Run76 series catalog: non-PuBe series and availability on nsdf-cli

Note 1 · Author(s): Anthony Villano · 14 Sept 2026 (revised 24 Sept 2026) · Status: Complete (2026-09-14)

Code Repository

Files this note rests on, each pinned to the commit that last changed it:

• R76/analysis_notes/nsdf_r76_availability.py — compares the run log with nsdf-cli’s list and, with --probe, with the live server; writes the CSV below. Pinned to 322b1b8.
• R76/nsdf_r76_series.csv — one row per Run76 series in the run log (235), with source, trigger, and what nsdf holds. Pinned to 322b1b8.
• R76/DataSeriesList.xlsx — the run log; sheet Run76 is the one used. Pinned to 8917088.
• R76/am_lead_series.csv — 317 Am/lead series from Runs 72–74 (see below). Pinned to c892af4.
• nsdf_dark_matter_cli/r_dataset.csv — the list bundled inside the installed nsdf-cli package (version 0.5.0; 78,088 files, each tagged R68 or R76). Not in this repository.

Question

Which Run76 data exists, and how much of it can actually be downloaded from NSDF, in particular anything other than PuBe?

How “available” was checked

In nsdf-cli a “file” is one dump of a series, e.g. 07221203_2025_F0001; a series is a run of many dumps F0001, F0002, … Two checks were made:

• The list bundled with the package. nsdf-cli ls reads this file and never asks the server, so it is frozen at the package release. Version 0.5.0 is still the newest release on PyPI, so there is no newer list to install.
• The live server (new in this revision). The download command asks a gen-url endpoint for signed URLs, and that endpoint answers 200 for any name, real or not. A 1-byte ranged request on a dump’s data file tells them apart: a real dump returns 206 with its true size (matching the local copy byte for byte), an absent one returns 404. Every series in the run log was checked this way.

Result: only PuBe is downloadable

The Run76 sheet lists 235 series. Sizes are download sizes (the whole dump folder, compressed; a dump is about four times larger once loaded into memory).

All 130 available series are PuBe, and none of the 65 non-PuBe series (42 Am, 2 Na-22, 21 with no source recorded) are on NSDF. The 130 hold 38,566 dumps, 1,550 GiB (1.51 TiB).

Has it changed since the table was last made?

No. The series counts, and which series are available, are identical to the previous version of this table. Checked against each source:

• Run log: the Run76 sheet is identical to the committed copy (235 series, same content in every row). The file was last saved on 14 Sept; its last commit (22 Sept) is described as a normalization after an Excel re-save. The sheet has an active filter that hides 62 of these rows; they are included in every count here (see the next section).
• Bundled list: the same, since 0.5.0 is still the latest release. All 130 available series appear in the run log, and no Run76 series in the log is listed there under the other run’s tag (R68).
• Live server: agrees with the list. The first and last listed dump of all 130 series exist, and none has dumps beyond its last listed one. None of the 105 series without a listing has anything on the server under F0001–F0005. The 317 Am/lead series from Runs 72–74 in am_lead_series.csv were also checked live (F0001 and F0002 each): none is on the server.

Limit of the live check: for series without a listing, only the first five dumps were tried. Of the 130 known series, 129 start at F0001 and one at F0002, so a series that exists only under higher dump numbers would be missed.

A filter in the run log hides 62 series

The Run76 sheet has an Excel filter switched on for its trigger column. It hides 62 rows: exactly the 62 DCRC3-triggered series. Opened in Excel, the sheet shows only 173 of the 235 series, and any list made from what is visible would be missing those 62.

Every count in this note includes the hidden rows. The reader used here returns hidden rows like any others, and this was checked against the raw file rather than assumed: 237 rows, 62 marked hidden, and 235 of the rows are series (the last row is a formula helper, not a series). The script prints the filter range and the number of hidden rows on every run, and the CSV marks each series the filter hides (hidden_by_filter). All 130 available series are among the 173 visible ones, so the filter hides only series that are not on NSDF. The workbook itself was not changed.

What the 130 available series look like

• Size varies enormously. 3 to 3,122 dumps per series (median 113), and a median of 4.8 GiB per series. The largest, 07221001_2314, has 3,122 dumps and 248.9 GiB. A single dump is 0.1 to 117 MiB.
• Dump numbering is not always complete. One series starts at F0002 (07221106_2320), and two have gaps: 07220706_1303 has 109 of the 210 dump numbers up to its last, and 07221119_1016 has 202 of 231.
• All 130 are type Side2.

Which PuBe series are not there

Of the 170 PuBe series, 40 are not on NSDF. They differ from the 130 that are in how they were triggered: 39 of the 40 use a DCRC3 trigger and one has no trigger recorded, whereas none of the 130 available series does. Across the whole run log, none of the 62 DCRC3-triggered series is on NSDF, whatever its source; these are also the 62 rows the sheet’s filter hides. The trigger is not the whole story, though: the 20 Am series triggered on DCRC1 or NaI are missing too.

The processing columns (proc0, proc1) do not explain it: 106 processed series are on NSDF and 86 processed ones are not. The reason these series were not uploaded is not in the data available here.

The 130 available series

One row per available series (all 235 are in the CSV). Scroll for the full list; “Downloaded” is the number of dumps of that series present in ~/idx on the machine this was run on.

Local copy: five empty stub folders

Only one Run76 series is downloaded: 07221203_2025, dumps F0001–F0005 of its 30. ~/idx also holds five empty folders for two series that are in the run log but not on NSDF: 07221022_1630_F0001–F0004 and 07221206_1738_F0001. They are left over from failed download attempts, since the download command creates the folder before fetching. This matters because it also skips any name whose folder already exists, so a repeated download of one of these names would silently do nothing even if the files were added later. They have not been deleted.

Am/lead series in Runs 72–74

R76/am_lead_series.csv lists the series tagged Am/lead in the neighbouring runs: 317 series (Run72: 8, Run73: 175, Run74: 134), none of them on NSDF, as above. This is a different set from the 42 Run76 Am series, and it does not include any Run76 series.

Status / Next Steps

• Complete for the question asked: there is no non-PuBe Run76 data reachable through nsdf-cli, and the available PuBe data is the 130 series above. All analysis so far uses PuBe series (notably 07221203_2025).
• Re-run nsdf_r76_availability.py --probe to refresh this. The live probe does not depend on the package version, so it will notice new uploads that nsdf-cli ls cannot.
• If Am, Na-22, or any DCRC3-triggered data are needed, they would have to come from outside nsdf-cli, or NSDF would have to add them. Whether there is a reason they were left out is a question for whoever uploaded the data.
• The 42 Run76 Am series and their run-log details are now in nsdf_r76_series.csv (filter source_shield starting with Am).
• Delete the five empty stub folders in ~/idx before any retry of those downloads.
• The filter in the run log is still switched on; anyone who opens it in Excel sees only 173 series. Clearing it in Excel (Data → Clear) would show all 235.