[jesshaley]

Builds the first half of part (2) of [handed-on classes carried at the end] (Decided by Jess Haley, 2026-11-02, did-schema docs/plans/did_v2_go_forward_class_audit.md; docs/V2_PLAN.md G98 (a)): "a small named step on the hand-on path writes V2's superclass list for the five classes pass 1 hands on unchanged (syncrule, epochfiles_ingested, stimulus_response_scalar, stimulus_response_scalar_parameters_basic, openminds), and ensureClassBlocks is deleted."

ensureClassBlocks is NOT deleted here (see "Why the padding stays" below).

Commit 1: the hand-on step

did2.convert.handOnSuperclasses writes did2.build.superclasses(<class>) on a document whose migrator hands it on unchanged, and drops an EMPTY v1 block of a parent V2 hosts no block for (a non-empty one is left for validation). v1_to_v2 calls it for HAND_ON_CLASSES; counted in summary.hand_on_superclasses (bodies_written, by_class).

Measured (scratch probe, branch claude/W-padding-probe, not for merge)

Commit 1 plus a probe in ensureClassBlocks printing every body it changed, with class and callers, on every path. Full Test code run 41396033849 against the V2 base run 41394449293 (c2749f8):

corpus	padded / changed, base	padded / changed, probe	final documents	quarantined	carried notes
20211116	4453 / 547	4453 / 0	5036 -> 5036	0 -> 0	24 -> 24
B	7355 / 2497	7355 / 0	33427 -> 33427	0 -> 0	7583 -> 7583
Dab	14543 / 4101	14543 / 0	67942 -> 67942	0 -> 0 (1605 in pass 1, all resolved, as base)	17786 -> 17786
JH	22368 / 8	22368 / 0	82845 -> 82845	0 -> 0	72050 -> 72050
PRED	28 / 0	28 / 0	42 -> 42	0 -> 0	0 -> 0

Probe lines in the corpus jobs: 0 (the padding changed nothing on any DID path). In the unit-test job: 19, from three emitters no corpus exercises: demo, daqreader_epochdata_ingested, daqreader_image_epochdata_ingested (superclass list only).

Commit 2: those emitters build complete documents

• jDemoFold writes did2.build.superclasses('demo').
• daqreader_epochdata_ingested, daqreader_image_epochdata_ingested: the document kept beside its clock references gets its class's superclass list; handed on unchanged (no epoch document), they take the hand-on step (added to HAND_ON_CLASSES: the decision named the five the corpora hold; these two are handed on to epochMint the same way, and no corpus holds one).
• resolveValidIntervals: writes an empty time_reference block when the source clock stated no tolerance (found by reading, not by the probe).

Why the padding stays

A deleting version of this branch (kept as W-delete-attempt locally) was run through NDI's end-to-end with NDI-matlab#1149: NDI run 41399719227 (Dab) quarantined 15212 documents (1605 on V2), 41399705406 (fixture e2e) failed the same way. NDI's second pass builds documents by hand (ndi.migrate.internal.etaShape and the assemblers) and relies on the padding inside v1_to_v2's re-validation for empty blocks: first five quarantined were relative_time_reference without its time_reference block, manipulation without interaction, chemical and formulation without data. The DID corpus harness runs no NDI second pass, so class_blocks_padding reading 0 there does not show this. Deleting ensureClassBlocks therefore waits on NDI's second pass emitting complete documents (siblings first).

https://claude.ai/code/session_01FxKLA8Tbf9cpvDe7fuVeKf

---

Generated by Claude Code