e2e gate: a device lane for adaptive_cartesian_gpu, and a host twin for the worst calibration row - #369
Merged
Conversation
…or the worst row Closes the two coverage gaps section 4 recorded rather than fixed, to keep #368 narrow. adaptive_cartesian_gpu is mcsamplerGPU standalone and the ILE driver's own default. It reaches compute_hist through its own integrate adaptation block, not through a portfolio member, and had no lane. It takes _AC_N_MAX, measured: at --n-max 20000 with the factor on it stops on the budget at n_eff 230. The device prior-only adaptive_cartesian row held the worst |z| in the calibration table and had no host twin. It has one now, and the twin returns the same three columns. Device table re-derived at 8 seeds on ldas-pcdev12 CUDA slot 0; all eight pre-existing rows identical to the ldas-pcdev2 numbers. Worst |z| is now 2.54. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
oshaughnessy-junior
deployed
to
private-review-dispatch-rift
September 18, 2026 22:18 — with 
GitHub Actions
Active
GitHub Actions
Active
Three findings, all in prose, none in the lanes. The comment on test_adaptive_cartesian said --n-max 60000 makes that lane stop on --n-eff at ntotal 30000. That is seed 1000. Over eight seeds ntotal runs 30000 to 60000, so seed 1004 still spends the whole budget; the cap buys the n_eff column (250-319 instead of 91-220), not the stopping criterion. I had carried the same claim into the new _LANE_KW note. Nothing asserts the _AC_N_MAX override is still applied: at 20000 the device lane passes. A ntotal < n_max guard was measured and rejected, because it would flake on adaptive_cartesian seed 1004. Recorded rather than added. expect_device broken on purpose for the new lane: the same arguments with cuda="" complete on ldas-grid with a correct answer (z -0.79), so only the marker separates a host run from a device run there. Comment-only; ldas-grid 18 passed, 11 skipped. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
oshaughnessy-junior
marked this pull request as ready for review
September 18, 2026 23:03
oshaughnessy-junior
had a problem deploying
to
private-review-dispatch-rift
September 18, 2026 23:03 — with
GitHub Actions
Error
GitHub Actions
Error
oshaughnessy-junior
deployed
to
private-review-dispatch-rift
September 18, 2026 23:03 — with
GitHub Actions
Active
GitHub Actions
Active
oshaughnessy-junior
deployed
to
private-review-dispatch-rift
September 19, 2026 00:03 — with
GitHub Actions
Active
GitHub Actions
Active
This branch was successfully deployed
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.
Closes the two coverage gaps section 4 of the e2e analytic gate recorded rather
than fixed, so that #368 could stay narrow. #368 has merged, so this sits straight
on
rift_O4d.--sampler-method adaptive_cartesian_gpuis mcsamplerGPU standalone and the ILEdriver's own default. It reaches
compute_histthrough its ownintegrateadaptation block rather than through a portfolio member, and had no lane. It now
has prior-only and A=8 B=2 device lanes at
--n-max 60000. Two mutations of thatconversion, on a device: sending the histogram to the host makes the new lane
fail, deleting the conversion leaves it passing, which is the portfolio lane's job.
The device prior-only
adaptive_cartesianrow held the worst |z| and was the onerow with no host twin. It has one now, agreeing per seed on ntotal and n_eff.
Verified
RIFT_CI_REQUIRE_GPU=1make_e2e_calibration.py --seeds 8, ten device rows, ldas-pcdev12prior-only, adaptive_cartesian, 8 seeds, ldas-gridexpect_devicebroken withcuda="".travis/test-ci-roster.pyWorst |z| in the table moves from 2.39 to 2.54, on the new
GPU A=8 B=2, adaptive_cartesian_gpurow.Z_TOLERANCEstays at 5.Pre-undraft review (second commit)
Three findings, all prose, none in the lanes.
test_adaptive_cartesiancomment claimed--n-max 60000makes that lane stop on--n-effat ntotal 30000. That holds at seed1000 only; over eight seeds ntotal runs 30000 to 60000. I had carried the claim
into the new
_LANE_KWnote. Both now state the distribution._AC_N_MAXoverride is stillapplied: at 20000 the device lane still passes. A
ntotal < n_maxguard wasmeasured and rejected, because it would flake on
adaptive_cartesianseed 1004.expect_devicebreak-it control to the file, since a hostrun of the new lane is otherwise a passing physics run.
🤖 Generated with Claude Code