Overview

This analysis set out to explain a recurring anomaly in U.S. interstate pipeline data: a systematic mid-week dip in natural gas production-receipt nominations. What began as a day-of-week observation in pipeline data grew into a three-layer investigation spanning 900 days of flow data, per-cycle nomination history, and a formal tie-out to documented maintenance events. Three distinct mechanisms emerged, but point to the "Tuesday" dip being driven by something very prosaic - Monday workorders that get started on Tuesdays, then get fixed through the week.

Layer 1 — The calendar dip is real and broad-based

Using 900 gas-days (2024-01 to 2026-06) of production-category receipt nominations, production noms on Tuesdays run about −0.37% (−0.22 Bcf/d) below the centered seven-day trend — the weekly trough — while Sundays and Mondays are the peaks. The effect is statistically strong (Welch t = −5.15, p < 1e-6) and structurally broad: *29 of 37 major production pipelines show the Tuesday dip individually, so it is not the artifact of one or two large systems.

Crucially, the dip shifts one day in Monday-holiday weeks. When Monday is a holiday, Tuesday's dip is muted (−0.12%) and Wednesday becomes the trough (−0.48%), with Thursday rebounding (+0.45%). This one-day slide is the fingerprint of a Monday work-order / field-start cadence — when the work week begins a day late, the whole nomination rhythm follows. Operating capacity at production points shows the same Tuesday/Wednesday softness (−0.7% / −1.4%), peaking Monday.

Layer 2 — The revision mechanism: opens low, revised up

Isolated to production receipt points only (730 gas-days, ~1,750 matched locations/day), the per-cycle history revealed the true nature of the dip. Early- and mid-week noms open low and are revised upward the most. Wednesday opens lowest at the Timely cycle (−1.06%) with Tuesday neutral (+0.01%); both then carry the largest open-to-final revisions (Wednesday +0.99%, Tuesday +0.66%) and the biggest Timely→ID3 ramps (Wednesday +1.10pp, Tuesday +0.81pp). Late-week (Friday) revises down.

This reframes the anomaly: the Tuesday dip seen in the single best-cycle snapshot is largely a low-open-plus-upward-revision effect, not evidence that final production noms are actually depressed. At the final (ID3) cycle, matched production is in fact highest Monday/Tuesday and lowest Friday. The holiday shift, however, reproduces independently on this separate data — Wednesday becomes the deep trough (−2.18% final), Thursday the rebound (+2.37%) across 11 Monday-holiday weeks — confirming the calendar cadence is genuine.

Layer 3 — Tying revisions to maintenance events

The final question was whether these cycle revisions trace to documented maintenance. We pulled production-receipt revisions (Timely→ID3, 23 months, 60 pipelines, 35,699 pipeline-days) and joined them to a panel of 4,623 maintenance-relevant notices (from `1005_hdl.recent_pipeline_notices`), flagging each pipeline-day falling inside a notice window.

The tie-out is strong and directional:

- Maintenance-active days carry 2.3× the revision magnitude of normal days (|Δ| 6.30% vs 2.75%), with a strongly upward mean (+3.89% vs +1.10%); Welch t = 7.96, p < 1e-14.

- The effect is notice-type-specific: Maintenance (+4.28%, p=1.8e-11) and Operational Alerts (+4.97%, p=2.2e-8) drive large upward revisions — conservative early noms restored or rerouted up as events resolve. Capacity Constraints do the opposite — revisions are suppressed and flat (−0.07%, magnitude below baseline, p=0.002) because a binding ceiling caps upward movement. Force Majeure and OFO are not significant.

- Timing aligns: revisions build and peak 1–2 days after a notice goes effective (+3.04% at k=+2 vs ~1.7% pre-event), decaying by day +5.

Reconciliation and conclusion

Critically, maintenance does not explain the weekly calendar dip — notice count versus per-pipeline dip magnitude is uncorrelated (r = −0.02, ns). The two phenomena are distinct: (a) a systemic weekly nomination cadence (the Tuesday dip and its holiday shift, driven by the Monday-start work rhythm) and (b) event-driven revision amplification around maintenance, which governs the size and direction of revisions on affected days but not which day of the week they occur.

Caveats: the production panel is an exact-name match (~68% coverage); maintenance windows cap open-ended end dates; revisions are clipped at ±50% to remove data errors; and the current partial month (2026-05) is the only material gap. All findings are captured in the dashboard's three analytical sections (29 components, datasets 2–11).

Read the full analysis on the dashboard