Enrol the second-generation comparison as a required phase

[gunbai-bot]

What

--required-regen-fixed-point already exists and already works. It re-emits from the same authority transaction and compares against the first generation, and its receipt type makes copy-forward fabrication structurally impossible — the FixedPoint variant has no first_generation_equal field to fill in. What it lacked was a required consumer, which is exactly what gunbc.rung_drop floor_cut_regen_second_generation_agreement declares.

This enrols it as a sixth phase on the build lane, after regen.

What it can see — narrowly

When the first generation equals the committed mirrors, this pass is green by construction: it re-emits from the tree that produced the first generation, so only a nondeterministic emit can separate them.

So it catches emit nondeterminism and nothing else. In particular it cannot see a self-consistently wrong seed — a producer built from a wrong mirror emits that same wrong mirror and every generation agrees. Enrolling it must not be read as closing that gap; the rung-drop row states the same bound.

It is enrolled anyway because it is nearly free and its red is real. The expensive variant — install the generation, rebuild the producer, re-emit — buys only build nondeterminism on top, at a crate build per required run, and is deliberately not what this enrols.

Executed, not asserted

arm	result
green path	first_generation_equal=true (156 adjudicated) → fixed_point_equal=true referencing it
receipt absent	refuses
receipt from a different commit	refuses by name — the stale-prior coupling is guarded
fixed-point receipt offered as prior	refuses rather than building a fixed point on a fixed point

A pass that cannot run refuses instead of skipping: "could not check" reported as "checked" is the absorbing fallback.

Cost

The second pass is one extra emit, ~5 minutes on this tree — no install, no rebuild.

Why the phase sits on the build lane

It reads the receipt regen writes at target/stage0-regen-receipt.json, and target/ does not survive checkout, so a lane boundary between them would leave it with no prior measurement to reference.

The .dag roster is the authority and the host enum mirrors it. The existing both-directions join refuses on disagreement, so the roster edit alone would have failed as "a phase enrolled with no executor" — the wall did the census.

cargo clippy --release --bin claim_executor -- -D warnings: clean.

Follow-up, not done here

This PR discharges the regen arm of the widened obligation. Generated-artifact second-generation agreement remains absent. This PR does not authorize retiring the rung-drop row.

gunbc#10321 widened floor_cut_regen_second_generation_agreement's bounded population from the regen phase alone to every required phase that regenerates a committed population — generated-artifact included. --required-regen-fixed-point covers the regen population (emitted_population.rs, 157 entries) and covers nothing in committed_generated_artifacts (38 entries), and the two populations are disjoint. So enrolling it satisfies one arm of a multi-arm trigger, which under §4b(3) retires nothing: a drop is retired by its trigger and by nothing else, and a trigger naming a capability is not discharged by an artifact contributing to part of it.

🤖 Generated with Claude Code

https://claude.ai/code/session_01YRSSLXXVjavzktziwdpEpH

[briansrls]

REQUEST_CHANGES on exact head d9817e1.

The implementation is a valid partial step: it enrolls the stage0 regen second-generation comparison on the required build lane, fail-closed when no fixed-point verdict is produced. I found no code blocker in that scope.

But #10321 has since widened floor_cut_regen_second_generation_agreement from the regen phase to every phase that regenerates a committed population, explicitly including generated-artifact. This body still says, under “Follow-up,” that this PR “retires the trigger” and that only marking the row retired remains. That is now false: this head supplies the regen arm only; the generated-artifact population still lacks second-generation agreement.

Correct the durable claim before merge. The body should say this discharges only the regen arm of the widened obligation and does not authorize retiring the row. No code expansion into generated-artifact is required in this PR.