simon-Pierre Boucher", "cache_key": "13b78e911df698d059161bf82f97d698", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/JPM", "fetched_utc": "2026-08-12T05:56:13Z", "first": "2024-01-02 04:03:00", "last": "2024-03-28 19:46:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 26645, "sha256": "d1383a06a9c8f047d23ffd71e0fa5913b1fb71c06df6711f3143ff27fd844714"}
66 +{"author": "Simon-Pierre Boucher", "cache_key": "4ae363aae108f48a4febf58acb28daf2", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/XOM", "fetched_utc": "2026-08-12T05:56:13Z", "first": "2024-01-02 04:00:00", "last": "2024-03-28 19:46:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 29547, "sha256": "f5ddcad08f5f5cb9f65d34e6925d2f2fd7a46ddf74bf37fdd3d6bc13d6996f50"}
67 +{"author": "Simon-Pierre Boucher", "cache_key": "b873f50315fc2655ca598a0e2ddb9823", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/UNH", "fetched_utc": "2026-08-12T05:56:14Z", "first": "2024-01-02 08:00:00", "last": "2024-03-28 19:26:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 25046, "sha256": "1248f316966a11334ca8fe4db38b0153403ef8cc1ddf2535e0e751b97496d1df"}
68 +{"author": "Simon-Pierre Boucher", "cache_key": "6728ecaa8020e9fa2802371e2213303c", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/etf/SPY", "fetched_utc": "2026-08-12T05:56:16Z", "first": "2024-01-02 04:00:00", "last": "2024-03-28 19:59:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 48128, "sha256": "81ccf6af42aee4933d8b1d935a715c0813a6c953092c58c9ca2c2823a72235d1"}
69 +{"author": "Simon-Pierre Boucher", "cache_key": "fc436a19a56c751cfca0771b9697dbb2", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/etf/QQQ", "fetched_utc": "2026-08-12T05:56:17Z", "first": "2024-01-02 04:00:00", "last": "2024-03-26 15:21:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 50000, "sha256": "28eebbcfe1b74e92cb9699374066c63cb7e4e164de00fd2183933c39ce931461"}
70 +{"author": "Simon-Pierre Boucher", "cache_key": "97754686f049c1e742200547469df859", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/etf/QQQ", "fetched_utc": "2026-08-12T05:56:17Z", "first": "2024-03-26 15:21:00", "last": "2024-03-28 19:59:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-03-26 15:21:00", "timeframe": "1min"}, "rows": 1834, "sha256": "6cc38c6f40bd383495f8c18250fe4f4107c52a0a8a1c17a34bd9fdb7d42a4b4e"}
71 +{"author": "Simon-Pierre Boucher", "cache_key": "93a7adbc0ac6f334920476e4fe0b83ec", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/ATRO", "fetched_utc": "2026-08-12T05:56:18Z", "first": "2024-01-02 09:30:00", "last": "2024-03-28 16:00:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 7444, "sha256": "1a41d1281a9fd1cf1a5062abcdaecb8ebf0eb61ebc7c084471e420cf4e75e757"}
72 +{"author": "Simon-Pierre Boucher", "cache_key": "238f080feba962d5bf39dd44016f0938", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/AUVI", "fetched_utc": "2026-08-12T05:56:18Z", "first": "2024-01-02 04:53:00", "last": "2024-03-28 19:34:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 10200, "sha256": "1d1cb461a53922fe481c8b8e984fb4ef53ec5e62dfee121f93a82eb079d9c191"}
73 +{"author": "Simon-Pierre Boucher", "cache_key": "1d654a3efff19034bcb6f9aee5807410", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/AXDX", "fetched_utc": "2026-08-12T05:56:18Z", "first": "2024-01-02 08:00:00", "last": "2024-03-28 16:05:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 4142, "sha256": "fbc451ad92fce083b33efc0447692b6e88cb65f3a9209f8e3d6271577e29dfad"}
74 +{"author": "Simon-Pierre Boucher", "cache_key": "d3b66988e16c0258b14b3e1b7fff1bb4", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/BKE", "fetched_utc": "2026-08-12T05:56:19Z", "first": "2024-01-02 09:30:00", "last": "2024-03-28 16:20:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 17146, "sha256": "93d22e4ec35d512177ef6cae24647b529df0d1c42326cf3b45a44e3592f542d1"}
75 +{"author": "Simon-Pierre Boucher", "cache_key": "1b8bc68e1d2199ccfe83459e85cd9f61", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/CECO", "fetched_utc": "2026-08-12T05:56:19Z", "first": "2024-01-02 09:30:00", "last": "2024-03-28 16:00:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 11324, "sha256": "a4fcc21e6b6f5b58f4226b4f5eb0d9dd34c0b01ade2f622a7428b977a151ce6d"}
76 +{"author": "Simon-Pierre Boucher", "cache_key": "8aae75bdf25ad09ca9c339b85068537d", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/CKX", "fetched_utc": "2026-08-12T05:56:19Z", "first": "2024-01-02 12:17:00", "last": "2024-03-28 15:57:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 241, "sha256": "10486e8127e6a78f1f750b18aee073f5b3d6d65b532b25874e46278d3d929b79"}
77 +{"author": "Simon-Pierre Boucher", "cache_key": "3ff2dc4c9bacfb07bfa39ea364085d78", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/GCL", "fetched_utc": "2026-08-12T05:56:20Z", "first": "2024-01-11 10:06:00", "last": "2024-03-25 11:47:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 44, "sha256": "1bf5b2ff559f35089c452dbd64c338478a77c3a469ab3799f25b12143ee04679"}
78 +{"author": "Simon-Pierre Boucher", "cache_key": "5e109c654808ac673f4ac3a4d06f6aaf", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/GRO", "fetched_utc": "2026-08-12T05:56:20Z", "first": null, "last": null, "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 0, "sha256": "8062687d3dbe7963af9b2fa82e28bdbf5ce3aee72af33fa08113f6e5bf11df4e"}
79 +{"author": "Simon-Pierre Boucher", "cache_key": "501c5e8944e9134f777351d9b7817327", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/HPE", "fetched_utc": "2026-08-12T05:56:21Z", "first": "2024-01-02 04:05:00", "last": "2024-03-28 16:22:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 26409, "sha256": "45cceca84f2bf737c7608d3639e659f948d1eb67262d9604fc7e16ce43945200"}
80 +{"author": "Simon-Pierre Boucher", "cache_key": "1aa56bb61edd614ee1c2bb3029c7292a", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/HTD", "fetched_utc": "2026-08-12T05:56:21Z", "first": "2024-01-02 09:30:00", "last": "2024-03-28 16:00:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 7436, "sha256": "b2d26e1cc068af78ee634ef0141e9a816b49aabc31b51e963ff5aa16c207768e"}
81 +{"author": "Simon-Pierre Boucher", "cache_key": "33aca3fe78b93c15bdbace2bf197f3f8", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/HWM", "fetched_utc": "2026-08-12T05:56:22Z", "first": "2024-01-02 08:00:00", "last": "2024-03-28 16:20:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 24023, "sha256": "ea220462ee70fa064a7753ef889fbac66c18e4f5ca5f4a6665f81934eea6bae4"}
82 +{"author": "Simon-Pierre Boucher", "cache_key": "4e4c68c953dfacd48f8c62cb8b5bbaa7", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/ICUI", "fetched_utc": "2026-08-12T05:56:22Z", "first": "2024-01-02 09:30:00", "last": "2024-03-28 16:02:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 12133, "sha256": "c2c1952a0879b2c87aaf3941f4a9d39170d972acab877cd7873fcd07c396cf2a"}
83 +{"author": "Simon-Pierre Boucher", "cache_key": "bd01b93fc349c1cbcf84303066980cd3", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/KEY.K", "fetched_utc": "2026-08-12T05:56:22Z", "first": "2024-01-02 09:44:00", "last": "2024-03-28 16:00:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 3202, "sha256": "fb5a39b256ddb08eb23883deacbfbf7b017673c32140bf879997b0f732e68341"}
84 +{"author": "Simon-Pierre Boucher", "cache_key": "c6e25c9763e17026cf5e77242c4ddb14", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/KSPI", "fetched_utc": "2026-08-12T05:56:23Z", "first": "2024-01-22 06:55:00", "last": "2024-03-28 16:00:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 10172, "sha256": "13ee966b9e0c91db8c60f24ddb8205c5595ae8901b34cee8a63fac7c6af547d2"}
85 +{"author": "Simon-Pierre Boucher", "cache_key": "9df1af4190dd161b545d783203a933da", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/LENZ", "fetched_utc": "2026-08-12T05:56:23Z", "first": "2024-01-02 08:30:00", "last": "2024-03-28 16:04:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 8308, "sha256": "ae3031a1f694e859277b4c254a9d857ec1fcae02d2050907809522c89842647e"}
86 +{"author": "Simon-Pierre Boucher", "cache_key": "71d2445f7392860a6f37c0412144d279", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/LTSL", "fetched_utc": "2026-08-12T05:56:23Z", "first": "2024-01-08 11:41:00", "last": "2024-03-28 15:36:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 122, "sha256": "3b57c6a65eaab5260245399d7c1ec2fcb12558028cf147bbf9f6bfe20d277d08"}
87 +{"author": "Simon-Pierre Boucher", "cache_key": "8cc13085f63434268be94cd37b1babb7", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/NSTS", "fetched_utc": "2026-08-12T05:56:23Z", "first": "2024-01-03 15:00:00", "last": "2024-03-28 16:00:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 255, "sha256": "aedfb55142adb7b8de418eb36a3288c705c6c9b8156eebdb4fff72823e993c8e"}
88 +{"author": "Simon-Pierre Boucher", "cache_key": "df097b931f431d8b8899af488b02f0b6", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/NWAX", "fetched_utc": "2026-08-12T05:56:24Z", "first": null, "last": null, "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 0, "sha256": "8062687d3dbe7963af9b2fa82e28bdbf5ce3aee72af33fa08113f6e5bf11df4e"}
89 +{"author": "Simon-Pierre Boucher", "cache_key": "688b80931b78ad8c6acfbac669cd1a7f", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/PBYI", "fetched_utc": "2026-08-12T05:56:24Z", "first": "2024-01-02 08:11:00", "last": "2024-03-28 17:19:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 16569, "sha256": "9a1044c49f888f7cae2dc495fca415722072d49f319e9af861fa105a3c42e875"}
90 +{"author": "Simon-Pierre Boucher", "cache_key": "075ac2c34491c2a583c1b6aac079ea33", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/PLNT", "fetched_utc": "2026-08-12T05:56:25Z", "first": "2024-01-02 07:55:00", "last": "2024-03-28 16:30:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 23291, "sha256": "3b5b52266a56332281ef275d5569740c806df91403897438361dbcd580f45fc1"}
91 +{"author": "Simon-Pierre Boucher", "cache_key": "19479bedf5c7a972d44a1e687895ce5b", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/PSA.G", "fetched_utc": "2026-08-12T05:56:25Z", "first": "2024-01-02 09:30:00", "last": "2024-03-28 16:00:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 1581, "sha256": "ea85a2704b7ecde1c408af8e4da8412784ba04251648b699f206967c8a809a95"}
