fix(recipes): make GB300 NCCL checks run and the nodewright assert name-agnostic

Two declared gates did not do what the recipes implied.

The nodewright-customizations health check hardcoded the Skyhook CR name
"tuning", but no-op.yaml renders a CR named "no-op". The render-aware
suppression only skips the assert when a component renders zero CRs, so
any component wiring a differently-named manifest asserted a CR that
never exists. The assert is now name-agnostic: the executor's
list-and-match path fails closed on an empty collection rather than
passing vacuously, and verifyNodewrightReady still enforces per-CR
completeness by name.

The GB300 NCCL net and nvls checks skipped rather than ran, because the
validator's applicability matrix registers only gb200 for EKS on both
variants and an unsupported combination returns Skip — leaving the
declared thresholds inert and the performance phase vacuously green.
Declare nccl-benchmark-profile: gb200/eks, the documented borrow for an
existing service with a new accelerator.

Also corrects two overlay comments: the fabric claim now states that the
overlays assume EFA to match the reference cluster rather than asserting
GB300 was qualified on it, and the Dynamo gate is marked as live from
the first run, since inference-perf has no accelerator gating.

Signed-off-by: Yuan Chen <yuanchen97@gmail.com>

Author: yuanchen8911

pkg/bundler/testdata/stock_render_golden.yaml

@@ -3,51 +3,51 @@
 #
 # One entry per leaf overlay: a digest over its fully rendered helm-deployer
 # bundle tree (sorted relative paths paired with per-file content hashes).
-a100-aks-ubuntu-training-kubeflow: c372718509da1b12167fb3b9dd73d5ab50cbd3e196bc1ccc907d760719dc070e
+a100-aks-ubuntu-training-kubeflow: d9e713041b0ecafb1f9bddcb9edce87d9a43bcc7e732a1cedd6226e79a4d1652
 a100-any: 443a3195639f0354c05d41c34e84310184e77fe19a1db4e4157248b423eda8be
-a100-eks-ubuntu-training-kubeflow: 2f95f78419ec6b24837a89b3168922b0c1cd0291234616b51977227ed0bc82ae
-a100-gke-cos-training-kubeflow: 98ae8ea5b594164b59aaec2b3abbd2d626ae94b1630f47562e49bfccb2d6958a
+a100-eks-ubuntu-training-kubeflow: efc43236df64d337bba806d9ecf1fee04749242d1c538e645f0b35125c66a111
+a100-gke-cos-training-kubeflow: f3744d9387bf227c38373674a0fe9a0812b561ff943ca55b9f94ab57393d2518
 a100-oke-ubuntu-training-kubeflow: 1539a64bf1eb692e759101fd13115083c657779060f40719d0d2f3b70d801d31
 b200-any: e64aac932ee176d835eee0a326ec4b77c30b2e97afd0dc5f22e8c89640c37da2
-b200-gke-cos-inference-dynamo: a19cac53536564f9cec76aa65723eaf1eb2c21095565c7d23809be3ca1b97ac2
-b200-gke-cos-training-kubeflow: 9fd82f03b9700181f1efd16f6c572d5b3a5f44874c9bcf255f43dc55f63c1e65
-bcm-inference: 3adcec67f5e7745ce21c2fcec49143b76b29f66abe207f47d42a7575d772ba27
+b200-gke-cos-inference-dynamo: 3bac9410068fa1f73e2dc61799d5cbedbfc8b721e564f7334b8b054b6e171b5c
+b200-gke-cos-training-kubeflow: 094c79ca01541a09c183cbc1d35c8f25408f015a8519b67d1c2370d4e4bb8fa4
