Android list-row PRESS: clickable container has no selective identity (label on child) — the identity gap #1269's id-demotion can't cover

[thymikee]

Summary

Sibling to #1269 / #1272. The #1269 fix (demote a non-unique id from identity) resolves the Android
list-get false-divergence class. It does not address list-presses, which are a distinct
identity gap with the same root spirit (non-selective identity → unstable positional disambiguation) but
no id to demote.

The gap

On Android, a list row's clickable node is a label-less container (role="linearlayout") — the visible
label lives on a child TextView. So a recorded press annotates to a target with no selective
identity at all: role only, no id, label-less ancestry. Measured (wave-3 E1, stripped-id fixture,
step-5 press "Connected devices"):

identity-mismatch, matchCount: 12, recorded={role:linearlayout} observed={role:linearlayout}, mismatches:[]

The 12 same-role containers are disambiguated purely by recorded position; on replay the position drifts
and it binds the wrong row — exactly the #1269 mechanism, but id-demotion is inapplicable because there is
no id.

Why #1272 doesn't cover it

#1272 demotes a non-unique id, falling back to the (selective) label. A press-container has neither a
unique id nor its own label — the label is one level down. So there is nothing for that fix to fall back to.

Fix direction (needs a design call)

Derive the press target's identity from its labeled descendant (or resolve the press to the labeled
node and record identity from there), so a list-row press carries a selective label identity instead of
a bare container role. This is an ADR-0012 identity-construction decision, distinct from #1269's
id-demotion — hence its own issue.

Evidence

• Full-flow stripped-id divergence (~/.agent-device-bench/wave3/E1/), the step-5 press targetBinding
  above. Get class fixed + measured in fix: demote non-unique ids from writer identity/selector chain (#1269) #1272; press class is what remains for Android list flows.

[thymikee]

Quantified post-#1272 (get-class fix merged): recorded an Android Settings gets→press flow with the demotion daemon and replayed n=20.

Step	Result
gets (Network, Connected) — the #1269 class	0/20 diverge (fixed)
press (Connected) — this issue	20/20 identity-mismatch, deterministic

The press annotates to a bare role="linearlayout" (matchCount high, no id, label on a child), so id-demotion doesn't apply and positional disambiguation misbinds every replay. With #1272 merged, this press class is now the 100% residual for Android list flows that include a row press — the single next identity gap.

[thymikee]

Follow-up refinement surfaced by the #1286 implementation (its design flag 3), recorded here so it survives: the identity-empty check's text probe uses extractNodeText, which falls back to a raw (un-demoted) node.identifier — so a container that carries its own NON-UNIQUE id directly (rather than being id-less like the measured Settings rows) reads as "having text" and skips descendant retargeting entirely. Narrow gap, not present in the measured repro shape, but the eventual refinement is for every identity-emptiness/selectivity judgment to consume the #1269-demoted identity view consistently, extractNodeText's identifier fallback included. Not blocking #1286.