Generated by python -m src.agents.scout, reading findings/scout-queue.json.
Regenerate whenever a domain's results.json changes; this file is a
human-readable snapshot, not the source of truth.
What the scout does and doesn't do, per Phase 5's design: it sweeps
per-period metric series already computed by each domain's own
compute.py, z-scores every point against that same series' own
mean/stdev (never an absolute threshold), and ranks whatever crosses
|z| ≥ 2.0. It does not decide anything and does not compute anything
new — every value below is exactly what the cited domain already
produced; the only new arithmetic is the z-score itself. Read each
candidate as a prompt to look closer, using the reproduce command given,
not as a conclusion.
Scope note: the Phase 2/3 archive (the harvested Parquet store) has only one real harvest so far, so it has no history yet to sweep. This queue is built entirely from the Phase 4 domains' own computed metrics. Domain C (Korea/Taiwan) is excluded — it has no genuine per-period time series in its results, only a per-lag correlation table, which isn't what "unusual relative to its own history" means.
Metric: H3 mirror reconciliation relative difference, Belgium export
vs. Netherlands import (pipeline gas)
Period: 2022-10 — value: 0.193 (series mean ≈ 0.70, stdev used
to compute z)
What this means: this is the one month where Belgium's and the
Netherlands' declared figures for the same gas flow came close to
reconciling (19.3% gap) against a history where every other month
differs by 60–90%+. The scout flags this because it's statistically
unusual — not because a large gap is unusual (that's the norm for this
pair, per Domain B's own writeup) but because this one small gap is the
outlier.
Verify: python -m src.domains.domain_b_eu_gas.compute, field
h2_mirror_reconciliation.relative_difference_by_month["2022-10-01"].
Metric: H1 waste-oil/virgin-oil price ratio, Spain (China vs.
Indonesia benchmark)
Period: 2025-03 — value: 0.59
What this means: this is the one month Spain's "waste" import price
from China dipped meaningfully below parity with virgin palm oil, inside
a series where every other month sat at or above the domain's own 0.70
flag threshold (mean ratio 1.01 across the whole series). Domain E's own
results.md already called this out as "the only material dip."
Verify: python -m src.domains.domain_e_eu_waste_oil_transparency.compute,
field h1_price_ratio.es.ratio_by_month["2025-03-01"].
Metric: H1 national divergence index (official industrial YoY vs.
composite satellite signal)
Period: 2025-03 — value: 2.97
What this means: the scout independently rediscovered the exact
outlier Domain A's own hypothesis-declared ±2σ check flagged — official
industrial growth spiked to 7.7% that month while the satellite composite
was mildly negative. Domain A's own writeup attributes this to a likely
March catch-up effect following NBS's Jan/Feb reporting gap, not a real
divergence. Useful as a sanity check on the scout's method: same answer,
arrived at generically, with no domain-specific logic.
Verify: python -m src.domains.domain_a_china.compute, field
h1_national_divergence_index.divergence_index["2025-03-01"].
Metric: H1 waste-oil/virgin-oil price ratio, Netherlands (China vs.
Indonesia benchmark)
Period: 2023-01 — value: 1.27
What this means: the highest price-ratio reading in the Netherlands'
series — "waste" oil priced 27% above the virgin-oil benchmark that
month. Worth checking against the raw quantity/value pair for a possible
reporting or unit anomaly before reading it as a real price signal.
Verify: python -m src.domains.domain_e_eu_waste_oil_transparency.compute,
field h1_price_ratio.nl.ratio_by_month["2023-01-01"].
Domain D's EIA production QoQ growth, XOM's and CVX's quarterly supply-language scores, and Domain E's other mirror-reconciliation and price-ratio series (Belgium, Italy) were all swept and produced no point at |z| ≥ 2.0 — consistent with those series being noisy but not containing a standout outlier, per each domain's own results.
Password required.