Which release did that tag name?

A client asks which build their customers are on, and the room answers with a tag. Tonight, three proofs, in a repository the agency or the client can open. Name the release that tag points at: the release id, the tag name, and whether it is a draft. Show a different release whose id or tag is not this one. Show another client's repository that does not list this release. If any of the three needs a person who is not in the room, you have not shown the release. You have a version string and a guess.

This is the release object. It is not the alert a dependency raised. Which alert did that dependency raise? is an alert number, a state, and an advisory. A fixed alert can sit on a commit that was never released, and a published release can ship a dependency that is still open. It is not the commit the client is running. Name the commit the client is running is a SHA. A tag can point at a commit, and the tag is still not the release id. It is not the review that blocked a merge. Which review blocked that merge? is a review state. A review can be required while the release is a draft, and draft is not a review. It is not the image a deploy ran. Which image did that deploy run? is a digest. A release can name a tag that no deploy has pulled.

None of what follows is a rate for this engagement, and none of it is legal advice. The fields below are one host's current description of a release. The scores below that are a 2022 paper about release notes on other people's repositories. A tag string is not a census of your clients. A paper that counted projects cannot name the release id in the repository you opened tonight. They do tell you which numbers people mix up when they try to sound precise about a release they have not opened.

GitHub's release API says a list of releases does not include regular Git tags that have not been associated with a release. A tag is not a release. The list defaults to 30 results per page, and the maximum is 100. The page parameter defaults to 1. 30 is not 100, and 100 is not 1. Creating a release takes tag_name, and the example body sets tag_name to v1.0.0, target_commitish to master, name to v1.0.0, body to Description of the release, draft to false, prerelease to false, and generate_release_notes to false. v1.0.0 is 6 characters. Description of the release is 26 characters. 6 is not 26. name and tag_name can print the same string and still be two fields. Do not delete name because tag_name already says v1.0.0. draft false is not prerelease false. Two booleans can both be false and still be two fields. The response schema says id is a required integer. The example request does not include an id, and the page does not print a sample integer for one. The id you can defend is the integer the repository returns, not v1.0.0 (GitHub, REST API endpoints for releases).

The generate-notes example is a second request, and it does not match the create example. Its tag_name is also v1.0.0, and its target_commitish is main. main is not master. previous_tag_name is v0.9.2, also 6 characters, and it is not v1.0.0. configuration_file_path is .github/custom_release_config.yml, 33 characters. The same page says an unspecified path uses .github/release.yml, 19 characters, or .github/release.yaml, 20 characters. 33 is not 19, and 19 is not 20. The generated notes are not saved. A generated body is not a release id. make_latest accepts true, false, or legacy. Three names. The default is true. Drafts and prereleases cannot be set as latest. true is not a draft, and legacy is not a tag. The latest-release endpoint returns the most recent non-prerelease, non-draft release, sorted by created_at, and the page says created_at is the date of the commit used for the release, not the date the release was drafted or published. A commit date is not published_at. published_at may be null. A null is not a tag.

Asset state is uploaded or open. Two names. open on an asset is not a draft release, and it is not an empty body. The get-by-tag endpoint returns a published release for that tag. A draft is not that response. Create returns 201. Get returns 200. Delete returns 204. A validation failure returns 422, and an unknown discussion category returns 404. 201 is not 200. 204 is not 201. 422 is not 404. The API version header on the page is 2026-03-10. That date is not v1.0.0, and it is not a commit date. Only users with push access receive draft releases in a list. A list that omitted drafts is not evidence that no draft exists. If the client cannot open the repository, you do not have the third proof. You have a changelog paragraph.

The study people reach for when they want a share of projects that write release notes is a different document, and it does not contain a release id. Huaxi Jiang, Jie Zhu, Li Yang, Geng Liang, and Chun Zuo study GitHub projects. They take nine languages from the TIOBE index and the top 100 repositories in each, which is 900. Nine times 100 is 900. The manual study says projects with release notes are 62.33% and projects without are 37.67%. 62.33 plus 37.67 is 100. The abstract says more than 54% of projects produce their release notes with pull requests. The manual-study paragraph says more than 54% of those projects with release notes document changes with links to pull requests. The abstract's denominator is projects. The paragraph's denominator is projects that already have notes. 62.33% is the share that have notes. It is not the share the abstract attaches to pull requests. The same paragraph says 41.9% of projects with release notes list changes directly, without categories. The introduction says 41% do not have the change category. 41 is not 41.9. 14.8% (133 of 900) use a pull-request template. 133 divided by 900 rounds to 14.8%. That cell matches. It is still not a release id (Jiang, Zhu, Yang, Liang, and Zuo, DeepRelease).

The dataset sentence and the category table do not share one count. They obtain 46,656 entries from 400 repositories. 400 is not 900. Table IV's quantities are issues fixed 18,564, non-functional 8,481, new features 7,311, and documentation 4,209. Those four sum to 38,565. 38,565 is not 46,656. The page does not print the remainder, so do not invent it. The note says the quantity is the number of each category in the entire dataset, and the method says they kept the top four categories. Four is not six. The introduction says the notes they studied can be classified into six main categories. 6 is not 4. The entry split is described as 8:1:1 of the 46K notes. 46,656 divided by 10 leaves a remainder of 6, so an 8:1:1 split of that integer is not three integers. The classification split is 7:1, a different ratio. 7 is not 8. The training set then grows from 42k items to more than 420k items, using four operations: synonym replacement, random swap, random insert, and random deletion. 4 is not 10. 42 times 10 is 420, and the page says more than 420k, so the result is not 420k.