92 +{"author": "Simon-Pierre Boucher", "cache_key": "9d63538e8455fa3108091b893d722950", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/RDZN", "fetched_utc": "2026-08-12T05:56:25Z", "first": "2024-01-02 08:00:00", "last": "2024-03-28 15:56:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 955, "sha256": "c918edc2142d679bbd880ce88ae1b4fe6d5f477349f3276066694c86bce84352"}
93 +{"author": "Simon-Pierre Boucher", "cache_key": "0ec4f590a02d68cf7d0337e89f3a1baa", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/RITM.B", "fetched_utc": "2026-08-12T05:56:26Z", "first": "2024-01-02 09:40:00", "last": "2024-03-28 16:00:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 2447, "sha256": "7a893130a1e57329626bfb66a6a9f5481811fa18732b7d08f845b7d37d1c1bc1"}
94 +{"author": "Simon-Pierre Boucher", "cache_key": "853e4a2cef1474fafec4e5a3b889cb3e", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/RPM", "fetched_utc": "2026-08-12T05:56:26Z", "first": "2024-01-02 09:30:00", "last": "2024-03-28 16:20:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 18682, "sha256": "6b6eb4d7ebce44fb5658b21bf61fdaf2d8e80750e846ec4d73faddc43b9c216e"}
95 +{"author": "Simon-Pierre Boucher", "cache_key": "7a9cc362158bc889c5a511f7582f4276", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/RUM", "fetched_utc": "2026-08-12T05:56:27Z", "first": "2024-01-02 07:01:00", "last": "2024-03-28 19:41:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 30567, "sha256": "59008c590811876cf576c6f09ce2cf94a539a01a26747badfed717e4c20a9dc1"}
96 +{"author": "Simon-Pierre Boucher", "cache_key": "6020217db7c98bfe1a82708406e50369", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/SLF", "fetched_utc": "2026-08-12T05:56:28Z", "first": "2024-01-02 09:30:00", "last": "2024-03-28 16:00:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 19536, "sha256": "20cfc5f4ee1b20c6336a19a1a895309750f06e12ec18dd92148e38f7640abbd5"}
97 +{"author": "Simon-Pierre Boucher", "cache_key": "f38e9c35f15ad96299c978a3acb5e068", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/SPB", "fetched_utc": "2026-08-12T05:56:28Z", "first": "2024-01-02 09:30:00", "last": "2024-03-28 16:05:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 15947, "sha256": "a78aed00d065b23860d4019cf031f02a88c217e2f546e14079be65b978334275"}
98 +{"author": "Simon-Pierre Boucher", "cache_key": "5d5e9ea1f442098aaf202a1c87d69c54", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/STRRP", "fetched_utc": "2026-08-12T05:56:28Z", "first": "2024-01-03 14:26:00", "last": "2024-03-26 15:28:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 174, "sha256": "9d3918a87642c2bc1086d37ddf0c79a1c13c959f3b88de55e9d4ed01e1591423"}
99 +{"author": "Simon-Pierre Boucher", "cache_key": "a2d1fc0ce26f55ac2dbcb8a56f027e70", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/USGOW", "fetched_utc": "2026-08-12T05:56:29Z", "first": "2024-01-02 10:52:00", "last": "2024-03-28 15:52:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 306, "sha256": "1c4689daece65d79483cc48b51dc7452ceaaaef134cdbce98cde5a97acbe4dc5"}
100 +{"author": "Simon-Pierre Boucher", "cache_key": "6641c48ba15e0036b7af55422eb3e0f8", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/stock/WTFCM", "fetched_utc": "2026-08-12T05:56:29Z", "first": "2024-01-02 09:30:00", "last": "2024-03-28 16:00:00", "params": {"adjustment": "adj_split", "end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 1545, "sha256": "85e69461dee2e1efb17b457d5770b7a9b4dc47695345fc3789f8ea3f8a6fecae"}
101 +{"author": "Simon-Pierre Boucher", "cache_key": "a13526fe4efaf465254003732bca9ecf", "data_source": "hfmarketdata.io", "endpoint": "/v1/bars/index/SPX", "fetched_utc": "2026-08-12T05:56:30Z", "first": "2024-01-02 09:30:00", "last": "2024-03-28 16:05:00", "params": {"end": "2024-04-01", "limit": 50000, "order": "asc", "start": "2024-01-02", "timeframe": "1min"}, "rows": 24156, "sha256": "24aeff152902c703c832f91ba7ff331776643cc8cbaf9b9d36194f846736ade4"}
modified experiments/micro/expB_artifact_baselines/README.md +1 −1
@@ -12,4 +12,4 @@ status: draft
12 12
13 13 Artifact baselines: null distributions of bounce, staleness, non-synchronous lead-lag
14 14
15 −Status: scaffolded 2026-08-12, not yet run.
15 +Status: **completed 2026-08-12** — see analysis.md; null magnitudes recorded in research/artifact_taxonomy.md.
modified experiments/micro/expB_artifact_baselines/analysis.md +46 −2
@@ -5,9 +5,53 @@ 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 # Analysis — expB_artifact_baselines
12 13
13 −*To be written after results exist. Must include the seven-field block and the evidence standard of CLAUDE.md §10 (never report an in-sample number as a finding).*
14 +Run: `results/expB_artifact_baselines/20260812T055602Z/results.json`
15 +(hardware manifest embedded; 47 network requests, 795 185 rows, protocol
16 +pre-specified in hypothesis.md; detectors passed the §8.1 synthetic gate —
17 +11 tests — before touching this data).
18 +
19 +## The measured artifact nulls (Q1 2024, RTH 1min, 31 tickers)
20 +
21 +| Staleness tercile | median staleness | median AC1 | Roll rel. spread | VR(30) | SPY leads +1min |
22 +|---|---|---|---|---|---|
23 +| fresh | 0.000 | −0.009 | 1.4 bp | 0.972 | +0.005 |
24 +| mid | 0.107 | −0.051 | 3.8 bp | 0.901 | +0.047 |
25 +| stale | 0.691 | −0.232 | 17.0 bp | 0.547 | +0.021 |
26 +
27 +* **Bounce/staleness dominate naive reversion metrics.** With *zero* planted
28 + economics, illiquid names show VR(30) = 0.55 and AC1 = −0.23 (extreme:
29 + RITM.B, 90 % stale minutes, VR30 = 0.35). Any reversion scan that does not
30 + clear these levels for its liquidity bucket is measuring market plumbing.
31 +* **Mega-caps show no measurable bounce at 1min**: AAPL/SPY/NVDA AC1 CIs
32 + cover 0 and the Roll estimator is undefined (positive lag-1 autocov) —
33 + the bounce null is liquidity-dependent, not universal.
34 +* **The stale-price lead-lag artifact is real and monotone**: SPY spuriously
35 + "leads" tickers by +1 min in proportion to their staleness
36 + (Spearman = +0.43). It peaks in the *mid* tercile (+0.047): the stalest
37 + names trade so rarely that even LOCF correlation collapses — the artifact
38 + is worst where it is least obvious.
39 +* **SPX-vs-SPY**: contemporaneous corr 0.965, and a +0.065 cross-correlation
40 + with SPY leading by 1 minute. An "ETF price discovery leads the index"
41 + finding is manufactured by index print staleness — measured here so Q2
42 + hypotheses must beat it.
43 +
44 +## Limitations
45 +
46 +Level 0 by construction (descriptive nulls; single quarter; one venue's
47 +bar convention). 11/42 tickers dropped for insufficient data — the null for
48 +*ultra*-illiquid names is therefore understated. Q1-2024-specific levels;
49 +expC should re-measure per period rather than reuse these constants blindly.
50 +
51 +## Verdict
52 +
53 +**Complete — nulls established and usable.** Both pre-specified
54 +falsification criteria failed to trigger. Numbers are recorded in
55 +`research/artifact_taxonomy.md`; expC (reversion scan) must report every
56 +effect *net of* the bucket-matched bounce null, and expD must run the
57 +synchronized-vs-raw timestamp comparison this experiment quantified.
modified experiments/micro/expB_artifact_baselines/benchmark.py +245 −10
@@ -1,7 +1,7 @@
1 1 # =============================================================================
2 2 # Project : anomaly-atlas
3 3 # File : experiments/micro/expB_artifact_baselines/benchmark.py
4 −# Purpose : Benchmark runner: Artifact baselines: null distributions of bounce, staleness, non…
4 +# Purpose : Measure the artifact nulls: bounce, staleness, LOCF lead-lag
5 5 # Author : Simon-Pierre Boucher
6 6 # Contact : contact@spboucher.ai
7 7 # Data src : hfmarketdata.io (sole data source)
@@ -10,25 +10,260 @@
10 10 # Platform : macOS / Apple Silicon (arm64)
11 11 # License : All rights reserved (research code)
12 12 # =============================================================================
13 +"""Experiment B — artifact baselines on real data (pre-specified protocol in
14 +hypothesis.md; detectors gated on synthetic ground truth first, §8.1).
13 15
14 −"""Benchmark entry point for expB_artifact_baselines.
15 −
16 −Must embed the hardware manifest in all result output
17 −(see benchmarks/hardware_manifest.py) and write results to
18 −results/expB_artifact_baselines/<timestamp>/. Uses hfmarketdata.io data ONLY, exclusively
19 −through src/anomaly_atlas/data/hf_client.py.
16 +Every number produced here is a NULL LEVEL (Level 0 by construction): the
17 +fake-signal magnitude that later experiments must exceed before claiming
18 +anything. Universe, window, seeds are pre-specified; all data flows through
19 +the cached hf_client.
20 20 """
21 21
22 +from __future__ import annotations
23 +
24 +import json
22 25 import sys
26 +from datetime import UTC, datetime
23 27 from pathlib import Path
24 28
25 −sys.path.insert(0, str(Path(__file__).resolve().parents[3] / "benchmarks"))
29 +import numpy as np
30 +
31 +REPO_ROOT = Path(__file__).resolve().parents[3]
32 +sys.path.insert(0, str(REPO_ROOT / "benchmarks"))
33 +sys.path.insert(0, str(REPO_ROOT / "src"))
34 +
26 35 from hardware_manifest import collect_manifest # noqa: E402
27 36
37 +from anomaly_atlas.data.hf_client import HFMarketDataClient # noqa: E402
38 +from anomaly_atlas.stats.bootstrap import moving_block_bootstrap, percentile_ci # noqa: E402
39 +from anomaly_atlas.stats.reversion import ac1, variance_ratio # noqa: E402
40 +from anomaly_atlas.validation.artifacts import roll_spread # noqa: E402
41 +
42 +LIQUID_STOCK = ["AAPL", "MSFT", "NVDA", "AMZN", "GOOGL", "META", "TSLA", "JPM", "XOM", "UNH"]
43 +LIQUID_ETF = ["SPY", "QQQ"]
44 +N_RANDOM, RANDOM_SEED = 30, 42
45 +START, END = "2024-01-02", "2024-04-01"
46 +ADJ = "adj_split"
47 +BOOT_N, BOOT_SEED = 300, 42
48 +MAX_LAG = 3
49 +
50 +RTH_MINUTES = [f"{h:02d}:{m:02d}" for h in range(9, 16) for m in range(60)]
51 +RTH_MINUTES = [t for t in RTH_MINUTES if "09:30" <= t < "16:00"] # 390 slots
52 +SLOT = {t: i for i, t in enumerate(RTH_MINUTES)}
53 +
54 +
55 +def rth_day_grids(bars: list[dict]) -> dict[str, np.ndarray]:
56 + """day -> 390-slot array of log close prices (NaN where no print)."""