+bcm-inference: eba0c9554ee16f0b79f169c2f69017233c69d7b418ea0b23d54cda98d3d0eec6
 gb200-any: 5ceba4eb7316ca448369bcb4dd20240247cd15c3d3215b354c738dec8f359162
-gb200-eks-ubuntu-inference-dynamo: 4dd15671a1126562b30389b73d93f6357aa30df13465d4b44a76cf834dbef52b
-gb200-eks-ubuntu-training-kubeflow: dcad2ea2c6b3742670d4f50925715d7a5cbd3cf0526d94789659b099fe75e444
-gb200-eks-ubuntu-training-slurm: 44538377243945fefb4b037fee2c9b316a74894bfeacdd3bc0bde13188c62930
+gb200-eks-ubuntu-inference-dynamo: b3e7aa31e71b6a40ac9be1e5c4320e2943ad64f2f19a000d4a1b60c45ac3aa24
+gb200-eks-ubuntu-training-kubeflow: 6d22eac954455791a7f5dfce7b1f301f6c57624838a717031c388539990b9c52
+gb200-eks-ubuntu-training-slurm: 22a4cccff15a46ba774ef981ea5f6035dab9283c57028b77219c2f7cfa95f12b
 gb200-oke-ubuntu-inference-dynamo: 0148baa4731d28f0251a8dc121117085ce445b638f3b27f257c6d7bc9ee35ab9
 gb200-oke-ubuntu-training-kubeflow: e6bcb925712d588a1596bb9825169a944609a7f6b0ab8981810877652d678f5d
 gb300-any: 2cd40fe86553cc08b3bfdf3f8f5c11e3c50e84a4d0b22494f274fa62e9ab01e9
-gb300-eks-ubuntu-inference-dynamo: 51fa43bea7ae52ec6d75a618e12672800d293105a7303adc8c87a8fb0237c799
-gb300-eks-ubuntu-training-kubeflow: 8f0bfaf20e300af5f3417fd5e1f0f2fa65e27841cba81d254360f09c47a9821b
-h100-aks-ubuntu-inference-dynamo: 5464797e850b026cf6b1750c5dfca195809f7d1d9a4e8550c6048139743e281e
-h100-aks-ubuntu-training-kubeflow: 56166fa188acd730492e67d8f0bf7b82bd47ec7ec3a9bdbf69ee66ae307c5241
-h100-aks-ubuntu-training-slurm: 41ef1f36d370c064e8b452ec2dffe6b9dae5193de3d073c766e97c24bc5b130a
+gb300-eks-ubuntu-inference-dynamo: 8f4a35471d993926d3877d22ee3a379755d5e8a7c37580a37bdf808fd6287187
+gb300-eks-ubuntu-training-kubeflow: 51d4cfa9314323d79c152f88157c2d589123417403d7c805c5908f1e78e9fb38
+h100-aks-ubuntu-inference-dynamo: 60106eca2412d6fad922a4524a09d2d68b7526eaa7f3ff49fd43d0cd7a26c4ce
+h100-aks-ubuntu-training-kubeflow: 231fb345f168a21fc81a5b2ee6fb3a072dd88ac5a64f4b83368a47c0f016bc15
+h100-aks-ubuntu-training-slurm: 4e358a78b920a7abe8f29562ee74c3ca93d9ed9cd89150ca4bf7ceb73ac680d7
 h100-any: b96b596863029f47ee8759d159e12db7c9863589379c62ab5d859287bb353667
-h100-bcm-ubuntu-training: cc6ec7a4dec6c0eb28fde65a92c64f4e1a225ef1b2f7d421e2829676e92157b8
-h100-eks-ubuntu-inference-dynamo: 865c0cde4d8a6bc2b37a097a21b016039db923e0c6a4f791b58bf94647af6662
-h100-eks-ubuntu-inference-nim: ef6db54bfd14fff7db0553524600071cf22315c0a049d2d86ef0d07b51ce20db
-h100-eks-ubuntu-training-kubeflow: bb1c03beb00233862006e4d7b741c756f4bcfcde89831c0a77a7dc4f25ecfb20
-h100-eks-ubuntu-training-slurm: 3188554d15b20ce5e5a042e8231bbb865a3c9bd236eade52902804d2bcbbd535
-h100-gke-cos-inference-dynamo: d3d3f1aab57c4204d00c2aa06b4e367d11372bf9d94762207c8d906d142c7a2c
-h100-gke-cos-training-kubeflow: ae4f3d89a6662adc47be27108cd5d1d211be0d9f02260bc8c5510c6059a8bff5
-h100-gke-cos-training-slurm: c5b1bc34d751cf7ef641df5bcea3643f5fb37945fef0dc3483ec7a6065681ab7
