Clean up the listing files and demos - #16
Merged
AlexCatarino merged 4 commits intoSep 25, 2026
Merged
Conversation
…ce norm Measured across the 85 published dataset pages, Data Summary rendered at 10,716 bytes against a maximum of 1,056 and Accessing Data at 8,148 against 4,744, and the 13F was the only repo using ### subsections in its listing files. The 20-field table dropped from Data Summary is exactly what the Data Point Attributes data-tree already publishes from the XML docs on SEC13FHolding, so it duplicated the page against itself. The value-scale and amendment warnings were written twice, once in each listing file, and now live only in the documentation. The partial-coverage disclosure survives as the Asset Coverage footnote, so the page still states that it reaches 97.0 percent of the reported lines rather than every reported name. The demo algorithms went from 81 and 140 non-blank lines to 65 and 111, against medians of 38 and 78 across the onboarded repos. Only prose and one duplicate log were cut: the trading logic is unchanged, because the demos cannot be run locally past 2021 and an unverified behaviour change is not worth the saving. vendorName is the registry's "Securities and Exchange Commission", which is what products 9 and 149 carry; the file said "U.S. Securities and Exchange Commission", which matches no vendor.
…ts dropped Asset Coverage said 8,520 US Equities, which is the number of securities that received a line in one week of August 2026, not what the dataset holds. The licensing card then read "Every position reported on Form 13F since 2013 ... across 8,520 US Equities", which is false by roughly a factor of two. The published output and the delivery tar both carry 18,877 ticker files, counted from disk, and that is now the figure in all three places, with the same footnote USPTO uses for the one other equity-linked dataset we onboarded. The count is of ticker files, which is one per point-in-time ticker, so it is slightly above the number of distinct securities the run itself logged; it is the number that can be verified from what ships. Three things the previous commit cut too far, found by reviewing that diff against the source: - The 97.0 percent resolution rate was one week's figure presented as the dataset-wide rate. It now carries its qualifier and the 95.0 percent of the fourth quarter of 2020 beside it, along with the reason they differ: the ticker crosswalk begins in late 2019, so the names most likely to be missing are the ones that delisted before then, which is a skew rather than noise. - Accessing Data said an amendment is published beside the filing it corrects. That is only true of a RESTATEMENT. A NEW HOLDINGS amendment supplements the original, so a reader who dropped the original on every 13F-HR/A would lose positions. - AmendmentType's summary described the two kinds without naming them, and the page tells the reader to apply the rule itself. The literals are now in the summary, which the data-tree publishes on both tabs. "1,462 filings" is gone from the Data Summary footnote: the count excludes 13F-NT notices, and once the sentence saying so was cut it named a set the reader could not identify.
Measured against the 85 published dataset pages, Introduction was the heaviest of all of them and About the Provider the heaviest of its section, while Historical Data was nearly twice the heaviest of ours. Every section is now inside the range the marketplace occupies. What went is either evidence written for a maintainer rather than a user -- the line by line agreement between EDGAR's daily index and the quarterly batches, the per quarter counts of option and debt lines -- or a field list that the Data Point Attributes data-tree already publishes from the XML summaries. Example Applications drops to four strategies, which is what the other repos carry. The measured facts a user needs all survive: the reporting lag and the share of filings that miss the deadline, the rule that nothing is aggregated, the shape of a point, and the difference between the Series, the flattened frame and the typed history.
…oved Every code block on the page was run in local LEAN against the published data, the classic and framework examples in both languages and the remaining fragments through a harness that executes each one verbatim. Two numbers did not survive it. Historical Data claimed that sixty days of AAPL history is a Series of 39 entries or a DataFrame of 6,333 rows. Those are the figures for a history call anchored on 1 December 2020, which nothing in the documentation tells the reader to make. Read at the start of the window Requesting Data sets up, the same call returns 36 entries and 4,144 rows, confirmed identically by the Python and the C# harness. The sentence now names the anchor so the number can be reproduced. The Requesting Data snippets carried "Coverage runs 2013-05-20 to 2026-05-29". The published history ends on 2026-09-18, and the end date moves every day the job runs, so the comment now gives the start alone. Checked and left alone: AAPL carries filings on 58 of the 66 weekdays of the fourth quarter of 2020, which is exact.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Match the other datasets.
Asset Coverage said 8,520 US Equities, which is one week's count rather than the history, so the licensing card claimed 8,520 names since 2013. It is now 18,877, the ticker files the delivery carries, counted from the output and from the tar. The page and the card are already updated.
The listing sections are cut to the size the rest of the marketplace uses. Data Summary was ten times the heaviest of the 85 published pages, and Introduction and About the Provider were the heaviest of their own sections. The field table that went is what the Data Point Attributes data-tree already publishes from the XML summaries, and those now name the two AmendmentType values so the page can stop spelling them out.
The demo algorithms lose prose and one duplicate log. Their trading logic is unchanged, because they cannot be run locally past 2021 and an unverified behaviour change is not worth the saving.