# CSX Weekly Service: methodology and publication review

Reviewed September 19, 2026. Scope: CSX only, reports dated March 29, 2017–September 16, 2026; weeks ending March 24, 2017–September 11, 2026. This is a descriptive within-railroad series. No other carrier, national total, ranking or adjustment for comparability is included.

## Evidence and revision status

- STB, [Rail Service Data](https://www.stb.gov/reports-data/rail-service-data/), retrieved September 19, 2026: official individual reports, consolidated reference and current correction notices. The current CN/NS notices do not resolve the CSX conflicts in this release.
- CSX, [Explanation of Methodology](https://www.stb.gov/wp-content/uploads/CSX_Methodology.pdf), retrieved September 19, 2026: STB identifies a partial revision effective January 24, 2018. That restatement concerns Item 6, outside these four measures. The document does not establish a complete history of unchanged collection practices for Items 1, 2, 3 and 5. Unknown implementation details remain unknown.
- Individual CSX EP 724 workbooks and their reporting-week headers, retrieved September 19, 2026, provide the period, field labels, source precision and train-type notes for each observation. Original URLs, hashes and UTC retrieval timestamps: [source manifest](/data/weekly-service-sources.json).
- STB, consolidated EP724 workbook through September 16, 2026, retrieved September 19, 2026: independent field-by-field reference, not a replacement for unavailable or conflicting weekly observations. Its source record is included in the manifest and data download.
- CSX, [August 22, 2017 operating-metrics announcement](https://www.csx.com/index.cfm/about-us/media/press-releases/csx-revises-service-metrics-to-better-reflect-performance/), reviewed September 19, 2026: company website measures are not mixed with the prescribed STB series.

## CSX carrier dossier and field concordance

This release has one carrier column. Cross-carrier proposals are classified **not reliably comparable on the reviewed evidence** and are excluded. Adding another railroad requires its own dossier, period-specific concordance and publication review. These reported values have not been normalized to another railroad's methodology.

| Field | System train speed (Item 1) | System terminal dwell (Item 2) | Cars online (Item 3) | Trains held (Item 5) |
|---|---|---|---|---|
| Workbook field | Rail Service (Item Nos. 1–6), B13 | Same worksheet, B27; ten terminal details B17:B26 | Same worksheet, B38 | Same worksheet, E59; cause totals B59:D59; train types rows 51–58 |
| Unit / aggregation | System average, mph | System weekly average, hours | Weekly average number of cars | System weekly average trains holding per day |
| Population | CSX line-haul movements between terminals | CSX system; ten largest terminals by cars processed shown separately | CSX reported cars online; eight reported car types | CSX network trains holding, by train type and cause |
| Exclusions / thresholds | Terminal time, local/passenger/maintenance-of-way trains and yard jobs excluded. Fewer than four trains of a type is N/A. | Workbook excludes cars on run-through trains. System average is separately reported, not averaged from ten locations. | Further operating-population exclusions not disclosed in reviewed methodology. | Fewer than four trains of a type is N/A. Train-type grouping and automotive notes are retained per workbook. |
| Operational collection | Internal train IDs classify trains; measurement system and weighting details not disclosed. | Collection system and weighting details not disclosed. | Collection system and snapshot cutoff not disclosed. | Field reports manually combined from approximately 03:30 daily snapshots. Time zone not disclosed. |
| Reporting period | Saturday–Friday | Saturday–Friday | Saturday–Friday | Saturday–Friday; daily snapshots averaged across week |
| Weighting / rounding | Source average used as filed; detailed weighting not disclosed. | Source average used as filed; detailed weighting not disclosed. | Source precision retained; component rounding checked separately. | Source averages retained, not multiplied into unique weekly trains or hours. |
| Changes / uncertainty | Company website changes are not substituted into STB data. Complete method history unknown. | Terminal list changes; no fixed-terminal panel is implied. | Complete population and method history unknown. | Automotive notes vary by week; complete method history unknown. Item 6 restatement is outside scope. |

Evidence for collection/population fields is the CSX methodology Items 1, 2, 3 and 5 and its reporting-week footnote, supplemented by the individual workbook labels. Exact implementation, weighting, time zone, unreported adjustments and historical continuity are not inferred where the sources do not disclose them. Partial or unexpected reporting weeks are withheld, not rescaled.

## Reported facts, calculations and interpretation

**Reported facts:** the original numeric cells, N/A entries, railroad notes, reporting periods and published methodology. `originalReported` retains source cells; `reported` is the chart-eligible projection, with nulls where publication checks withhold a value. A null is not a railroad-reported zero. Terminal and train-type details are source observations, not rankings.

**RailStats.org calculations:** weekly change = current − preceding week. Percent change = (current − preceding week) / preceding week × 100. Both inputs must be available and their week-ending dates exactly seven days apart; a zero prior value makes percentage change unavailable. Prior-week values may be outside the displayed window. Index = current / first displayed week × 100. All selected measures share the same starting week; missing or zero baselines make that series' index unavailable throughout the window. A later missing value produces a gap. No interpolation, smoothing, annualization or adjusted service score is used. Display rounds to two decimal places (cars to whole numbers); calculations retain source precision.

**Interpretation limits:** relative direction and co-movement are descriptive, not causal. An increase need not mean better service. Line-haul speed is not end-to-end shipment speed. Cars online is inventory, not cars delivered. Held-train snapshots may count the same train on multiple days and do not measure delay duration. Changes may reflect reporting populations or methods as well as operations; undocumented causes are not assigned.

## Validation and unresolved source issues

The preserved compilation contains 495 weekly slots; 494 individual reports were available. The August 21, 2019 report returned 404. Its expected week ending August 16 is retained with no values, explicitly marked expected rather than verified.

Twenty slots have at least one withheld chart measure. Including the missing workbook: speed has 5 withheld observations, dwell 4, cars 6 and trains held 18. In three reports the source dates conflict with the Saturday–Friday schedule; all four measures are withheld. These flags indicate unresolved source questions, not a finding that the railroad's data is erroneous.

Every original file is preserved and hash-validated. The extractor checks carrier/source identity, sheet and field labels, nonnegative finite headline inputs, reporting periods and chronological uniqueness. Individual fields are compared with the compilation. Differences compatible with the compilation's one-decimal rounding retain the weekly precision; other numeric conflicts withhold the affected measure. Blank/N/A/zero availability differences are recorded, with the individual report's markers retained.

Cars-online component differences of up to four cars are consistent with rounding eight integer components: retain the reported total and disclose the difference. Larger discrepancies withhold cars online. Trains-held cause totals, train-type totals, cause columns and complete train-type rows must reconcile within 0.000001; unresolved differences withhold the total and cause charts. Original source values remain available with source-review warnings. Three period anomalies and one missing file remain unresolved; no corrected figure is invented.

See [reconciliation notes](/data/weekly-service-review.json) for exact differences and the page's range-specific source notes for affected observations. Future corrections require re-extraction, comparison with this preserved release, methodology review where relevant and renewed validation before changing published figures.

## Storage and visitor behavior

Original STB files are preserved separately from derived outputs. The website serves one approximately 1 MB CSX JSON file when this page opens, and custom ranges filter that file in the visitor's browser. Visitors do not download hundreds of workbooks or query STB for each date change. No database or year-by-year loading is claimed. [Download the publication data](/data/weekly-service-csx.json).