+h100-bcm-ubuntu-training: aa4c2bd1b395fd29b57dd9226ac0b04eaf3edcd39a853a113a379dfea2f31cb9
+h100-eks-ubuntu-inference-dynamo: 71d5ea9ccb46dd4cb1cf1685faa6d3f8b8404dacf38935f6ce5ca319c6c8b99b
+h100-eks-ubuntu-inference-nim: 1d79a4e312bc47486493bf835ac2264d8786c849b6e6c88c355d7c9084381df6
+h100-eks-ubuntu-training-kubeflow: a121f87063837539f4c26710d37e587d8c3c3be95a7446b6e80088273c1e7625
+h100-eks-ubuntu-training-slurm: d37883a04011b600fc1a24350c20044f08c317138f40bfe674cb0a7ba235d726
+h100-gke-cos-inference-dynamo: fbae6625cfea5c37079cae55bfbd05d177f59c9c8c766b098f3291065b0c660c
+h100-gke-cos-training-kubeflow: 4486d628426f471c3b1ec087947e798ae6998cbe339bf8188570fddb9450a135
+h100-gke-cos-training-slurm: fa6d26058170f02e3ae099c57ba6cb2afa8d890298d6efc52cc21d929d34c325
 h100-kind-inference-dynamo: 3b5c3aa86fe396d355fc77a6a7bd2fb6b87d7b7d9a6abf272b2ed41023920673
 h100-kind-training-kubeflow: c2a29960328fc03e6ce67e5e8f25d6fe750886909ca31ee63750b5150a1e6a76
 h100-kind-training-slurm: 3cfbdfb63490de808042b535b8ad92ba45e1dd63d02401ecf57163765f36a4fe
 h200-any: c84e22d03d87a174b5215d883e9131e8a6c5f1befedc179a2cd2ca2bded39144
-h200-eks-inference: 17d5b7b4ee2d82443a89e9c403625aaa38a1f70ec6f4fee3a34d02c3210fc648
-h200-eks-training: c7b713eb05c1fb221e7e85c4e84f6845e34197a6375077942800e033199faeab
+h200-eks-inference: bbae5686447c4b4c0dec4b8a5e263710fe67f6fce61b2635bab3a9a335502430
+h200-eks-training: 57b10be6b4878f1b9aabfa2df09721cb7fa362cba411335ad3052f50ab3822c2
 l40s-any: 384bb817ae61f86388cb87e8ff28287cc518988c3e0f65468d64f4e488be58ca
 l40s-oke-inference: 4fd3eaa4c3b3b9f9aae2597abe450c8c6f69ed44c8a25c858d38d264e175d377
 l40s-oke-training: b18da6c239e730f5e11e8da9a109e9e95fc5ddd5f1e947666b8119db5b517d3f
 monitoring-hpa: b35a6c42d73788facd17b104b554bf6d2ae32bf73b0193424a72e566f1b6ceaf
 ocp-inference-nim: bb00cdb191823b32da334bea70826c8a92c11b62d0e096918d43a8e1b043c361
 ocp-training: d7a213263630f2c25982d6f4a144df7d9d5784428d296ccda1b4dab5a42b98bb
 rtx-pro-6000-any: 5500f71ae8812ed1541588222a1dc61f0d4e1c8f8cd7fd9c6a21e29f4d647507
-rtx-pro-6000-eks-ubuntu-inference-dynamo: 112dea463beb0f2328a4c61f908c934fe0944b066b6eb650d69e160f3c187cee
-rtx-pro-6000-eks-ubuntu-inference-nim: 9f675d1c39282b7aa4eb884a1f2b2512a1186dfa529c8b5200967726f4be65dd
-rtx-pro-6000-eks-ubuntu-training-kubeflow: 87207e073c6c9234f655d997d4f605b76ddc7d04905686b94a224774af2f6843
+rtx-pro-6000-eks-ubuntu-inference-dynamo: 16a4de23f7d145e56f8da9ee3e067c47cd354e37b26cfd1f3c52ba4ff7728a93
+rtx-pro-6000-eks-ubuntu-inference-nim: eb765d6af75d71c346a3ce1f8bd8c6c8ab9f3282327988a2eb8a100386764b62
+rtx-pro-6000-eks-ubuntu-training-kubeflow: bc961a695a0e639e6fe79727f0284c9635e26c208824c63bac8526648446b53c
 rtx-pro-6000-lke-ubuntu-inference: d08362d22a96a8992c4669d6bf334fc06f6cbf050a883f456fe10674d9d925eb
 rtx-pro-6000-lke-ubuntu-training: 5b0b3d9cc55eea2f60bcc80a54a250e5a03a204dc506863d137d75b8a70a4072