57 + days: dict[str, np.ndarray] = {}
58 + for b in bars:
59 + dt = b["datetime"]
60 + t = dt[11:16]
61 + if not ("09:30" <= t < "16:00"):
62 + continue
63 + grid = days.setdefault(dt[:10], np.full(390, np.nan))
64 + grid[SLOT[t]] = np.log(b["close"])
65 + return days
66 +
67 +
68 +def trade_time_returns(days: dict[str, np.ndarray]) -> np.ndarray:
69 + """Within-day log returns between consecutive PRINTS (no grid, no LOCF)."""
70 + out = []
71 + for day in sorted(days):
72 + p = days[day]
73 + obs = p[np.isfinite(p)]
74 + if len(obs) >= 2:
75 + out.append(np.diff(obs))
76 + return np.concatenate(out) if out else np.array([])
77 +
78 +
79 +def locf_grid_returns(days: dict[str, np.ndarray], day_list: list[str]) -> np.ndarray:
80 + """Concatenated per-day LOCF grid returns, NaN before first print and at
81 + day boundaries — the join that MANUFACTURES the stale-price artifact."""
82 + out = []
83 + for day in day_list:
84 + p = days.get(day)
85 + if p is None:
86 + out.append(np.full(389, np.nan))
87 + continue
88 + filled = p.copy()
89 + for i in range(1, 390):
90 + if not np.isfinite(filled[i]):
91 + filled[i] = filled[i - 1]
92 + out.append(np.diff(filled)) # NaN propagates before first print
93 + return np.concatenate(out)
94 +
95 +
96 +def nan_xcorr(x: np.ndarray, y: np.ndarray, max_lag: int) -> dict[int, float]:
97 + """corr(x_{t-k}, y_t) over finite pairs only; k>0 = x leads y."""
98 + n = min(len(x), len(y))
99 + x, y = x[:n], y[:n]
100 + out: dict[int, float] = {}
101 + for k in range(-max_lag, max_lag + 1):
102 + a = x[: n - k] if k >= 0 else x[-k:]
103 + b = y[k:] if k >= 0 else y[: n + k]
104 + m = np.isfinite(a) & np.isfinite(b)
105 + if m.sum() < 100 or a[m].std() == 0 or b[m].std() == 0:
106 + out[k] = float("nan")
107 + continue
108 + out[k] = float(np.corrcoef(a[m], b[m])[0, 1])
109 + return out
110 +
111 +
112 +def analyze_ticker(days: dict[str, np.ndarray], day_list: list[str]) -> dict | None:
113 + present = (
114 + np.concatenate([np.isfinite(days[d]) for d in day_list if d in days])
115 + if any(d in days for d in day_list)
116 + else np.array([])
117 + )
118 + n_days_covered = sum(d in days for d in day_list)
119 + if n_days_covered < 30:
120 + return None
121 + r = trade_time_returns(days)
122 + if len(r) < 2_000:
123 + return None
124 + block = max(50, len(r) // max(n_days_covered, 1))
125 + boot = moving_block_bootstrap(r, ac1, block=block, n_boot=BOOT_N, seed=BOOT_SEED)
126 + lo, hi = percentile_ci(boot)
127 + spread = roll_spread(r)
128 + return {
129 + "days_covered": n_days_covered,
130 + "staleness": round(1.0 - present.mean() * len(present) / (390 * n_days_covered), 4)
131 + if n_days_covered
132 + else None,
133 + "rth_fill_ratio": round(present.sum() / (390 * n_days_covered), 4),
134 + "n_trade_returns": int(len(r)),
135 + "ac1": round(ac1(r), 5),
136 + "ac1_ci95": [round(lo, 5), round(hi, 5)],
137 + "roll_rel_spread": round(spread, 6) if np.isfinite(spread) else None,
138 + "vr5": round(variance_ratio(r, 5), 4),
139 + "vr30": round(variance_ratio(r, 30), 4),
140 + }
141 +
28 142
29 143 def main() -> None:
30 − collect_manifest() # embedded in results once implemented
31 − raise NotImplementedError("experiment not yet implemented")
144 + run_utc = datetime.now(UTC)
145 + client = HFMarketDataClient()
146 +
147 + # deterministic random universe (seed pre-specified)
148 + all_stock = client.tickers("stock", timeframe="1min", adjustment=ADJ)
149 + rng = np.random.default_rng(RANDOM_SEED)
150 + random_universe = sorted(rng.choice(sorted(all_stock), N_RANDOM, replace=False))
151 + universe = (
152 + [("stock", t, "liquid") for t in LIQUID_STOCK]
153 + + [("etf", t, "liquid") for t in LIQUID_ETF]
154 + + [("stock", t, "random") for t in random_universe]
155 + )
156 +
157 + # fetch + grid everything
158 + grids: dict[str, dict[str, np.ndarray]] = {}
159 + for asset, ticker, _ in universe:
160 + bars = client.get_bars(asset, ticker, "1min", ADJ, START, END)
161 + grids[ticker] = rth_day_grids(bars)
162 + print(f"{ticker}: {sum(len(v[np.isfinite(v)]) for v in grids[ticker].values())} RTH bars")
163 + day_list = sorted(grids["SPY"].keys()) # trading calendar := SPY days
164 +
165 + # B1-B3: per-ticker artifact levels
166 + per_ticker: dict[str, dict] = {}
167 + for asset, ticker, bucket in universe:
168 + m = analyze_ticker(grids[ticker], day_list)
169 + if m is not None:
170 + m["bucket"] = bucket
171 + m["asset"] = asset
172 + per_ticker[ticker] = m
173 +
174 + # B4: LOCF lead-lag vs SPY
175 + spy_r = locf_grid_returns(grids["SPY"], day_list)
176 + for ticker, m in per_ticker.items():
177 + if ticker == "SPY":
178 + continue
179 + r = locf_grid_returns(grids[ticker], day_list)
180 + xc = nan_xcorr(spy_r, r, MAX_LAG)
181 + m["xcorr_vs_spy"] = {str(k): round(v, 5) if np.isfinite(v) else None for k, v in xc.items()}
182 + m["spy_leads_+1"] = round(xc[1], 5) if np.isfinite(xc[1]) else None
183 +
184 + # SPX (index) vs SPY — the non-synchronous-session case
185 + spx_bars = client.get_bars("index", "SPX", "1min", None, START, END)
186 + spx_grid = rth_day_grids(spx_bars)
187 + spx_r = locf_grid_returns(spx_grid, day_list)
188 + spx_xc = nan_xcorr(spx_r, spy_r, MAX_LAG)
189 +
190 + # staleness -> artifact monotonicity (Spearman)
191 + pairs = [
192 + (m["staleness"], m["spy_leads_+1"])
193 + for m in per_ticker.values()
194 + if m.get("spy_leads_+1") is not None and m["staleness"] is not None
195 + ]
196 + xs = np.array([p[0] for p in pairs])
197 + ys = np.array([p[1] for p in pairs])
198 + rx = np.argsort(np.argsort(xs)).astype(float)
199 + ry = np.argsort(np.argsort(ys)).astype(float)
200 + spearman = float(np.corrcoef(rx, ry)[0, 1]) if len(pairs) > 5 else float("nan")
201 +
202 + # aggregates by staleness tercile
203 + stale_vals = sorted(m["staleness"] for m in per_ticker.values())
204 + t1, t2 = np.percentile(stale_vals, [33.3, 66.7])
205 +
206 + def tercile(s: float) -> str:
207 + return "fresh" if s <= t1 else "mid" if s <= t2 else "stale"
208 +
209 + agg: dict[str, dict] = {}
210 + for name in ("fresh", "mid", "stale"):
211 + rows = [m for m in per_ticker.values() if tercile(m["staleness"]) == name]
212 + if not rows:
213 + continue
214 + agg[name] = {
215 + "n": len(rows),
216 + "median_staleness": round(float(np.median([m["staleness"] for m in rows])), 4),
217 + "median_ac1": round(float(np.median([m["ac1"] for m in rows])), 5),
218 + "median_roll_spread": round(
219 + float(np.median([m["roll_rel_spread"] for m in rows if m["roll_rel_spread"]])), 6
220 + ),
221 + "median_vr5": round(float(np.median([m["vr5"] for m in rows])), 4),
222 + "median_vr30": round(float(np.median([m["vr30"] for m in rows])), 4),
223 + "median_spy_leads_+1": round(
224 + float(
225 + np.median(
226 + [m["spy_leads_+1"] for m in rows if m.get("spy_leads_+1") is not None]
227 + )
228 + ),
229 + 5,
230 + ),
231 + }
232 +
233 + results = {
234 + "experiment": "expB_artifact_baselines",
235 + "run_utc": run_utc.isoformat(),
236 + "author": "Simon-Pierre Boucher",
237 + "contact": "contact@spboucher.ai",
238 + "data_source": "hfmarketdata.io",
239 + "protocol": {
240 + "window": [START, END],
241 + "adjustment": ADJ,
242 + "rth": "09:30-16:00",
243 + "liquid": LIQUID_STOCK + LIQUID_ETF,
244 + "random_universe": list(random_universe),
245 + "random_seed": RANDOM_SEED,
246 + "boot": [BOOT_N, BOOT_SEED],
247 + "confidence_level": 0,
248 + "note": "artifact NULL levels — descriptive, in-sample by design",
249 + },
250 + "per_ticker": per_ticker,
251 + "terciles": {"cuts": [round(float(t1), 4), round(float(t2), 4)], "agg": agg},
252 + "staleness_vs_spy_lead_spearman": round(spearman, 4),
253 + "spx_vs_spy_xcorr": {
254 + str(k): round(v, 5) if np.isfinite(v) else None for k, v in spx_xc.items()
255 + },
256 + "client_stats": vars(client.stats) | {"refreshes": list(client.stats.refreshes)},
257 + "manifest": collect_manifest(),
258 + }
259 +
260 + out_dir = REPO_ROOT / "results" / "expB_artifact_baselines" / run_utc.strftime("%Y%m%dT%H%M%SZ")
261 + out_dir.mkdir(parents=True)
262 + (out_dir / "results.json").write_text(json.dumps(results, indent=2) + "\n")
263 + print(f"\nwrote {out_dir.relative_to(REPO_ROOT)}/results.json")
264 + print("terciles:", json.dumps(agg, indent=1))
265 + print("spearman(staleness, SPY leads +1):", round(spearman, 4))
266 + print("SPX vs SPY xcorr:", results["spx_vs_spy_xcorr"])
32 267
33 268
34 269 if __name__ == "__main__":
modified experiments/micro/expB_artifact_baselines/hypothesis.md +52 −12
@@ -5,34 +5,74 @@ 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: final
9 10 ---
10 11
11 12 # Hypothesis — expB_artifact_baselines
12 13
14 +*Pre-specified 2026-08-12, before any real-data measurement. Detectors passed
15 +the §8.1 synthetic gate first (11 tests, benchmarks/synthetic/).*
16 +
13 17 ```text
14 18 Hypothesis
15 − <what we believe and why — pre-specified BEFORE looking at results>
19 + The three mechanical artifacts identified in expA are MEASURABLE and
20 + MATERIAL in this dataset at 1-minute resolution:
21 + (i) bid-ask bounce produces negative AC1 in 1min returns, larger for
22 + less liquid names (Roll-implied relative spread as the null level);
23 + (ii) illiquid names are stale on a large fraction of RTH minutes;
24 + (iii) LOCF-gridding makes SPY spuriously "lead" stale tickers at +1 min,
25 + with the artifact magnitude increasing in staleness.
16 26
17 27 Falsification criterion
18 − <the concrete measurable outcome that would prove this wrong>
28 + The experiment fails if the nulls are unusable as baselines: bounce AC1
29 + indistinguishable from 0 across the liquidity spectrum (|median AC1| CI
30 + covering 0 for the bottom-liquidity tercile), or no monotone relation
31 + between staleness and the +1min SPY cross-correlation (Spearman rho <= 0
32 + across tickers).
