DFlash: use the R2 keys the goldens were actually uploaded to - #826
Merged
Conversation
The operator uploaded to gautham-experiments/correctness_prompts/laguna-xs-2.1-dflash/ and the pinned keys omit the leading segment, so the fetch would fail 404 NoSuchKey after a full dispatch. This is the second time the prefix has been lost. First I stripped it myself, assuming "gautham-experiments" was the bucket already carried by R2_BUCKET_ENDPOINT -- the serial keys are written with no bucket segment, which made the assumption feel safe, and it was still me overriding what the operator had told me. Then the commit that corrected it (18ccd58) missed the merge of #824, which went in from the preceding commit, so main kept the wrong key. #825 did not touch paths. Fixed in both places -- the correctness step's env and the timed_prompt_pool entry -- and pinned by DFlashGoldenKeyTests so a third loss fails in CI rather than 30-40 minutes into a ranked dispatch. The test also asserts each key stays inside the signer's charset guard (^[A-Za-z0-9._/-]+$ from #822), because that guard refuses a key BEFORE signing and would otherwise turn a typo into a confusing dispatch-time abort. If the objects genuinely move, change that test in the same commit that moves them. swift test: 582 tests, 28 suites, green. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
anupsv
added a commit
that referenced
this pull request
Jul 31, 2026
…828) #826 added a "gautham-experiments/" prefix to the golden keys. That was wrong, and #824's original keys were right. Probe run 30613434387 established it in about a minute: correctness_prompts/laguna-xs-2.1-dflash/dflash_correctness_golden_hidden.json FOUND 185394b correctness_prompts/laguna-xs-2.1-dflash/dflash_benchmark_golden_hidden.json FOUND 185433b gautham-experiments/correctness_prompts/laguna-xs-2.1-dflash/... 404 NoSuchKey Those byte counts are exactly the generated goldens', and the SERIAL control key was FOUND in the same run, so the credentials and the bucket were never the issue. "gautham-experiments" is the BUCKET, already carried by R2_BUCKET_ENDPOINT. Worth being plain about how this went, because the pattern cost four dispatches: the operator's path string is ambiguous between bucket-plus-key and key, I picked one reading on the strength of the serial keys' shape, was told I had overridden them, switched to the other reading, and was wrong that time. Two confident guesses in opposite directions, ~35 minutes each. Neither reading was unreasonable; arguing about it was. The probe that ends the argument took a minute to write and a minute to run. Fixed in all three places -- the correctness step's env, the timed_prompt_pool entry, and DFlashGoldenKeyTests' pinned prefix -- and the test's doc comment now records the probe output as the evidence, so the next person reads a measurement instead of a rationale. swift test: 582 tests, 28 suites, green. Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
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
1 participant
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.
The goldens are at
gautham-experiments/correctness_prompts/laguna-xs-2.1-dflash/, and main's pinnedkeys omit the leading segment — so the fetch would fail 404 NoSuchKey after a full
dispatch.
This is the second time the prefix has been lost. First I stripped it myself,
assuming
gautham-experimentswas the bucket already carried byR2_BUCKET_ENDPOINT(the serial keys are written with no bucket segment, which made that feel safe — it was
still me overriding what I'd been told). Then the commit that corrected it,
18ccd58,missed the merge: #824 went in from the preceding commit. #825 didn't touch paths.
Fix
Both places — the correctness step's env and the
timed_prompt_poolentry:DFlashGoldenKeyTestspins them, so a third loss fails in CI rather than 30–40 minutesinto a dispatch. It also asserts each key stays inside the signer's charset guard
(
^[A-Za-z0-9._/-]+$, from #822) — that guard refuses a key before signing, so a typowould otherwise surface as a confusing dispatch-time abort rather than a bad key.
If the objects genuinely move, change that test in the same commit that moves them.
Verification
swift test: 582 tests, 28 suites, green.🤖 Generated with Claude Code
Need help on this PR? Tag
@codesmithwith what you need. Autofix is disabled.