Human-directed independent review (Claude / @NF0T)

Four improvements to the automation bridge. Code is sound — flagging for @ten9876 to approve due to AGENTS.md (Tier 1).

Claims independently verified against current codebase:

• EqualizerModel::BandCount, bandLabel(), rxBand(), txBand(), rxEnabled(), txEnabled() confirmed at src/models/EqualizerModel.h:18–42 ✓
• RadioModel::equalizerModel() at src/models/RadioModel.h:65 returns EqualizerModel&; &radio->equalizerModel() → const EqualizerModel* — correct C++ ✓
• All new SliceModel fields — apfLevel(), squelchOn(), squelchLevel(), agcMode(), agcThreshold() — confirmed at src/models/SliceModel.h:69–328 ✓
• QDoubleSpinBox already in use at line 687 of AutomationServer.cpp — no missing include ✓
• matchRecursive() (line 120) and matchByButtonText() (line 141) pre-exist; scoped lookup correctly calls both within the ancestor subtree; if scope string resolves no ancestor sc stays null and falls through to steps 1–N as documented ✓
• doInvoke disabled check fires before the TX-keying guard — correct ordering ✓
• Bridge and discover_socket confirmed in tools/automation_probe.py:33,45 — automation_validate.py import is valid ✓
• Squelch threshold MODEL exclusion correctly documented (mode-gated, not a desync) ✓

No blocking issues found. Tier 1 (AGENTS.md) — @ten9876 to approve.

Reviewed the full diff against the source tree. This is clean, well-scoped work — thanks @jensenpat.

Verified

• Model APIs all resolve. EqualizerModel::{rxBand,txBand,rxEnabled,txEnabled,bandLabel,BandCount}, RadioModel::equalizerModel(), and the new SliceModel::{apfLevel,squelchOn,squelchLevel,agcMode,agcThreshold} getters all exist, and RadioModel.h includes both SliceModel.h/EqualizerModel.h, so equalizerSnapshot() compiles against full types (not forward decls).
• range cast order is correct. QAbstractSlider → QSpinBox → QDoubleSpinBox is right, since QDoubleSpinBox does not derive from QSpinBox; QDial correctly inherits the slider branch.
• Scoped resolver matches scope by objectName/findChild/class/accessibleName and falls through to flat resolution when either half doesn't resolve, so a literal / in a control name still works (good call documenting that).
• Disabled refusal uses w->isEnabled(), which also returns false when an ancestor is disabled — that's the right semantics for "the radio won't accept this."
• Conventions honored (QStringLiteral, qobject_cast, no QSettings), Python imports (Bridge/discover_socket) and the tree["roots"] key check out, and all 6 CI checks are green.

Minor (non-blocking, optional)

• In doInvoke, the disabled check runs before the TX-safety guard, so a disabled TX-keying control returns disabled:true rather than blocked. Harmless — both refuse — but worth being aware of if any caller keys off the specific error.
• In automation_validate.py, classify_numeric computes mid = int((lo+hi)/2). For a QDoubleSpinBox with a sub-unit range (e.g. [0,1]) mid collapses to lo, making the mid-probe and model cross-check degenerate. None of the mapped MODEL controls are double-valued, so this is purely defensive — a round() or float mid would cover it if such a control is ever added.

LGTM as a COMMENT review. Nice that the validation tool encodes the wrapping/disabled/duplicate-name lessons directly rather than leaving them as tribal knowledge.

_{🤖 aethersdr-agent · cost: $2.8373 · model: claude-opus-4-8}

Approving (Tier 1 / automation-bridge area, per @NF0T's routing).

Verified the security-relevant points for a TX-reachable bridge:

• Disabled-refusal doesn't regress the TX guard. !w->isEnabled() runs before isTransmitControl(), so a disabled keying control returns disabled:true rather than blocked — both refuse; no path drives a keying control. It's a net safety win (silent no-op → assertable error), and isEnabled() correctly accounts for a disabled ancestor.
• Scoped targeting can't bypass guards. <scope>/<name> only changes resolution; the resolved widget still passes through doInvoke's disabled check and the aetherTxKeying marker guard. Falls through to flat matching so a literal / in a name still works.
• range / equalizer / slice additions are read-only (dumpTree/get) — no write path. range cast order (QAbstractSlider→QSpinBox→QDoubleSpinBox) is correct since QDoubleSpinBox doesn't derive from QSpinBox.
• automation_validate.py skips disabled + keying controls, and the server enforces both guards regardless — defense-in-depth.

Model APIs (EqualizerModel, new SliceModel getters) all resolve, conventions honored, CI green on all six. The two bot nits (disabled-vs-blocked error string for disabled TX controls; classify_numeric mid collapsing for sub-unit double ranges) are non-blocking and correctly noted as defensive. Nice that the QA sweep's wrapping/disabled/duplicate-name lessons are encoded in the tool rather than left tribal. Thanks @jensenpat, @NF0T.