19 33
20 34 Artifact null(s)
21 − <the fake-signal baseline(s) this must beat: bounce / staleness /
22 − non-synchronous timestamps / permuted calendar / random walk>
35 + This experiment BUILDS the artifact nulls; its own null is the synthetic
36 + ground truth (§8.1 gate) — detectors verified to read 0 on random walks.
23 37
24 38 Method
25 − <exact procedure, universe, split (train/validation/holdout), seeds,
26 − number of hypotheses tested, correction applied>
39 + Universe (pre-specified): LIQUID = {AAPL MSFT NVDA AMZN GOOGL META TSLA
40 + JPM XOM UNH} + {SPY QQQ}; RANDOM = 30 tickers drawn from the full 1min
41 + stock ticker list with numpy seed 42 (deterministic given the cached
42 + list). Window: 2024-01-02 → 2024-04-01, RTH only (09:30 ≤ t < 16:00),
43 + adjustment adj_split, all data via hf_client (cached).
44 + Per ticker: staleness ratio on the 390-min RTH grid; trade-to-trade 1min
45 + log-return AC1 with moving-block bootstrap CI (block = 1 day, n=300,
46 + seed=42); Roll implied relative spread; VR(5), VR(30).
47 + Cross: lagged xcorr (±3 min, NaN-aware, day-boundary safe) of LOCF-grid
48 + returns vs SPY; SPX-vs-SPY as the index-staleness case.
49 + This is a DESCRIPTIVE measurement of artifact levels, not an anomaly
50 + claim: no OOS split; every number is Level 0 by construction.
27 51
28 52 Result
29 − <filled after the run: effect size, bootstrap CIs, corrected p-values,
30 − OOS status, cost-adjusted effect, credits used>
53 + Run 20260812T055602Z (47 requests, 795 185 rows, 0 retries). 31/42
54 + tickers had enough data (>=30 days, >=2000 returns); 11 dropped, listed
55 + in results.json. Median by staleness tercile (staleness | AC1 | Roll rel
56 + spread | VR30 | SPY-leads-+1min):
57 + fresh 0.000 | -0.009 | 1.4 bp | 0.972 | +0.005
58 + mid 0.107 | -0.051 | 3.8 bp | 0.901 | +0.047
59 + stale 0.691 | -0.232 | 17.0 bp | 0.547 | +0.021
60 + Spearman(staleness, SPY-leads-+1) = +0.43 (> 0: monotone). Extreme case
61 + RITM.B: staleness 0.90, AC1 -0.253, VR30 0.35. Mega-caps: AC1 CI covers
62 + 0 and the Roll estimator is undefined (positive autocov) — no measurable
63 + bounce at the top. SPX-vs-SPY: corr 0.965 at lag 0 and +0.065 with SPY
64 + leading by 1 min (index prints lag the tradable ETF).
31 65
32 66 Interpretation
33 − <what the numbers mean, WITH confidence level (0-3); alternative
34 − explanations considered — artifact first>
67 + Both falsification criteria FAILED to trigger: the nulls are usable.
68 + Headline: an uncorrected VR/AC1 scan on mid/low-liquidity names is
69 + DOMINATED by artifacts — VR30 of 0.55 and AC1 of -0.23 arise with no
70 + economic mean reversion whatsoever. All numbers Level 0 (descriptive
71 + null levels), as pre-specified. Magnitudes recorded in
72 + research/artifact_taxonomy.md.
35 73
36 74 Next experiment
37 − <the most informative follow-up given this result>
75 + expC (reversion scan) consumes these nulls: any reversion claim must
76 + exceed the bounce null for its liquidity bucket. Phase 1 literature
77 + sweep proceeds in parallel.
38 78 ```
modified research/LOG.md +28 −0
@@ -54,3 +54,31 @@ the expB nulls.
54 54
55 55 **Decision.** Full findings in research/data_source_profile.md (status:
56 56 reviewed). Next: expB_artifact_baselines + Phase 1 literature sweep.
57 +
58 +## 2026-08-12 02:15 ET — expB complete: artifact nulls measured (after the §8.1 gate caught a real bug)
59 +
60 +**Question.** Are the bounce/staleness/non-synchronicity artifacts measurable
61 +and material at 1min on this data?
62 +
63 +**Experiment.** Built generators with known ground truth + first real stats
64 +modules (reversion, leadlag, bootstrap, artifacts). The mandatory synthetic
65 +gate (11 tests) CAUGHT A REAL BUG before any real data was touched: the
66 +variance-ratio estimator divided by q twice (VR ≈ 1/q on a pure random walk).
67 +Fixed; all detectors then read 0 on random walks and recover planted effects.
68 +expB then ran the pre-specified protocol (hypothesis.md written first):
69 +12 liquid + 30 seed-42 random tickers, Q1 2024, RTH 1min, 47 requests,
70 +795k rows.
71 +
72 +**Result.** Median by staleness tercile (staleness | AC1 | Roll spread |
73 +VR30 | SPY-leads-+1min): fresh 0.00 | −0.009 | 1.4bp | 0.97 | +0.005;
74 +mid 0.11 | −0.051 | 3.8bp | 0.90 | +0.047; stale 0.69 | −0.232 | 17bp |
75 +0.55 | +0.021. Spearman(staleness, SPY-lead) = +0.43. SPX-vs-SPY: SPY leads
76 +by 1min at +0.065. Mega-caps: no measurable bounce (AC1 CI covers 0).
77 +
78 +**Interpretation.** Uncorrected VR/AC1 scans on mid/low-liquidity names are
79 +dominated by plumbing, not economics. Nulls usable; both falsification
80 +criteria failed to trigger. All Level 0 by construction.
81 +
82 +**Decision.** artifact_taxonomy.md now carries 7 entries with measured
83 +magnitudes (T1–T7). Next: expC consumes the bucket-matched bounce null;
84 +Phase 1 literature sweep in parallel.
modified research/artifact_taxonomy.md +93 −5
@@ -5,13 +5,101 @@ 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 # Artifact taxonomy (living document)
12 13
13 −*Primary deliverable (Q4). Not yet written.*
14 +The catalogue of mechanisms **in this specific dataset** that manufacture
15 +fake anomalies — a primary deliverable (Q4). Every entry: mechanism,
16 +detection, measured magnitude on this data, neutralization. Magnitudes from
17 +`results/expA_data_reality/20260812T054515Z/` and
18 +`results/expB_artifact_baselines/20260812T055602Z/` (Q1 2024, RTH 1min,
19 +pre-specified universe). Detectors validated on synthetic ground truth first
20 +(`benchmarks/synthetic/test_synthetic_gate.py`, 11 tests).
14 21
15 −Catalogue of artifacts in THIS dataset that masquerade as anomalies —
16 −bid-ask bounce, stale prices, non-synchronous timestamps, survivorship,
17 −look-ahead, corporate-action gaps — with detection and neutralization for each.
22 +## T1 — Bid-ask bounce (Roll)
23 +
24 +* **Mechanism.** Trades alternate bid/ask; observed returns gain a negative
25 + lag-1 autocovariance (−s²/4) with no economics. Masquerades as
26 + mean-reversion (Q1).
27 +* **Detection.** Roll implied relative spread `2·√(−autocov1)`;
28 + bounce-implied AC1; `excess_reversion()` (validation/artifacts.py).
29 +* **Measured.** Median AC1 by staleness tercile: −0.009 (fresh) / −0.051
30 + (mid) / **−0.232 (stale)**; Roll spread 1.4 / 3.8 / 17.0 bp. Mega-caps:
31 + no measurable bounce (AC1 CI covers 0; Roll undefined ~half the time —
32 + positive autocov).
33 +* **Neutralize.** Report reversion net of the liquidity-bucket bounce null;
34 + never average AC1 across liquidity buckets; treat Roll-undefined as
35 + "no bounce measurable", not zero spread.
36 +
37 +## T2 — Stale prices / missing minutes
38 +
39 +* **Mechanism.** Bars exist only where trades occurred (expA: zero
40 + zero-volume bars; an illiquid name printed 38 bars/day). LOCF joins add a
41 + large mass of zero returns and depress variance-ratio statistics.
42 +* **Detection.** `staleness_ratio` on the 390-min RTH grid.
43 +* **Measured.** Staleness up to 0.90 (RITM.B); **VR(30) = 0.55 median for
44 + the stale tercile — 0.35 extreme — with zero planted economics.** 11/42
45 + pre-specified tickers had too little data to analyze at all.
46 +* **Neutralize.** Explicit grids with observed-masks (never silent LOCF);
47 + liquidity filters pre-specified; VR/AC1 claims benchmarked against the
48 + staleness-matched null, not against 1.0.
49 +
50 +## T3 — Non-synchronous lead-lag (LOCF cross-correlation)
51 +
52 +* **Mechanism.** A fresh series LOCF-joined to a stale one appears to LEAD
53 + it: the stale print reflects old common information (classic
54 + non-synchronous trading bias).
55 +* **Detection.** Lagged cross-correlation vs SPY on the LOCF grid; synthetic
56 + ground truth: rho=0.7 pair with 30 % observation → spurious +1 lag corr.
57 +* **Measured.** SPY "leads" mid-staleness tickers by +0.047 at +1 min
58 + (Spearman vs staleness +0.43). **SPX-vs-SPY: +0.065 with SPY leading
59 + 1 min** at 0.965 contemporaneous corr — index prints lag the ETF.
60 +* **Neutralize.** Any Q2 lead-lag claim must exceed the staleness-predicted
61 + cross-correlation; test on synchronized (both-fresh) subsamples; index
62 + series are stale by construction.
63 +
64 +## T4 — Auction close vs last bar
65 +
66 +* **Mechanism.** Daily bars carry the official closing-auction print; 1min
67 + bars do not (expA: AAPL 312.41 daily close vs 312.49 last RTH 1min close).
68 + Mixing conventions manufactures phantom overnight/close-to-close returns.
69 +* **Measured.** 8 bp discrepancy on a calm day for the most liquid stock.
70 +* **Neutralize.** Pick ONE close convention per experiment and state it;
71 + never compute overnight returns across mixed conventions.
72 +
73 +## T5 — Rolling adjustment anchor
74 +
75 +* **Mechanism.** `adj_splitdiv` re-bases the whole history to the dataset
76 + build date (expA: AAPL 2020-08-31 close = 125.17 adjusted vs 129.04
77 + traded). Adjusted series are not point-in-time stable → silent look-ahead
78 + and irreproducibility if the cache is refreshed mid-study.
79 +* **Neutralize.** Frozen local cache (hf_client never silently refetches);
80 + data-manifest hash in every provenance; intraday work uses within-day
81 + returns (adjustment-invariant) or UNADJUSTED plus explicit factors.
82 +
83 +## T6 — Daily vs intraday volume conventions
84 +
85 +* **Mechanism.** Daily volume includes auction/consolidated prints absent
86 + from 1min bars (expA: 46.1 M daily vs 34.7 M extended-1min sum vs 25.7 M
87 + RTH-1min sum for AAPL on one day — a 1.8× spread across conventions).
88 +* **Neutralize.** Volume-based signals pick one convention; never mix daily
89 + and intraday volume in one feature.
