[on8st]

Not covered: Icom and the Demo. Both send dB rows down the same path (IcomCivBackend scope rows through toDbm, SimSignalSource at a -120 floor) but declare no binsAbsolute, so on either of them Off still draws every row black. The maintainer ruled on 2026-10-03 to merge this as it stands, correct for HL2, ANAN and RTL-SDR. A row-unit signal emitted where onBackendSpectrumFrame sends the row on comes as a separate follow-up PR. radio-capabilities-map.md states the same gap.

Second update, 700c2d7e: current main is merged in; the docs now call binsAbsolute a proxy for the row unit and list the Demo as not covered (Icom added in a3758582). The merged tree is 614 of 616 on macOS: the two failing tests fail identically on clean main and are not from this PR.

Summary

Refs #5678 (the HL2 parity rows). No Fixes line: I found no issue for this.

Update after review, a54dff9c. Two more things to read first, and one rename:

• The shipped default (Black Level 15) is still dark on these radios. It is a black point of −75; the HL2 floors first light at slider 52 to 98. Retest with the slider above that. A Flex at 15 is in the same position (threshold 145 against a 96..120 tile).
• The rows are dBFS under a dBm label on all three families. No RTL-SDR bin goes below −120 dBFS (RtlSdrDdc.cpp:179), so slider 0 (−60) cannot black an RTL signal stronger than −60 dBFS. The end points are unchanged; the options are in the review reply.
• The API now says rowsAreAbsoluteDb. Full ctest at a54dff9c: 611 of 611 (macOS, Qt 6.12.0, Ninja).

Read first.

• The fix has not been seen in the running app. The defect was observed there. The fix is tested as arithmetic only.
• Two constants are chosen, not derived, and they are yours to move: the manual black point on a dBm row runs from -60 dBm (slider 0) to -160 dBm (slider 100).
• It reaches three families. The switch is RadioCapabilities::panBinsAbsolute(), which HL2, ANAN and RTL-SDR declare. Only the HL2 was observed.
• The test ends in a source-text pin on SpectrumWidget.cpp. It names its claim; if you would rather not carry it, it is four checks at the end of one file.

With the Display panel's Black Level button on Off (manual), a Hermes-Lite 2 draws every new waterfall row black, wherever the slider sits.

Observed (the defect, before the change)

Hermes-Lite 2, gateware 74, receive only, through the automation bridge, 2026-10-01, on a local integration build over 25da98de (Qt 6.12.0). None of the PRs in that build touched this function.

Black Level button	mean luma of the newest 130 rows
SW, WtrFall Gain 0 / 50 / 100	5.6 to 5.8 / 12.7 to 13.8 / 43.5 to 43.6
SW, offset 20 / 50 / 80	2.6 / 14.8 / 61.4
Off, Black Level 0, 25, 50, 75, 100, WtrFall Gain 0, 50, 100	0.00 in 12 of 12 frames, max 0.0, 653 856 px each

get display read back each wfBlackLevel and wfAutoBlack: false, so the slider reached the widget.

Cause

The manual black point in SpectrumWidget::intensityToWaterfallLevel was 160 - level: 160 down to 60, in Flex tile-intensity units (a tile is int16(raw) / 128, about 96..120 on HF).

A radio whose spectrum is computed on this host has no waterfall plane. RadioModel::onBackendSpectrumFrame hands the pan frame on as the waterfall row, so the row is dBm: negative. Every sample is below a threshold of +60..+160. SW is unaffected because its black point is measured from the row, so it is already in the row's unit.

The change

• The value-to-level law moves from SpectrumWidget into gui/WaterfallLevelMap.h as pure functions, because SpectrumWidget links into no test target. The widget now gathers its state and calls it.
• The manual branch takes the row's unit from m_panBinsAbsolute, the flag the widget already holds for the auto-floor gate:
  • false (Flex): 160 - level. Unchanged.
  • true (HL2, ANAN, RTL-SDR): -60 dBm - level, one dB a step.
• SW and HW do not read the flag. No capability is added and no family is named.
• docs/HERMES.md and docs/architecture/radio-capabilities-map.md record the flag's second consumer.

Design choice, yours to move: the two end points. The span has to reach below the lowest floor this path produces and leave only strong signals lit at slider 0, as on a Flex. On this HL2 on a dummy load the displayed floor ran from -111 dBm (LNA -12 dB) to -148 dBm (LNA +40 dB) at the 384 kHz span; the 48 kHz span has bins eight times narrower, 9 dB lower. -160 dBm clears that. These figures come from one radio whose dBm axis is uncalibrated; I have none for an ANAN or an RTL-SDR. Both constants are in WaterfallLevelMap.h (kManualBlackAtZeroAbsoluteDb, and the 0..100 slider at one dB a step).

