docs(007): post-rc1 HOLD sync + seed-rc2 prep

[scrimshawlife-ctrl]

Summary

Operator surfaces still said rc1 was "launching" after Spark already REJECTed it. This sync records the real gate, then lands seed-rc2 prep (authorized by Danny continue training according to our updated contents).

• BEST stays seed-morph78
• rc1 REJECT (test non-inferiority misses)
• holdout SCORED_SPENT
• selection-surface audit NO_RULE_MATCH / HOLD
• rc2 AUTHORIZED — cold start + RELEASE_SET + strict filler + new holdout + HLX_FORCE_TRAIN_DISJOINT
• Train launch parked: this Cloud Agent has no morpheus SSH key (secrets pointed at bc-3b73859e-… are not readable from the New Project temp-repo run)

Files

• NEXT_MOVES_007.md (+ specs twin)
• STATUS.md / specs/007-hyperlexical-model/status.md
• HYPERLEXICAL-PRODUCT-PLAN.md
• receipt receipts/20260926-post-rc1-hold.md
• scripts/spark/soft_ceiling/rc2-train-env.json
• SPARK-RC2-RUN.md
• receipts/20260926-rc2-preregistration.md
• receipts/20260926-spark-ssh-blocked-rc2.md
• tests/shadow/test_hyperlexical_rc_tooling.py (rc2 env guard)

Not in this PR

No train weights, Hub, T13, or SoT changes. No fake local train.

Unblock train

Continue on the Hyperlex Cursor Environment that holds SPARK_SSH_PRIVATE_KEY (agent bc-3b73859e-…), or inject that secret into a run that can ssh spark, then follow SPARK-RC2-RUN.md.

[scrimshawlife-ctrl]

Closing without merge.

This draft records the rc1 REJECT, which is right, and then authorizes rc2 and adds a train env with HYPERLEX_ALLOW_TRAIN=1. That authorization is stale. rc4 did not launch: the pre-registered probe returned CHAR_WINS. BEST stays seed-morph78. The branch also includes host paths that do not belong on the public repo.

Do not reopen this as a train recipe. A later train needs a new pre-registration.

NEXT_MOVES_007.md on main still says the rc1 retrain is launching. That sentence is stale. This close does not rewrite it.