90 +
91 +## T7 — Vendor session / timezone semantics
92 +
93 +* **Mechanism.** All timestamps are US/Eastern wall-clock without a marker;
94 + sessions differ per class (equities 04:00–19:59, SPX prints to 16:20,
95 + futures ≈24 h, fx ET-week, crypto 24/7). Cross-asset joins on naive
96 + timestamps silently compare different market states.
97 +* **Neutralize.** One canonical calendar module (`data/calendars.py`),
98 + explicit session filters per asset class, DST-aware conversions.
99 +
100 +---
101 +
102 +*Open items: intraday-seasonality of spread/staleness (U-shape) as a
103 +confounder for Q3 calendar scans (to be measured in expE); continuous-futures
104 +splice choice (3 variants exposed by the API) as a testable artifact for
105 +futures-based hypotheses.*
added results/expB_artifact_baselines/20260812T055602Z/results.json +963 −0
@@ -0,0 +1,963 @@
1 +{
2 + "experiment": "expB_artifact_baselines",
3 + "run_utc": "2026-08-12T05:56:02.079317+00:00",
4 + "author": "Simon-Pierre Boucher",
5 + "contact": "contact@spboucher.ai",
6 + "data_source": "hfmarketdata.io",
7 + "protocol": {
8 + "window": [
9 + "2024-01-02",
10 + "2024-04-01"
11 + ],
12 + "adjustment": "adj_split",
13 + "rth": "09:30-16:00",
14 + "liquid": [
15 + "AAPL",
16 + "MSFT",
17 + "NVDA",
18 + "AMZN",
19 + "GOOGL",
20 + "META",
21 + "TSLA",
22 + "JPM",
23 + "XOM",
24 + "UNH",
25 + "SPY",
26 + "QQQ"
27 + ],
28 + "random_universe": [
29 + "ATRO",
30 + "AUVI",
31 + "AXDX",
32 + "BKE",
33 + "CECO",
34 + "CKX",
35 + "GCL",
36 + "GRO",
37 + "HPE",
38 + "HTD",
39 + "HWM",
40 + "ICUI",
41 + "KEY.K",
42 + "KSPI",
43 + "LENZ",
44 + "LTSL",
45 + "NSTS",
46 + "NWAX",
47 + "PBYI",
48 + "PLNT",
49 + "PSA.G",
50 + "RDZN",
51 + "RITM.B",
52 + "RPM",
53 + "RUM",
54 + "SLF",
55 + "SPB",
56 + "STRRP",
57 + "USGOW",
58 + "WTFCM"
59 + ],
60 + "random_seed": 42,
61 + "boot": [
62 + 300,
63 + 42
64 + ],
65 + "confidence_level": 0,
66 + "note": "artifact NULL levels \u2014 descriptive, in-sample by design"
67 + },
68 + "per_ticker": {
69 + "AAPL": {
70 + "days_covered": 61,
71 + "staleness": 0.0,
72 + "rth_fill_ratio": 1.0,
73 + "n_trade_returns": 23729,
74 + "ac1": 0.00162,
75 + "ac1_ci95": [
76 + -0.01747,
77 + 0.02147
78 + ],
79 + "roll_rel_spread": null,
80 + "vr5": 1.011,
81 + "vr30": 1.0477,
82 + "bucket": "liquid",
83 + "asset": "stock",
84 + "xcorr_vs_spy": {
85 + "-3": 0.00528,
86 + "-2": -0.0079,
87 + "-1": 0.00678,
88 + "0": 0.53789,
89 + "1": 0.01237,
90 + "2": -0.00536,
91 + "3": -0.00986
92 + },
93 + "spy_leads_+1": 0.01237
94 + },
95 + "MSFT": {
96 + "days_covered": 61,
97 + "staleness": 0.0,
98 + "rth_fill_ratio": 1.0,
99 + "n_trade_returns": 23729,
100 + "ac1": -0.01245,
101 + "ac1_ci95": [
102 + -0.04532,
103 + 0.01955
104 + ],
105 + "roll_rel_spread": 0.000111,
106 + "vr5": 0.9523,
107 + "vr30": 0.9179,
108 + "bucket": "liquid",
109 + "asset": "stock",
110 + "xcorr_vs_spy": {
111 + "-3": -0.0054,
112 + "-2": 0.00084,
113 + "-1": 0.00872,
114 + "0": 0.59532,
115 + "1": 0.00645,
116 + "2": -0.00534,
117 + "3": -0.00479
118 + },
119 + "spy_leads_+1": 0.00645
120 + },
121 + "NVDA": {
122 + "days_covered": 61,
123 + "staleness": 0.0,
124 + "rth_fill_ratio": 1.0,
125 + "n_trade_returns": 23729,
126 + "ac1": 0.01298,
127 + "ac1_ci95": [
128 + -0.01676,
129 + 0.04068
130 + ],
131 + "roll_rel_spread": null,
132 + "vr5": 0.9979,
133 + "vr30": 0.9733,
134 + "bucket": "liquid",
135 + "asset": "stock",
136 + "xcorr_vs_spy": {
137 + "-3": 0.00212,
138 + "-2": 0.01372,
139 + "-1": 0.02509,
140 + "0": 0.53888,
141 + "1": 0.00859,
142 + "2": -0.01068,
143 + "3": -0.01252
144 + },
145 + "spy_leads_+1": 0.00859
146 + },
147 + "AMZN": {
148 + "days_covered": 61,
149 + "staleness": 0.0,
150 + "rth_fill_ratio": 1.0,
151 + "n_trade_returns": 23729,
152 + "ac1": -0.00916,
153 + "ac1_ci95": [
154 + -0.03589,
155 + 0.01421
156 + ],
157 + "roll_rel_spread": 0.000117,
158 + "vr5": 0.946,
159 + "vr30": 0.8658,
160 + "bucket": "liquid",
161 + "asset": "stock",
162 + "xcorr_vs_spy": {
163 + "-3": -0.01025,
164 + "-2": -0.00672,
165 + "-1": 0.00274,
166 + "0": 0.5434,
167 + "1": 0.0037,
168 + "2": -0.00722,
169 + "3": 0.01063
170 + },
171 + "spy_leads_+1": 0.0037
172 + },
173 + "GOOGL": {
174 + "days_covered": 61,
175 + "staleness": 0.0,
176 + "rth_fill_ratio": 1.0,
177 + "n_trade_returns": 23729,
178 + "ac1": -0.02624,
179 + "ac1_ci95": [
180 + -0.06462,
181 + 0.01378
182 + ],
183 + "roll_rel_spread": 0.000196,
184 + "vr5": 0.9769,
185 + "vr30": 0.9116,
186 + "bucket": "liquid",
187 + "asset": "stock",
188 + "xcorr_vs_spy": {
189 + "-3": 0.00045,
190 + "-2": -0.00117,
191 + "-1": -0.00146,
192 + "0": 0.4932,
193 + "1": -0.00674,
194 + "2": -0.00411,
195 + "3": 0.00345
196 + },
197 + "spy_leads_+1": -0.00674
198 + },
199 + "META": {
200 + "days_covered": 61,
201 + "staleness": 0.0,
202 + "rth_fill_ratio": 1.0,
203 + "n_trade_returns": 23729,
204 + "ac1": -0.02418,
205 + "ac1_ci95": [
206 + -0.05678,
207 + 0.0063
208 + ],
209 + "roll_rel_spread": 0.000244,
210 + "vr5": 0.9732,
211 + "vr30": 0.9473,
212 + "bucket": "liquid",
213 + "asset": "stock",
214 + "xcorr_vs_spy": {
215 + "-3": 0.00716,
216 + "-2": 0.00212,
217 + "-1": 0.01956,
218 + "0": 0.48589,
219 + "1": 0.00367,
220 + "2": -0.00819,
221 + "3": 0.00193
222 + },
223 + "spy_leads_+1": 0.00367
224 + },
225 + "TSLA": {
226 + "days_covered": 61,
227 + "staleness": 0.0,
228 + "rth_fill_ratio": 1.0,
229 + "n_trade_returns": 23729,
230 + "ac1": 0.01645,
231 + "ac1_ci95": [
232 + -0.00215,
233 + 0.03578
234 + ],
235 + "roll_rel_spread": null,
236 + "vr5": 1.0053,
237 + "vr30": 1.0155,
238 + "bucket": "liquid",
239 + "asset": "stock",
240 + "xcorr_vs_spy": {
241 + "-3": -0.00901,
242 + "-2": -0.00031,
243 + "-1": 0.01181,
244 + "0": 0.40916,
245 + "1": 0.00585,
246 + "2": -0.00239,
247 + "3": 0.00135
248 + },
249 + "spy_leads_+1": 0.00585
250 + },
251 + "JPM": {
252 + "days_covered": 61,
253 + "staleness": 0.0,
254 + "rth_fill_ratio": 1.0,
255 + "n_trade_returns": 23729,
256 + "ac1": -0.03379,
257 + "ac1_ci95": [
258 + -0.05157,
259 + -0.01069
260 + ],
261 + "roll_rel_spread": 0.000162,
262 + "vr5": 0.9574,
263 + "vr30": 0.9499,
264 + "bucket": "liquid",
265 + "asset": "stock",
266 + "xcorr_vs_spy": {
267 + "-3": 0.00115,
268 + "-2": -0.00947,
269 + "-1": 0.00796,
270 + "0": 0.4252,
271 + "1": 0.00377,
272 + "2": 0.01337,
273 + "3": -0.00602
274 + },
275 + "spy_leads_+1": 0.00377
276 + },
277 + "XOM": {
278 + "days_covered": 61,
279 + "staleness": 0.0,
280 + "rth_fill_ratio": 1.0,
281 + "n_trade_returns": 23729,
282 + "ac1": -0.01076,
283 + "ac1_ci95": [
284 + -0.03129,
285 + 0.00755
286 + ],
287 + "roll_rel_spread": 0.000105,
288 + "vr5": 0.9895,
289 + "vr30": 1.0847,
290 + "bucket": "liquid",
291 + "asset": "stock",
292 + "xcorr_vs_spy": {
293 + "-3": 0.00277,
294 + "-2": -0.01127,
295 + "-1": 0.00023,
296 + "0": 0.20757,
297 + "1": -0.00728,
298 + "2": 0.00613,
299 + "3": 0.00189
300 + },
301 + "spy_leads_+1": -0.00728
302 + },
303 + "UNH": {
304 + "days_covered": 61,
305 + "staleness": 0.0024,
306 + "rth_fill_ratio": 0.9976,
307 + "n_trade_returns": 23673,
308 + "ac1": -0.05529,
309 + "ac1_ci95": [
310 + -0.08843,
311 + -0.02247
312 + ],
313 + "roll_rel_spread": 0.000279,
314 + "vr5": 0.8895,
315 + "vr30": 0.8411,
316 + "bucket": "liquid",
317 + "asset": "stock",
318 + "xcorr_vs_spy": {
319 + "-3": -0.00305,
320 + "-2": -0.01299,
321 + "-1": 0.00162,
322 + "0": 0.12185,
323 + "1": 0.02334,
324 + "2": -4e-05,
325 + "3": 0.00096
326 + },
327 + "spy_leads_+1": 0.02334
328 + },
329 + "SPY": {
330 + "days_covered": 61,
331 + "staleness": 0.0,
332 + "rth_fill_ratio": 1.0,
333 + "n_trade_returns": 23729,
334 + "ac1": 0.00628,
335 + "ac1_ci95": [
336 + -0.01811,
337 + 0.02379
338 + ],
339 + "roll_rel_spread": null,
340 + "vr5": 1.0101,
341 + "vr30": 0.9718,
342 + "bucket": "liquid",