pkg/recipe/testdata/catalog_parity_golden.yaml

@@ -3,51 +3,51 @@
 #
 # One entry per leaf overlay: sha256 of its deterministically-marshalled
 # resolved recipe. A moved digest means that recipe's resolved bytes changed.
-a100-aks-ubuntu-training-kubeflow: ffbf217dc9fd1b596e621db0c59d8ea5b5e6866caf263dbc232e250c9f1d3368
+a100-aks-ubuntu-training-kubeflow: 3dfa350aa517fa401b8474629659be7146aacbc25805e70de241e287f4b2e659
 a100-any: 62d1581b21fecd3465b69659bac3dc9e5438eb2612c98c1acdd7b19cc9237084
-a100-eks-ubuntu-training-kubeflow: 8221fea2ed20a162728161059975571733630817a9c517988cc747ce2652cf45
-a100-gke-cos-training-kubeflow: 56853465e3f4922b64cf4306b9d991cf5dd1d4d9070afa52df03dc27dfef8555
+a100-eks-ubuntu-training-kubeflow: 974a3c313452457605a42ae6e4669ca15ba747e91a4b8d0da85944323ff45291
+a100-gke-cos-training-kubeflow: 17df2d55ab7b62e3d1357dba60a20289881c2f647797dd5efb8b34b999e790fa
 a100-oke-ubuntu-training-kubeflow: 310a6ee2abdd7f8508882429d657c01bd23458ba635236c410934825e4f617dc
 b200-any: 63dbc4fe27b84321e8b1991cd98c13f85395a3919d13e0c6891599eaa1f1e729
-b200-gke-cos-inference-dynamo: 51431391a0d6d470709174ab9be8103fa64baa21f9bd1fc6c0373959b6bfe21c
-b200-gke-cos-training-kubeflow: bffdd1b939344ea9a8f40abdfcc00edea3908304e0e9f668d5b07d113fc1224b
-bcm-inference: 1871d4b08c3d7221d3eb72a70c765f696fff0c7e96b6d4111440882901d2287b
+b200-gke-cos-inference-dynamo: 6efe04d7048b6d9848c07bd90a315ed0a0641905d9c6794b22271c49323d6fdb
+b200-gke-cos-training-kubeflow: e151a51e1ad7ec31f77a7ec7c24961b8fdce94d569127cf5114864ba40ba1462
+bcm-inference: 43ab64c51d714450cbdcdb6b16022b9a89ddab55cab0c56e6de7a5d8d0f1e4a1
 gb200-any: 9ee6ab187b4c1b93cd859d86ada35e922d0873ffc6fe892cb084ad98baf09727
-gb200-eks-ubuntu-inference-dynamo: 77b97f2841e6c58754b22d45b7cbfc20ffc0582dc804ccdb7d72776429c53c20
-gb200-eks-ubuntu-training-kubeflow: 4329cd8773d0bd00428bbf6bdc53d2695742d7bd7bc16f07fd3019ed7a5854e4
-gb200-eks-ubuntu-training-slurm: cf2147f07cae42f8c4b171ab79022dac223e7c102e46469b202ee6ba2b27efbe
+gb200-eks-ubuntu-inference-dynamo: c0de54a7ade04e826a1f7cb1ef9c88c28b60c4aad6357eb0ee8fe9063112a3ec
+gb200-eks-ubuntu-training-kubeflow: 19220d5a2e9ec32954bd1b32bb4e874cc9f57ca1119d7e04990330ea2e3a59c7
+gb200-eks-ubuntu-training-slurm: 86a9358483ebb372d6a62d4702a585c6302f56518c509f2f42e414915fac8d28
 gb200-oke-ubuntu-inference-dynamo: 138c28fdd64d4a08a2c1349b5b998f2ee68cdf8de452724510d00f7296b1a5cc
 gb200-oke-ubuntu-training-kubeflow: 3959f47474edfac9b53a65939117d4364551fde0c9324246f195f144fe3beb12
 gb300-any: 7f68607dffbbfc912f19b00ffa833caa81ca675d25bc22cf612dbd0b76f39475
-gb300-eks-ubuntu-inference-dynamo: f83c02d93b81c1c774527755ad79d0c520fc9e569e0659389415eb5a2241b0ee
-gb300-eks-ubuntu-training-kubeflow: e6a6d171d7511a4fc9ca950cf48b20ac68f8230af36d1e6294565402ead55d21
-h100-aks-ubuntu-inference-dynamo: 61a6cdb6c3f1a5460332cef094fd76e14da87dc1b39bc7229fbbb8ccae0e5aac
-h100-aks-ubuntu-training-kubeflow: b4bf459485d8f5882393220de96e8ea4aef65fadae6b86f395b26fa6b3767c1b
-h100-aks-ubuntu-training-slurm: cd9488c6ae6eed9c43a08eeaaaa4ff5e9eb378f19256ac0c02d478706efb6048
+gb300-eks-ubuntu-inference-dynamo: 44bdad10e2340b164d1b2f906cdc61791f4eb5d2ab7ac83257ec725284f401e4
+gb300-eks-ubuntu-training-kubeflow: 23ff8f6bc4ac65d877ae79215ed7e22b817c62e2fbc974793e8fe67f7a1de89c
+h100-aks-ubuntu-inference-dynamo: 07117b4d11ec6042258f42fb08781b218880f58c301ec71f641beedef8d62806
+h100-aks-ubuntu-training-kubeflow: 2b23859469f426f76d39136a561eacee0e12cfe43a56c2abe72b0501bf9236ef
+h100-aks-ubuntu-training-slurm: 9776f8101cc5d7755681d6424f8b3f74f159af8f5c1436847bfa2c72ccf0c3ea
 h100-any: f5a55f03948075adde9b1b24fd650d98fa4d35b85f394d071ad3365ba05588a5
