Monitor-hour rows
1,632
24 archived hourly positions.
Frozen · final
DAH-2026-09-07-r1Delhi Air Health Research Release 1
A reproducible snapshot of the Delhi Air Health archive from 6 Sept 2026, 4:00 pm through 7 Sept 2026, 3:00 pm. The release is separate from the continuously growing live archive.
PM2.5 completeness
78.0%
1,121/1,438 eligible monitor-hours.
PM10 completeness
83.9%
1,206/1,438 eligible monitor-hours.
Station feeds / sites
68 / 67
Feed identity is kept separate from coordinate-deduplicated physical sites.
Normalized snapshots
1,229
Frozen source-snapshot layer.
Raw ledger rows
142
Only provenance-tracked runs from v2.7 onward have raw source rows.
Provenance runs
1
Frozen collector-run records included in this release.
QC flags
0
Rules version 1.0.0.
Release files
Download exactly what an analysis used.
Primary analysis table
One frozen row per station feed × archive hour, with explicit observed / eligible-but-missing / pre-entry states.
monitor-hours.csvStation registry
Stable feed IDs, physical-site grouping, operator, region and frozen coordinates.
stations.csvCampus links
Frozen campus coordinates and the campus-to-reference-monitor link used at release time.
campus-links.csvNormalized snapshots
Frozen source-snapshot layer from which the monitor-hour analysis table was derived.
normalized-snapshots.csvCollection provenance
Collector versions, timestamps, source resource and ingestion counts for provenance-tracked runs.
collection-runs.csvQC flags
Versioned review flags. A flag never means a row was automatically excluded.
quality-flags.csvMetadata events
Observed station rename, provider, owner or coordinate-change audit events.
metadata-events.csvRaw source ledger
Raw CPCB OGD source records for provenance-tracked runs included in this release.
raw-observations.ndjsonRelease notes
Plain-text interpretation and provenance notes bundled with the exports.
README.txtIntegrity
Verify before analysis.
The manifest records byte length and SHA-256 for every release file. SHA256SUMS also includes the manifest itself.
Historical provenance boundary
1,162 normalized snapshot rows in this release predate per-run raw-ledger provenance. They remain in the frozen analysis dataset, but the project does not claim raw source-record lineage for those legacy rows.
Version lock
- Code
- 2.8.0
- Methods
- 1.1.0
- Analysis spec
- 1.0.0
- Schema
- 2
Finalized 7 Sept 2026, 4:41 pm. Release identifiers are never reused for a different dataset.
Independent reproduction
The website is not the analysis environment.
The source package includes research/reproduce_release.py, which uses only Python's standard library to verify checksums and recompute core completeness, network and campus-comparison statistics from these exported files. This is intentional: a headline number should be reproducible without Next.js, Supabase or the production UI.