343 + "asset": "etf"
344 + },
345 + "QQQ": {
346 + "days_covered": 61,
347 + "staleness": 0.0,
348 + "rth_fill_ratio": 1.0,
349 + "n_trade_returns": 23729,
350 + "ac1": 0.01507,
351 + "ac1_ci95": [
352 + -0.00676,
353 + 0.0325
354 + ],
355 + "roll_rel_spread": null,
356 + "vr5": 1.0233,
357 + "vr30": 1.0059,
358 + "bucket": "liquid",
359 + "asset": "etf",
360 + "xcorr_vs_spy": {
361 + "-3": 0.00069,
362 + "-2": 0.00181,
363 + "-1": 0.01376,
364 + "0": 0.90007,
365 + "1": 0.01068,
366 + "2": -0.00847,
367 + "3": -0.00557
368 + },
369 + "spy_leads_+1": 0.01068
370 + },
371 + "ATRO": {
372 + "days_covered": 61,
373 + "staleness": 0.6919,
374 + "rth_fill_ratio": 0.3081,
375 + "n_trade_returns": 7269,
376 + "ac1": -0.05122,
377 + "ac1_ci95": [
378 + -0.08802,
379 + -0.01873
380 + ],
381 + "roll_rel_spread": 0.001022,
382 + "vr5": 0.8975,
383 + "vr30": 0.9242,
384 + "bucket": "random",
385 + "asset": "stock",
386 + "xcorr_vs_spy": {
387 + "-3": 0.00995,
388 + "-2": -0.00133,
389 + "-1": -0.00238,
390 + "0": 0.09732,
391 + "1": 0.06282,
392 + "2": 0.03343,
393 + "3": 0.0201
394 + },
395 + "spy_leads_+1": 0.06282
396 + },
397 + "AUVI": {
398 + "days_covered": 61,
399 + "staleness": 0.6646,
400 + "rth_fill_ratio": 0.3354,
401 + "n_trade_returns": 7917,
402 + "ac1": -0.21019,
403 + "ac1_ci95": [
404 + -0.28933,
405 + -0.11986
406 + ],
407 + "roll_rel_spread": 0.008811,
408 + "vr5": 0.6587,
409 + "vr30": 0.6087,
410 + "bucket": "random",
411 + "asset": "stock",
412 + "xcorr_vs_spy": {
413 + "-3": 0.0033,
414 + "-2": 0.00165,
415 + "-1": -0.0025,
416 + "0": 0.01784,
417 + "1": 0.0057,
418 + "2": 0.01093,
419 + "3": -0.00067
420 + },
421 + "spy_leads_+1": 0.0057
422 + },
423 + "AXDX": {
424 + "days_covered": 61,
425 + "staleness": 0.8335,
426 + "rth_fill_ratio": 0.1665,
427 + "n_trade_returns": 3901,
428 + "ac1": -0.32299,
429 + "ac1_ci95": [
430 + -0.35528,
431 + -0.27751
432 + ],
433 + "roll_rel_spread": 0.019373,
434 + "vr5": 0.4941,
435 + "vr30": 0.3858,
436 + "bucket": "random",
437 + "asset": "stock",
438 + "xcorr_vs_spy": {
439 + "-3": 0.00371,
440 + "-2": 0.01291,
441 + "-1": -0.0068,
442 + "0": 0.00065,
443 + "1": -0.00701,
444 + "2": 0.00379,
445 + "3": 0.01124
446 + },
447 + "spy_leads_+1": -0.00701
448 + },
449 + "BKE": {
450 + "days_covered": 61,
451 + "staleness": 0.288,
452 + "rth_fill_ratio": 0.712,
453 + "n_trade_returns": 16878,
454 + "ac1": -0.04574,
455 + "ac1_ci95": [
456 + -0.07188,
457 + -0.019
458 + ],
459 + "roll_rel_spread": 0.000494,
460 + "vr5": 0.9229,
461 + "vr30": 0.8628,
462 + "bucket": "random",
463 + "asset": "stock",
464 + "xcorr_vs_spy": {
465 + "-3": -0.00949,
466 + "-2": -0.00335,
467 + "-1": -0.01292,
468 + "0": 0.21079,
469 + "1": 0.08981,
470 + "2": 0.03184,
471 + "3": -0.00685
472 + },
473 + "spy_leads_+1": 0.08981
474 + },
475 + "CECO": {
476 + "days_covered": 61,
477 + "staleness": 0.5311,
478 + "rth_fill_ratio": 0.4689,
479 + "n_trade_returns": 11093,
480 + "ac1": -0.16489,
481 + "ac1_ci95": [
482 + -0.23411,
483 + -0.07077
484 + ],
485 + "roll_rel_spread": 0.001807,
486 + "vr5": 0.7867,
487 + "vr30": 1.0073,
488 + "bucket": "random",
489 + "asset": "stock",
490 + "xcorr_vs_spy": {
491 + "-3": -0.01082,
492 + "-2": 0.00211,
493 + "-1": 0.00371,
494 + "0": 0.09421,
495 + "1": 0.06433,
496 + "2": 0.03838,
497 + "3": 0.01926
498 + },
499 + "spy_leads_+1": 0.06433
500 + },
501 + "HPE": {
502 + "days_covered": 61,
503 + "staleness": 0.0003,
504 + "rth_fill_ratio": 0.9997,
505 + "n_trade_returns": 23723,
506 + "ac1": -0.06183,
507 + "ac1_ci95": [
508 + -0.08556,
509 + -0.03994
510 + ],
511 + "roll_rel_spread": 0.000504,
512 + "vr5": 0.8673,
513 + "vr30": 1.0189,
514 + "bucket": "random",
515 + "asset": "stock",
516 + "xcorr_vs_spy": {
517 + "-3": 0.00082,
518 + "-2": 0.00395,
519 + "-1": -0.00307,
520 + "0": 0.27357,
521 + "1": 0.00666,
522 + "2": -0.02272,
523 + "3": -0.00514
524 + },
525 + "spy_leads_+1": 0.00666
526 + },
527 + "HTD": {
528 + "days_covered": 61,
529 + "staleness": 0.6906,
530 + "rth_fill_ratio": 0.3094,
531 + "n_trade_returns": 7300,
532 + "ac1": -0.31279,
533 + "ac1_ci95": [
534 + -0.35006,
535 + -0.27744
536 + ],
537 + "roll_rel_spread": 0.00128,
538 + "vr5": 0.5425,
539 + "vr30": 0.4852,
540 + "bucket": "random",
541 + "asset": "stock",
542 + "xcorr_vs_spy": {
543 + "-3": 0.00289,
544 + "-2": 0.00058,
545 + "-1": -0.00173,
546 + "0": 0.03006,
547 + "1": 0.05151,
548 + "2": 0.04017,
549 + "3": 0.01686
550 + },
551 + "spy_leads_+1": 0.05151
552 + },
553 + "HWM": {
554 + "days_covered": 61,
555 + "staleness": 0.0052,
556 + "rth_fill_ratio": 0.9948,
557 + "n_trade_returns": 23605,
558 + "ac1": -0.02708,
559 + "ac1_ci95": [
560 + -0.05571,
561 + 0.00259
562 + ],
563 + "roll_rel_spread": 0.00019,
564 + "vr5": 0.9716,
565 + "vr30": 0.9134,
566 + "bucket": "random",
567 + "asset": "stock",
568 + "xcorr_vs_spy": {
569 + "-3": 0.00501,
570 + "-2": -0.01232,
571 + "-1": 0.00072,
572 + "0": 0.33034,
573 + "1": 0.02045,
574 + "2": 0.00126,
575 + "3": -0.00136
576 + },
577 + "spy_leads_+1": 0.02045
578 + },
579 + "ICUI": {
580 + "days_covered": 61,
581 + "staleness": 0.4941,
582 + "rth_fill_ratio": 0.5059,
583 + "n_trade_returns": 11975,
584 + "ac1": -0.10807,
585 + "ac1_ci95": [
586 + -0.13625,
587 + -0.07268
588 + ],
589 + "roll_rel_spread": 0.001372,
590 + "vr5": 0.8884,
591 + "vr30": 0.8928,
592 + "bucket": "random",
593 + "asset": "stock",
594 + "xcorr_vs_spy": {
595 + "-3": 0.00216,
596 + "-2": 0.01038,
597 + "-1": -0.00739,
598 + "0": 0.12354,
599 + "1": 0.0769,
600 + "2": 0.03793,
601 + "3": -0.00457
602 + },
603 + "spy_leads_+1": 0.0769
604 + },
605 + "KEY.K": {
606 + "days_covered": 61,
607 + "staleness": 0.8674,
608 + "rth_fill_ratio": 0.1326,
609 + "n_trade_returns": 3093,
610 + "ac1": -0.3469,
611 + "ac1_ci95": [
612 + -0.38592,
613 + -0.28806
614 + ],
615 + "roll_rel_spread": 0.005038,
616 + "vr5": 0.4251,
617 + "vr30": 0.2543,
618 + "bucket": "random",
619 + "asset": "stock",
620 + "xcorr_vs_spy": {
621 + "-3": 0.01268,
622 + "-2": 0.01152,
623 + "-1": 0.00011,
624 + "0": 0.00819,
625 + "1": 0.02132,
626 + "2": -0.00214,
627 + "3": 0.00553
628 + },
629 + "spy_leads_+1": 0.02132
630 + },
631 + "KSPI": {
632 + "days_covered": 48,
633 + "staleness": 0.471,
634 + "rth_fill_ratio": 0.529,
635 + "n_trade_returns": 9855,
636 + "ac1": -0.20949,
637 + "ac1_ci95": [
638 + -0.25785,
639 + -0.16733
640 + ],
641 + "roll_rel_spread": 0.001597,
642 + "vr5": 0.6778,
643 + "vr30": 0.6525,
644 + "bucket": "random",
645 + "asset": "stock",
646 + "xcorr_vs_spy": {
647 + "-3": -0.00345,
648 + "-2": -0.00331,
649 + "-1": 0.01628,
650 + "0": 0.01032,
651 + "1": 0.02003,
652 + "2": 0.00966,
653 + "3": -0.00787
654 + },
655 + "spy_leads_+1": 0.02003
656 + },
657 + "LENZ": {
658 + "days_covered": 61,
659 + "staleness": 0.7067,
660 + "rth_fill_ratio": 0.2933,
661 + "n_trade_returns": 6916,
662 + "ac1": -0.27902,
663 + "ac1_ci95": [
664 + -0.35931,
665 + -0.19937
666 + ],
667 + "roll_rel_spread": 0.007069,
668 + "vr5": 0.623,
669 + "vr30": 0.464,
670 + "bucket": "random",
671 + "asset": "stock",
672 + "xcorr_vs_spy": {
673 + "-3": 0.01234,
674 + "-2": 0.00659,
675 + "-1": 0.00154,
676 + "0": 0.00207,
677 + "1": 0.0143,
678 + "2": 0.00425,
679 + "3": -0.00163
680 + },
681 + "spy_leads_+1": 0.0143
682 + },
683 + "PBYI": {
684 + "days_covered": 61,
685 + "staleness": 0.3251,
686 + "rth_fill_ratio": 0.6749,
687 + "n_trade_returns": 15996,
688 + "ac1": -0.11171,
689 + "ac1_ci95": [
690 + -0.14442,
691 + -0.07365
692 + ],
693 + "roll_rel_spread": 0.002591,
694 + "vr5": 0.8282,
695 + "vr30": 0.82,
696 + "bucket": "random",
697 + "asset": "stock",
698 + "xcorr_vs_spy": {
699 + "-3": -0.00863,
700 + "-2": -0.00987,
701 + "-1": 0.0077,
702 + "0": 0.04659,
703 + "1": 0.03636,
704 + "2": 0.00891,
705 + "3": 0.00204
706 + },