-h100-bcm-ubuntu-training: 22d3283e415997b5cfd747a842db026bff1a297afcce6b23247c12a384097d59
-h100-eks-ubuntu-inference-dynamo: dd981d3b52fcf02e2c24b1a940b236b02fd96a57e6e020e091090d7d3643c3af
-h100-eks-ubuntu-inference-nim: c75f6502b92431467692a77e5eb372f0633385c52e0c46fd8d8cdc49a1c1d72c
-h100-eks-ubuntu-training-kubeflow: 67b3879633e1d47fc932a4292719e692848aaba1ac6bb07a6ab52c3e5b9548e1
-h100-eks-ubuntu-training-slurm: 9b69a63af76b3595af773368147b73dc0f6ab3708eb21aa2a4dac93b0bce3e1d
-h100-gke-cos-inference-dynamo: a236779e81f74303bbc13c36fafb71158d05b6a9398d27fd5de9c5c013a183a9
-h100-gke-cos-training-kubeflow: 10db012bf1053759ef2610e7cc46eeee242a61f92c745e5d8175829b7b1cda6d
-h100-gke-cos-training-slurm: b5fd06a9b6764c596e2e33e7e79c193f01004a17af0ecf4901c64399ff2547a2
+h100-bcm-ubuntu-training: a4c56385b4f0908ebfb022899d5075555e3179d5edb55f08cc062548c196c318
+h100-eks-ubuntu-inference-dynamo: b79ae736a7d9bb20ca2ddb8117aae7e6cdd955f89b8face685b614701a7d64c9
+h100-eks-ubuntu-inference-nim: 9b52a271c4afd205b5b6c0346513e5980f052bb3d6b8a5c8c04c1ef9d2d101e1
+h100-eks-ubuntu-training-kubeflow: 501e3007f9d1b2391b6f196c5fdd769e9a2f180f33c25a9b4592c01dc4dc282a
+h100-eks-ubuntu-training-slurm: 4f5a721af39c8ea372fcdb85e5e740065b99ed4f1287d355d39174d09d023657
+h100-gke-cos-inference-dynamo: 3ba2ec40aecbd3910785df7343896773e07c563901373fdff34fcfb4bca0aba6
+h100-gke-cos-training-kubeflow: 52a6517d46a6be029a12115d8d0f7ccc0a4516270cc03ce9b172f5b3c03b1cd5
+h100-gke-cos-training-slurm: 6899367250adb995f0383eaee6cf0762673495b2b7442ff98e7296731edfb917
 h100-kind-inference-dynamo: 0e2d552b62b55b91f960c9721ed128b1d07f68be8517799473fbaa3baa1cc6e5
 h100-kind-training-kubeflow: 4bb03659faaa6a9a66d58dfc95572bdb54b75010c219c4bbefbd0eafeb8ed867
 h100-kind-training-slurm: 0245c786638ade1023ba4dc49e0f2efd56d2f789bdf447e946bbc3a3e652e734
 h200-any: acf1986d76037dddd025125eaa34eb7ecf3d31fe62cb80e7dc89dd042ebb42f5
-h200-eks-inference: 02fc22eeddf62c684fdc58d35b2728ed5b0aaf147d63424a96b2814764dd21b5
-h200-eks-training: 57b28d6401d77a2443462876fd39dd1dd85b57c73c797ee4549cc8cbed3b2ada
+h200-eks-inference: badc48bec4aa553f3b9aad5e89ee0d8fcbae3ee5d393b885f1aee0377024d764
+h200-eks-training: 3b7d225d76c247691cfbb0fe91bd98f5e4bb2863f4403df89ff2daa0a49f7b8f
 l40s-any: 89210ba5815f93ce3d7b1a2a78f7ba75d083556e940603a1bc812eac9aeccd0a
 l40s-oke-inference: 6c75ef93e86fb21dbdf73ac97089b894ca8da832164dc807e94592493ab7f7cc
 l40s-oke-training: 2a35011bd59cf3c727cea158d504c51bd59cba069620d3d6a4988d68972dee74
 monitoring-hpa: 15e93304e7e68997e7b15a0be4af360a29f10c869502e6dccb6a6cefb6519fd5
 ocp-inference-nim: 49245ecfc91f67fcc2ed0b305b00046a4164628cafc75ab98542e6dafa626821
 ocp-training: aaa8cd7f45ef42235b075936461d715e2e017ea2802df9e8017de38304a17915
 rtx-pro-6000-any: 9639d67639256a44be8566f3f8eda49ceb4025b6b140a3bc2a4b6d4768d07ffa
-rtx-pro-6000-eks-ubuntu-inference-dynamo: 941af41effe3efff910e1fe88eb0915621426a0080efb54682af30b597455580
-rtx-pro-6000-eks-ubuntu-inference-nim: 10cf9bcd387410a5441c9956c941238d50efcccf2f085a24e195cdd8b26753c1
-rtx-pro-6000-eks-ubuntu-training-kubeflow: ce9b10a01a63dc079033206a6a26e601fb7f84c8f809bc6322a0ee9dca57d9fe
+rtx-pro-6000-eks-ubuntu-inference-dynamo: 14c5d90b608eda5dd48cfb22a4ceabc50cc8ca50acf08df5adea5e2059f6fe60
+rtx-pro-6000-eks-ubuntu-inference-nim: 59cafd3a23b4cd52cdfff318fd39dcc5bd8b587ccd49146d5868bde69127d3b1
+rtx-pro-6000-eks-ubuntu-training-kubeflow: ff3a0be78e21b186458ebe1183742ba80077a7608b4d1d7ed1c27f9fdec486cf
 rtx-pro-6000-lke-ubuntu-inference: 28cdab0531f8ed6018391aba7cab1fbfad8ba551392de1b9138f3b88042f5340
 rtx-pro-6000-lke-ubuntu-training: 312c4cf49d0e8cec0377cd9858338d7106226a2252d81828a75c1d426a194a2f

