Defend the number the README publishes, and stop quoting the best of three - #163
Merged
Merged
Conversation
…three The native-image job measured the time to first search, printed it, and passed whatever it said. Only a missing value failed it. So a dependency that started doing real work at startup could have taken it to 300 ms with no build going red, while the comparison table kept claiming 37 ms, and the paragraph under that table claimed the job would stop the sentence quietly becoming false. That was true of the reflection claim and not of the time. It now fails over a 75 ms ceiling. Loose on purpose: the measurement lands between 29 and 42 ms across recent runs on a shared runner, so the ceiling has room for a noisy neighbour and still catches a regression that doubles it. 37 ms was also no longer the number. The last four runs on main measured 31, 29, 42 and 34, so a single frozen figure in a table was going stale by design. The table now states the bound the build defends and says where the measurement usually lands. And the CLI is not 5.7 MB. That is the macOS arm64 binary; Windows is 6.8 and Linux x64 is 14.1. Quoting the smallest of three without saying so is picking a winner in the paragraph that exists to stop a reader drawing the wrong conclusion from exactly these three facts.
|
This pull request adds no capability the code did not already have. 2 files read · 1051 ms · against |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Found while checking whether the repository's About description was still true. Two of its numbers were
not, and one of them was a claim nothing was defending.
A measurement with no assertion under it
The
native-imagejob measures the time from process start to first search, prints it, and passeswhatever it says. The only failure path is a missing value. So a dependency that began doing real work at
startup could have taken it to 300 ms and no build would have gone red, while the README's comparison
table kept saying 37 ms.
The paragraph under that table claims the job stops the sentence quietly becoming false. That was true of
the reflection claim, because
--no-fallbackfails the build, and it was not true of the time.It now fails over a 75 ms ceiling. Loose deliberately: across recent runs on main the measurement lands
between 29 and 42 ms on a shared runner, so the ceiling leaves room for a noisy neighbour and still
catches the thing worth catching, which is a regression that doubles it.
37 ms was not the number any more
The last four runs on
mainmeasured 31, 29, 42 and 34. A single frozen figure for a measurement thatmoves was going stale by design, and it had. The table now states the bound the build defends and says
where the measurement usually lands, which is a claim that stays true without anyone editing it.
The CLI is not 5.7 MB
That is the macOS arm64 binary. Windows is 6.8 MB and Linux x64 is 14.1 MB.
Quoting the smallest of three without saying so is bad anywhere, and the paragraph it sits in is the one
added in
2.3.0to stop a reader concluding from an ARM target, a fast cold start and a small binary thatKdrant runs as the vector database. Picking a winner inside that argument undercuts it.
Also updated, outside this PR
The repository About description carried both numbers and did not mention the MCP server, while the repo
topics already carry
mcpandmodel-context-protocol. It now reads:The two fragile numbers are gone from it. They live in the README, next to the job that produces them and
now defends them, which is where a number that changes on every push belongs.