707 + "spy_leads_+1": 0.03636
708 + },
709 + "PLNT": {
710 + "days_covered": 61,
711 + "staleness": 0.0305,
712 + "rth_fill_ratio": 0.9695,
713 + "n_trade_returns": 23004,
714 + "ac1": -0.01553,
715 + "ac1_ci95": [
716 + -0.03743,
717 + 0.00607
718 + ],
719 + "roll_rel_spread": 0.000235,
720 + "vr5": 0.9942,
721 + "vr30": 1.0098,
722 + "bucket": "random",
723 + "asset": "stock",
724 + "xcorr_vs_spy": {
725 + "-3": -0.00276,
726 + "-2": 0.01403,
727 + "-1": 0.00527,
728 + "0": 0.20839,
729 + "1": 0.03961,
730 + "2": -0.00434,
731 + "3": -0.00844
732 + },
733 + "spy_leads_+1": 0.03961
734 + },
735 + "RITM.B": {
736 + "days_covered": 61,
737 + "staleness": 0.8984,
738 + "rth_fill_ratio": 0.1016,
739 + "n_trade_returns": 2356,
740 + "ac1": -0.25345,
741 + "ac1_ci95": [
742 + -0.30424,
743 + -0.17296
744 + ],
745 + "roll_rel_spread": 0.001261,
746 + "vr5": 0.5839,
747 + "vr30": 0.3502,
748 + "bucket": "random",
749 + "asset": "stock",
750 + "xcorr_vs_spy": {
751 + "-3": 0.00127,
752 + "-2": 0.01134,
753 + "-1": -0.01608,
754 + "0": -0.00305,
755 + "1": 0.01628,
756 + "2": 0.00606,
757 + "3": 0.00448
758 + },
759 + "spy_leads_+1": 0.01628
760 + },
761 + "RPM": {
762 + "days_covered": 61,
763 + "staleness": 0.2196,
764 + "rth_fill_ratio": 0.7804,
765 + "n_trade_returns": 18504,
766 + "ac1": -0.0903,
767 + "ac1_ci95": [
768 + -0.13912,
769 + -0.06355
770 + ],
771 + "roll_rel_spread": 0.000398,
772 + "vr5": 0.8569,
773 + "vr30": 0.8519,
774 + "bucket": "random",
775 + "asset": "stock",
776 + "xcorr_vs_spy": {
777 + "-3": 0.00269,
778 + "-2": 0.01379,
779 + "-1": -0.01292,
780 + "0": 0.25954,
781 + "1": 0.09142,
782 + "2": 0.01975,
783 + "3": 0.02268
784 + },
785 + "spy_leads_+1": 0.09142
786 + },
787 + "RUM": {
788 + "days_covered": 61,
789 + "staleness": 0.0291,
790 + "rth_fill_ratio": 0.9709,
791 + "n_trade_returns": 23037,
792 + "ac1": -0.05627,
793 + "ac1_ci95": [
794 + -0.09747,
795 + -0.01213
796 + ],
797 + "roll_rel_spread": 0.001713,
798 + "vr5": 0.9298,
799 + "vr30": 0.8893,
800 + "bucket": "random",
801 + "asset": "stock",
802 + "xcorr_vs_spy": {
803 + "-3": 0.01104,
804 + "-2": -0.00292,
805 + "-1": 0.00475,
806 + "0": 0.13733,
807 + "1": 0.05423,
808 + "2": 0.0155,
809 + "3": -0.00228
810 + },
811 + "spy_leads_+1": 0.05423
812 + },
813 + "SLF": {
814 + "days_covered": 61,
815 + "staleness": 0.1826,
816 + "rth_fill_ratio": 0.8174,
817 + "n_trade_returns": 19386,
818 + "ac1": -0.01447,
819 + "ac1_ci95": [
820 + -0.03021,
821 + 0.01033
822 + ],
823 + "roll_rel_spread": 0.000111,
824 + "vr5": 0.997,
825 + "vr30": 0.9691,
826 + "bucket": "random",
827 + "asset": "stock",
828 + "xcorr_vs_spy": {
829 + "-3": 0.00193,
830 + "-2": 0.00374,
831 + "-1": -0.00431,
832 + "0": 0.32152,
833 + "1": 0.11852,
834 + "2": 0.01432,
835 + "3": 0.01368
836 + },
837 + "spy_leads_+1": 0.11852
838 + },
839 + "SPB": {
840 + "days_covered": 61,
841 + "staleness": 0.3332,
842 + "rth_fill_ratio": 0.6668,
843 + "n_trade_returns": 15801,
844 + "ac1": -0.04111,
845 + "ac1_ci95": [
846 + -0.08175,
847 + -0.01481
848 + ],
849 + "roll_rel_spread": 0.000364,
850 + "vr5": 1.008,
851 + "vr30": 1.0611,
852 + "bucket": "random",
853 + "asset": "stock",
854 + "xcorr_vs_spy": {
855 + "-3": 0.00032,
856 + "-2": -0.00068,
857 + "-1": 0.00845,
858 + "0": 0.14854,
859 + "1": 0.07764,
860 + "2": 0.02411,
861 + "3": -0.01058
862 + },
863 + "spy_leads_+1": 0.07764
864 + }
865 + },
866 + "terciles": {
867 + "cuts": [
868 + 0.0,
869 + 0.3346
870 + ],
871 + "agg": {
872 + "fresh": {
873 + "n": 11,
874 + "median_staleness": 0.0,
875 + "median_ac1": -0.00916,
876 + "median_roll_spread": 0.00014,
877 + "median_vr5": 0.9895,
878 + "median_vr30": 0.9718,
879 + "median_spy_leads_+1": 0.00481
880 + },
881 + "mid": {
882 + "n": 10,
883 + "median_staleness": 0.1066,
884 + "median_ac1": -0.05052,
885 + "median_roll_spread": 0.000381,
886 + "median_vr5": 0.9264,
887 + "median_vr30": 0.9013,
888 + "median_spy_leads_+1": 0.04692
889 + },
890 + "stale": {
891 + "n": 10,
892 + "median_staleness": 0.6912,
893 + "median_ac1": -0.23182,
894 + "median_roll_spread": 0.001702,
895 + "median_vr5": 0.6408,
896 + "median_vr30": 0.547,
897 + "median_spy_leads_+1": 0.02067
898 + }
899 + }
900 + },
901 + "staleness_vs_spy_lead_spearman": 0.4287,
902 + "spx_vs_spy_xcorr": {
903 + "-3": 0.00179,
904 + "-2": 0.00012,
905 + "-1": 0.06453,
906 + "0": 0.96524,
907 + "1": 0.00747,
908 + "2": -0.00387,
909 + "3": -0.00233
910 + },
911 + "client_stats": {
912 + "network_requests": 47,
913 + "cache_hits": 0,
914 + "rows_fetched": 795185,
915 + "seconds_waiting": 0.426904123001441,
916 + "errors_retried": 0,
917 + "refreshes": []
918 + },
919 + "manifest": {
920 + "author": "Simon-Pierre Boucher",
921 + "contact": "contact@spboucher.ai",
922 + "project": "anomaly-atlas",
923 + "data_source": "hfmarketdata.io",
924 + "collected_utc": "2026-08-12T05:56:30.682310+00:00",
925 + "chip": {
926 + "brand": "Apple M5 Max",
927 + "arch": "arm64",
928 + "cores_total": 18,
929 + "cores_performance": 6,
930 + "cores_efficiency": 12,
931 + "gpu_cores": 40
932 + },
933 + "memory": {
934 + "unified_bytes": 51539607552,
935 + "unified_gb": 48.0,
936 + "pagesize": 16384
937 + },
938 + "ssd": {
939 + "model": "APPLE SSD AP2048Z",
940 + "size": "2 TB",
941 + "smart_status": "Verified"
942 + },
943 + "os": {
944 + "product": "macOS",
945 + "version": "27.0",
946 + "build": "26A5388g",
947 + "kernel": "27.0.0"
948 + },
949 + "software": {
950 + "python": "3.14.4",
951 + "numpy": "2.5.2",
952 + "pandas": "3.0.5",
953 + "polars": "1.43.2",
954 + "duckdb": "1.5.5",
955 + "statsmodels": "0.14.6",
956 + "arch": "8.0.0"
957 + },
958 + "git": {
959 + "commit": "f59e89156a980def4efceec445f9a8d23c88513b",
960 + "dirty_tree": true
961 + }
962 + }
963 +}
modified src/anomaly_atlas/stats/bootstrap.py +45 −3
@@ -1,7 +1,7 @@
1 1 # =============================================================================
2 2 # Project : anomaly-atlas
3 3 # File : src/anomaly_atlas/stats/bootstrap.py
4 −# Purpose : Block/stationary bootstrap and confidence intervals
4 +# Purpose : Moving-block bootstrap and percentile confidence intervals
5 5 # Author : Simon-Pierre Boucher
6 6 # Contact : contact@spboucher.ai
7 7 # Data src : hfmarketdata.io (sole data source)
@@ -10,7 +10,49 @@
10 10 # Platform : macOS / Apple Silicon (arm64)
11 11 # License : All rights reserved (research code)
12 12 # =============================================================================
13 −"""Block/stationary bootstrap and confidence intervals.
13 +"""Moving-block bootstrap for serially dependent data (Künsch 1989).
14 14
15 −Stub scaffolded 2026-08-12; implemented in later phases (see CLAUDE.md).
15 +Blocks preserve short-range dependence, so statistics like AC1 or variance
16 +ratios get honest sampling distributions. Every call takes an explicit seed.
16 17 """
18 +
19 +from __future__ import annotations
20 +
21 +from collections.abc import Callable
22 +
23 +import numpy as np
24 +
25 +
26 +def moving_block_bootstrap(
27 + x: np.ndarray,
28 + stat: Callable[[np.ndarray], float],
29 + block: int,
30 + n_boot: int = 500,
31 + seed: int = 0,
32 +) -> np.ndarray:
33 + """Bootstrap distribution of `stat` using moving blocks of length `block`."""
34 + x = np.asarray(x, dtype=float)
35 + n = len(x)
36 + if n < 2 * block:
37 + return np.array([])
38 + rng = np.random.default_rng(seed)
39 + n_blocks = int(np.ceil(n / block))
40 + starts_max = n - block + 1
41 + out = np.empty(n_boot)
42 + for i in range(n_boot):
43 + starts = rng.integers(0, starts_max, n_blocks)
44 + sample = np.concatenate([x[s : s + block] for s in starts])[:n]
45 + out[i] = stat(sample)
46 + return out
47 +
48 +
49 +def percentile_ci(samples: np.ndarray, alpha: float = 0.05) -> tuple[float, float]:
50 + """Two-sided percentile confidence interval."""