recipes/checks/nodewright-customizations/health-check.yaml

@@ -50,13 +50,25 @@ spec:
   steps:
     - name: validate-skyhook-cr-complete
       try:
-        # Assert the Skyhook CR (e.g. tuning) in the skyhook namespace has
-        # status.status == "complete" per the Skyhook CRD schema.
+        # Assert a Skyhook CR in the skyhook namespace has status.status ==
+        # "complete" per the Skyhook CRD schema.
+        #
+        # Deliberately name-agnostic. The manifests do not agree on a name —
+        # tuning.yaml / tuning-gke.yaml / tuning-generic.yaml render "tuning"
+        # while no-op.yaml renders "no-op" — so a hardcoded name asserts a CR
+        # that never exists for any component wiring a differently-named
+        # manifest. A nameless assert uses the executor's list-and-match path
+        # (pkg/chainsaw/inprocess.go): it fails closed when the collection is
+        # empty (ErrCodeNotFound, not a vacuous pass) and passes when a live,
+        # non-terminating Skyhook reports complete.
+        #
+        # Per-CR completeness is not weakened by this: the render-aware Go
+        # readiness check (verifyNodewrightReady) derives the expected CR names
+        # from the component's rendered manifests and requires every one of
+        # them to reach complete, reporting each failure by name.
         - assert:
             resource:
               apiVersion: skyhook.nvidia.com/v1alpha1
               kind: Skyhook
-              metadata:
-                name: tuning
               status:
                 status: complete