Table III prints DeepRelease at ROUGE-1 precision 77.38, recall 71.97, and F1 73.21. The row against PR-Title prints 8.59%, 0.24%, and 4.99%. 77.38 minus 71.26 is 6.12, and 6.12 divided by 71.26 is 8.59%. 73.21 minus 69.73 is 3.48, and 3.48 divided by 69.73 is 4.99%. Those two cells match the subtraction. The prose says the F1 scores are typically between 0.2 and 0.4, and that DeepRelease can reach 0.6 to 0.7. The table prints 73.21, not 0.7321, in the same column as 4.99%. Read the cell that is printed. Do not move a decimal because the sentence used a different scale. Table IV's non-functional recall is 90.34 for the keyword baseline and 89.38 for DeepRelease. 89.38 is lower than 90.34, which is what the prose says, and it is not an F1. DeepRelease's F1 on that row is 88.01. 88.01 is not 89.38.

Table VI is milliseconds, and the average row matches the column. The ten entry times sum to 296.990, and 296.990 divided by 10 is 29.699, the printed average. The category times average to the printed 0.013. At 10 pull requests the entry time is 26.830 and the category time is 0.017. 26.830 is not 0.017. The prose says a cited earlier study put manual release notes at as much as eight hours. Eight hours is not 26.830 milliseconds. The runs were repeated ten times on one RTX 3090. 10 repetitions is not the vocabulary size of 30K, and it is not the API's default page size of 30. Embeddings are 128-dimensional. Hidden states are 256. 128 is not 256. Batch size is 8. Beam size is 4. LeadCM uses the first 9 tokens because 9 is the average entry length. 9 is not 8, and 8 is not 4. Random guess is repeated 30 times. 30 repetitions is not page size 30. The Electron snippet is tag v11.4.0, which is 7 characters, not 6, and the pull requests it names are 27993, 27941, and 28101. None of those is v1.0.0, and none of them is a release id the API page prints, because that page does not print one.

Do the three proofs on one real release, in the repository that actually holds it. First, the binding. Copy the release id, the tag name, and the draft flag. If the list omitted drafts because the token cannot see them, write that. Do not infer v1.0.0 because an example request uses it. If name is null, write null. Do not copy the tag into the name to fill the hole. Second, the other release. A different id, or the same repository with a different tag, is not this one. A paper's share of projects is not that second release. If you cannot show a release that is not this one, you have a counter, not a record. Third, the other client. A repository that belongs to someone else does not list this release id. If the only list is a shared organization and you cannot say which client owns it, write that. A client you cannot name is the finding. It is not a reason to tell them the customers are on v1.0.0.

Three moves feel like the proof and are a second problem. Publishing the draft because the tag was embarrassing, and then describing the missing id as a cleanup. Quoting 54% because it looks precise, while the release you were asked about is one row, and the two sentences that print 54% do not share a denominator. Asking a model which release the tag names. The model will pick a plausible version. It will not open the repository. A commit you have not kept, while the client is asking what is running, is name the commit the client is running. Opening that SHA does not name the release id.

The path is ordinary when an operator can open one client's release without opening the next client's release as the way to prove it. Gigbase is one agency workspace: projects, clients, and a subscription, distinct from the next workspace. The shape that matters here is one project you can point at, in a repository the client or the agency can open, distinct from the next project. That case is not a release ledger. Nothing in the description is a headcount, a duration, or a result figure. If the only control on the screen changes every client's tag together, you do not have this proof. You have a version banner.

There is a version of this afternoon where the agency should not take the work. The work is not sold under the agency's name. The client asked to meet the builders. The room only has a slide. The buyer only wants a template. Or the buyer wants a legal opinion about what a tag promises. A white label software development partner is the page for capacity sold under the agency's name. It does not rank anyone, including us. 62.33% is not this repository, and v1.0.0 is not this repository.

When the repository is real and the release is not on a screen you can open, the missing object is specific. The id, the tag, and the draft flag, a different release that is not that one, and another client's repository that does not list it. A white-label engagement that expects this will put the repository in the client's organization, or in yours, before the first tag, because a version that lives only in a vendor dashboard is not an id you can hand the client. One that can only paste v1.0.0 into chat is telling you who holds the release. Where the next slice is one capability and the acceptance sentence names the release id, an AI integration sprint is the shape that can require the three proofs as the definition of done. Where you are not sure which repository the current engagement tags, a two-week AI readiness assessment maps the systems. The findings are yours. They are not a promise that 54% of this tree writes notes, and they are not a tag.

The ruleset a push hit is a different object: which ruleset did that push hit?. Do not record it on this page. A ruleset id is not a release id. An enforcement value is not a tag.

The invoice a charge paid, and the schema a call required, are different objects: which invoice did that charge pay? and which schema did that call require?. Do not record either on this page. An invoice id is not a release id. A schema name is not a tag.

Before you tell the client which build is live, write the three lines. The release id, the tag, and the draft flag, or the field that is missing. A different release that is not that one. Another repository that does not hold it. Do not publish the draft to avoid the flag, and do not ask a model which tag it was. A paper can tell you how often other repositories wrote notes from pull requests. The binding is the release you can defend.