51 + s = np.asarray(samples, dtype=float)
52 + s = s[np.isfinite(s)]
53 + if len(s) == 0:
54 + return (float("nan"), float("nan"))
55 + return (
56 + float(np.percentile(s, 100 * alpha / 2)),
57 + float(np.percentile(s, 100 * (1 - alpha / 2))),
58 + )
modified src/anomaly_atlas/stats/leadlag.py +45 −3
@@ -1,7 +1,7 @@
1 1 # =============================================================================
2 2 # Project : anomaly-atlas
3 3 # File : src/anomaly_atlas/stats/leadlag.py
4 −# Purpose : Lead-lag tests: cross-correlation, Granger, lagged regressions
4 +# Purpose : Lead-lag tests: lagged cross-correlation and asymmetry
5 5 # Author : Simon-Pierre Boucher
6 6 # Contact : contact@spboucher.ai
7 7 # Data src : hfmarketdata.io (sole data source)
@@ -10,7 +10,49 @@
10 10 # Platform : macOS / Apple Silicon (arm64)
11 11 # License : All rights reserved (research code)
12 12 # =============================================================================
13 −"""Lead-lag tests: cross-correlation, Granger, lagged regressions.
13 +"""Lagged cross-correlation between two return series.
14 14
15 −Stub scaffolded 2026-08-12; implemented in later phases (see CLAUDE.md).
15 +Sign convention: lag k > 0 means x LEADS y by k bars — corr(x_{t-k}, y_t).
16 +Validated on synthetic ground truth (charter §8.1) before real data.
16 17 """
18 +
19 +from __future__ import annotations
20 +
21 +import numpy as np
22 +
23 +
24 +def lagged_xcorr(x: np.ndarray, y: np.ndarray, max_lag: int) -> dict[int, float]:
25 + """corr(x_{t-k}, y_t) for k in [-max_lag, +max_lag]; k>0 = x leads y."""
26 + x = np.asarray(x, dtype=float)
27 + y = np.asarray(y, dtype=float)
28 + n = min(len(x), len(y))
29 + x, y = x[:n], y[:n]
30 + out: dict[int, float] = {}
31 + for k in range(-max_lag, max_lag + 1):
32 + if k >= 0:
33 + a, b = x[: n - k] if k else x, y[k:] if k else y
34 + else:
35 + a, b = x[-k:], y[: n + k]
36 + if len(a) < 3 or a.std() == 0.0 or b.std() == 0.0:
37 + out[k] = float("nan")
38 + continue
39 + out[k] = float(np.corrcoef(a, b)[0, 1])
40 + return out
41 +
42 +
43 +def peak_lag(xc: dict[int, float]) -> int:
44 + """Lag with the largest |corr| (ties: smallest |lag|)."""
45 + finite = {k: v for k, v in xc.items() if np.isfinite(v)}
46 + if not finite:
47 + return 0
48 + return min(finite, key=lambda k: (-abs(finite[k]), abs(k)))
49 +
50 +
51 +def leadlag_asymmetry(xc: dict[int, float]) -> float:
52 + """Sum of corr at positive lags minus sum at negative lags.
53 +
54 + Zero (in expectation) for synchronous series; positive when x leads y.
55 + """
56 + pos = sum(v for k, v in xc.items() if k > 0 and np.isfinite(v))
57 + neg = sum(v for k, v in xc.items() if k < 0 and np.isfinite(v))
58 + return float(pos - neg)
modified src/anomaly_atlas/stats/reversion.py +70 −3
@@ -1,7 +1,7 @@
1 1 # =============================================================================
2 2 # Project : anomaly-atlas
3 3 # File : src/anomaly_atlas/stats/reversion.py
4 −# Purpose : Mean-reversion tests: variance ratios, Hurst, AR, half-life
4 +# Purpose : Mean-reversion tests: variance ratios, AC1, AR half-life
5 5 # Author : Simon-Pierre Boucher
6 6 # Contact : contact@spboucher.ai
7 7 # Data src : hfmarketdata.io (sole data source)
@@ -10,7 +10,74 @@
10 10 # Platform : macOS / Apple Silicon (arm64)
11 11 # License : All rights reserved (research code)
12 12 # =============================================================================
13 −"""Mean-reversion tests: variance ratios, Hurst, AR, half-life.
13 +"""Mean-reversion statistics on return series.
14 14
15 −Stub scaffolded 2026-08-12; implemented in later phases (see CLAUDE.md).
15 +Validated on synthetic ground truth before touching real data
16 +(benchmarks/synthetic/test_synthetic_gate.py — charter §8.1).
16 17 """
18 +
19 +from __future__ import annotations
20 +
21 +import numpy as np
22 +
23 +
24 +def ac1(returns: np.ndarray) -> float:
25 + """Lag-1 autocorrelation of a return series."""
26 + r = np.asarray(returns, dtype=float)
27 + if len(r) < 3:
28 + return float("nan")
29 + a, b = r[:-1], r[1:]
30 + sa, sb = a.std(), b.std()
31 + if sa == 0.0 or sb == 0.0:
32 + return float("nan")
33 + return float(((a - a.mean()) * (b - b.mean())).mean() / (sa * sb))
34 +
35 +
36 +def autocov1(returns: np.ndarray) -> float:
37 + """Lag-1 autocovariance (input to the Roll spread estimator)."""
38 + r = np.asarray(returns, dtype=float)
39 + if len(r) < 3:
40 + return float("nan")
41 + a, b = r[:-1], r[1:]
42 + return float(((a - a.mean()) * (b - b.mean())).mean())
43 +
44 +
45 +def variance_ratio(returns: np.ndarray, q: int) -> float:
46 + """Lo-MacKinlay variance ratio VR(q) with overlapping q-period sums.
47 +
48 + Ground truth: VR = 1 for a random walk, < 1 under mean reversion,
49 + > 1 under momentum. Unbiased variance estimators, demeaned.
50 + """
51 + r = np.asarray(returns, dtype=float)
52 + n = len(r)
53 + if n < q + 2 or q < 2:
54 + return float("nan")
55 + mu = r.mean()
56 + var1 = ((r - mu) ** 2).sum() / (n - 1)
57 + rq = np.convolve(r, np.ones(q), mode="valid") # overlapping q-sums
58 + # Lo-MacKinlay bias-corrected PER-PERIOD variance of q-sums: the factor q
59 + # lives inside m, so the ratio below is varq/var1 (NOT varq/(q*var1)).
60 + m = q * (n - q + 1) * (1 - q / n)
61 + varq = ((rq - q * mu) ** 2).sum() / m
62 + if var1 == 0.0:
63 + return float("nan")
64 + return float(varq / var1)
65 +
66 +
67 +def half_life(log_prices: np.ndarray) -> float:
68 + """Mean-reversion half-life from an AR(1) fit: dp_t = a + b*p_{t-1} + e.
69 +
70 + Returns ln(2)/-ln(1+b) in bars for b in (-1, 0); +inf if b >= 0
71 + (no reversion). Matches the OU generator's ln(2)/-ln(1-kappa).
72 + """
73 + p = np.asarray(log_prices, dtype=float)
74 + if len(p) < 10:
75 + return float("nan")
76 + x, y = p[:-1], np.diff(p)
77 + vx = x.var()
78 + if vx == 0.0:
79 + return float("nan")
80 + b = ((x - x.mean()) * (y - y.mean())).mean() / vx
81 + if b >= 0.0 or b <= -1.0:
82 + return float("inf")
83 + return float(np.log(2.0) / -np.log1p(b))
modified src/anomaly_atlas/validation/artifacts.py +84 −3
@@ -1,7 +1,7 @@
1 1 # =============================================================================
2 2 # Project : anomaly-atlas
3 3 # File : src/anomaly_atlas/validation/artifacts.py
4 −# Purpose : Artifact detectors: bid-ask bounce, staleness, look-ahead
4 +# Purpose : Artifact detectors: Roll bounce, staleness, LOCF resampling
5 5 # Author : Simon-Pierre Boucher
6 6 # Contact : contact@spboucher.ai
7 7 # Data src : hfmarketdata.io (sole data source)
@@ -10,7 +10,88 @@
10 10 # Platform : macOS / Apple Silicon (arm64)
11 11 # License : All rights reserved (research code)
12 12 # =============================================================================
13 −"""Artifact detectors: bid-ask bounce, staleness, look-ahead.
13 +"""Detectors for the mechanisms that manufacture fake anomalies in bar data.
14 14
15 −Stub scaffolded 2026-08-12; implemented in later phases (see CLAUDE.md).
15 +Doctrine (charter §2.1): every candidate anomaly must first be explained by
16 +these nulls before it may be called a regularity. Each function is validated
17 +on synthetic ground truth (charter §8.1).
16 18 """
19 +
20 +from __future__ import annotations
21 +
22 +import numpy as np
23 +
24 +from anomaly_atlas.stats.reversion import autocov1
25 +
26 +
27 +def roll_spread(returns: np.ndarray) -> float:
28 + """Roll (1984) implied effective spread: 2*sqrt(-Cov(r_t, r_{t-1})).
29 +
30 + In log-return space this is the RELATIVE spread. Returns NaN when the
31 + lag-1 autocovariance is non-negative (estimator undefined — typical for
32 + momentum or noise-free series).
33 + """
34 + cov = autocov1(returns)
35 + if not np.isfinite(cov) or cov >= 0.0:
36 + return float("nan")
37 + return float(2.0 * np.sqrt(-cov))
38 +
39 +
40 +def bounce_implied_ac1(returns: np.ndarray) -> float:
41 + """The lag-1 autocorrelation a pure Roll bounce would produce for this
42 + series: -s^2/4 divided by Var(r), with s the Roll implied spread.
43 +
44 + Because s is estimated FROM the lag-1 autocovariance, this equals the
45 + measured AC1 whenever AC1 < 0 — the useful output is the DECOMPOSITION:
46 + ``excess_reversion`` reports how much reversion remains after removing
47 + the bounce explainable by the observed spread level.
48 + """
49 + r = np.asarray(returns, dtype=float)
50 + s = roll_spread(r)
51 + if not np.isfinite(s):
52 + return 0.0
53 + var = r.var()
54 + if var == 0.0:
55 + return float("nan")
56 + return float(-(s**2) / 4.0 / var)
57 +
58 +
59 +def excess_reversion(returns: np.ndarray, rel_spread: float) -> float:
60 + """Artifact-adjusted AC1: measured AC1 minus the bounce null implied by an
61 + INDEPENDENT spread estimate ``rel_spread`` (e.g. a liquidity-matched
62 + spread level, or a quoted/estimated spread from another source).
63 +
64 + For a pure Roll series with the true spread supplied, this is ≈ 0.
65 + A genuinely mean-reverting series keeps a negative excess.
66 + """
67 + r = np.asarray(returns, dtype=float)
68 + var = r.var()
69 + if var == 0.0 or len(r) < 3:
70 + return float("nan")
71 + from anomaly_atlas.stats.reversion import ac1
72 +
73 + bounce_ac1 = -(rel_spread**2) / 4.0 / var
74 + return float(ac1(r) - bounce_ac1)
75 +
76 +
77 +def staleness_ratio(observed_mask: np.ndarray) -> float:
78 + """Fraction of grid slots WITHOUT a fresh print (0 = fully fresh)."""
79 + m = np.asarray(observed_mask, dtype=bool)
80 + if len(m) == 0:
81 + return float("nan")
82 + return float(1.0 - m.mean())
83 +
84 +
85 +def locf_fill(values: np.ndarray, observed_mask: np.ndarray) -> np.ndarray:
86 + """Last-observation-carried-forward fill of a gridded series.
87 +
88 + Slots before the first observation keep their original value. This is
89 + the (dangerous) join that manufactures stale-price artifacts — it exists
90 + here so experiments can measure that artifact explicitly.
91 + """
92 + v = np.asarray(values, dtype=float).copy()
93 + m = np.asarray(observed_mask, dtype=bool)
94 + for t in range(1, len(v)):
95 + if not m[t]:
96 + v[t] = v[t - 1]
97 + return v
17 98