recipes/overlays/gb300-eks-inference.yaml

@@ -27,8 +27,10 @@ spec:
     accelerator: gb300
     intent: inference
 
-  # Qualified on p6e-gb300r.36xlarge (GB300 NVL72, 4 GPUs per node, Grace
-  # ARM64 host, EFA scale-out fabric).
+  # Targets p6e-gb300r.36xlarge (GB300 NVL72, 4 GPUs per node, Grace ARM64
+  # host). This overlay assumes the EFA scale-out fabric, matching the
+  # reference EKS cluster whose GPU nodes carry an efa-only ENI. RoCE-fabric
+  # GB300 is out of scope here; fabric-aware selection is tracked in #1410.
   #
   # Specific constraints for GB300 on EKS inference workloads.
   # Constraint names use fully qualified measurement paths: {type}.{subtype}.{key}

recipes/overlays/gb300-eks-training.yaml

@@ -27,8 +27,10 @@ spec:
     accelerator: gb300
     intent: training
 
-  # Qualified on p6e-gb300r.36xlarge (GB300 NVL72, 4 GPUs per node, Grace
-  # ARM64 host, EFA scale-out fabric).
+  # Targets p6e-gb300r.36xlarge (GB300 NVL72, 4 GPUs per node, Grace ARM64
+  # host). This overlay assumes the EFA scale-out fabric, matching the
+  # reference EKS cluster whose GPU nodes carry an efa-only ENI. RoCE-fabric
+  # GB300 is out of scope here; fabric-aware selection is tracked in #1410.
   #
   # Specific constraints for GB300 on EKS training workloads.
   # Constraint names use fully qualified measurement paths: {type}.{subtype}.{key}
@@ -125,6 +127,18 @@ spec:
         - nccl-all-reduce-bw-net
         - nccl-all-reduce-bw-nvls
       constraints:
+        # Without this, both checks SKIP rather than run: the validator's
+        # applicability matrix registers only gb200 for EKS on the net and nvls
+        # variants, and an unsupported combination returns Skip — the
+        # thresholds below would be inert and the phase would report a vacuous
+        # pass. gb200/eks is the documented "existing service, new accelerator"
+        # borrow (see docs/user/validation.md), and it is the right template
+        # here because GB300 on EKS runs the same EFA scale-out fabric and the
+        # same NVL72 MNNVL/IMEX topology the gb200/eks profile was built for.
+        # Replace with a native gb300 matrix entry once GB300 runtime templates
+        # exist.
+        - name: nccl-benchmark-profile
+          value: gb200/eks
         - name: nccl-all-reduce-bw-net
           value: ">= 40"
         - name: nccl-all-reduce-bw-nvls

recipes/overlays/gb300-eks-ubuntu-inference-dynamo.yaml

@@ -56,12 +56,14 @@ spec:
     performance:
       checks:
         - inference-perf
-      # Conservative smoke gate mirrored from the GB200 EKS Dynamo overlay
-      # until GB300-specific inference baselines are available. GB300 NVL72
-      # exceeds GB200 per-GPU, so the GB200-derived floor should not false-fail
-      # a healthy run — it must be re-baselined on real GB300 hardware before
-      # it is treated as a regression gate. Model + concurrency are pinned so
-      # the gate stays valid independent of the compiled validator defaults.
+      # Smoke gate mirrored from the GB200 EKS Dynamo overlay until
+      # GB300-specific inference baselines are available. Unlike the NCCL
+      # checks, inference-perf has no accelerator gating, so this gate is live
+      # on GB300 from the first run — it is a real threshold, not a
+      # placeholder, and must be re-baselined on real GB300 hardware. GB300
+      # NVL72 exceeds GB200 per-GPU, so the GB200-derived floor should not
+      # false-fail a healthy run. Model + concurrency are pinned so the gate
+      # stays valid independent of the compiled validator defaults.
       #
       # NOTE: this throughput floor is a fixed absolute full-node value (here a
       # 4-GPU GB300 node) and is not normalized for GPU count, so a smaller SKU