Direction is the Flex one in both units: a higher slider value is a lower black point ("Decrease to darken the noise floor", the slider's tooltip). WtrFall Gain keeps its range law (120 - gain * 0.91, in the row's unit), so Off and SW at the same black point draw the same picture.

Tests

waterfall_manual_black_test (new, no Qt, 16 checks):

• A Flex is unchanged. A frozen copy of the law as it stood is compared with the extracted one bit for bit over 8 604 792 samples: Off, SW and HW, every 5th value of gain and of the slider, three radio black levels, values from -200 to +600, NaN and both infinities. Zero mismatches with the flag false; zero for SW and HW with it true. (The header keeps qBound's comparison order, which sends a NaN sample to black; std::clamp would not.)
• The tile law blanks a dBm row at every slider position, also for a signal 60 dB over the floor.
• On a dBm row, for seven measured floors and gain 0 / 50 / 100: slider 100 lights the floor, slider 0 blacks it, some position blacks the floor and keeps a signal 30 dB above it, and raising the slider never darkens.
• Off at a black point equals SW at the same black point.
• Source-text pin (four checks). The claim: SpectrumWidget.cpp passes m_panBinsAbsolute as rowsAreAbsoluteDb, calls the shared law, and keeps no second copy of 160 - level. No behavioural seam reaches it: SpectrumWidget is compiled only into the application, and no test target links it. It shows how the call is written, not that the widget runs. That HL2 declares panBinsAbsolute() and a Flex does not is pinned on real backends by noise_floor_auto_adjust_gate_test.

Red before green, re-run at this head. On main the law is a private member of a widget no test links, so the test cannot be built against main; M1 is main's law inside the new header.

mutation	failing checks of 16
M1: the manual branch ignores the unit (main's law)	4
M2: the widget passes false	1 (the source-text pin)
M3: the tile constant 160 becomes 150	2
M4: the dBm slider direction reversed	5
M5: the unit leaks into SW	2

Mutations were run with the test compiled on its own with Apple clang.

Full ctest on macOS (Qt 6.12.0, Ninja) at 67999c5c, which contains main at c7513e12: 100% tests passed out of 611.

Every script step of the Static checks job in static-checks.yml was run locally with stdin from /dev/null and is clean, the colour ratchet at +0. check_a11y on the two changed GUI files: no findings.

Not verified

• Not observed in the running app. The observation above is the defect, on a build without this change. No build carrying the fix has been launched against a radio.
• ANAN and RTL-SDR take the same path; nothing was run on them.
• The source-text pin does not see MainWindow pushing the capability into the widget.
• macOS only. The sanitizer-configure job was not run locally.

Beside this on main

Nothing on main overlaps: the branch merges c7513e12 cleanly. The NB Blank fix for the same unit problem (branch hl2/waterfall-nb-blank-dbm) reads the same flag and merges cleanly with this one. The keyed (TX) waterfall rows go through dbmToWaterfallLevel, which has its own black-level law and is untouched.

73, ON8ST

🤖 Generated with Claude Code

[aethersdr-agent]

1. Issue fit

Yes, for the radio it was measured on. There is no linked issue (the body says so, and Refs #5678); GOVERNANCE.md does not require one for a bug fix with a clear root cause, and this one has a root cause I was able to re-derive independently rather than take on trust: RadioModel::onBackendSpectrumFrame emits the same bins array as both the pan frame and the waterfall row (src/models/RadioModel.cpp:6752, "the backend has no separate waterfall plane, so the pan row is the waterfall row"), that row lands on the native-tile path, and the manual branch there compared it against a tile-intensity threshold of +60..+160. Every sample is below that. The diagnosis is correct and the fix is aimed at the right line.

Two things the fix does not reach, both arithmetic rather than opinion, both detailed below: the shipped default slider position still draws an HL2 waterfall black, and the two chosen end points do not fit RTL-SDR's scale, which is hard-bounded to roughly −120…−6 dBFS by its own FFT code. Neither is a correctness defect in the law; both are the constants question the body explicitly hands to the maintainer, and I have numbers for it now.

2. Scope

File / group	What it changes	Claimed?	Verdict
src/gui/WaterfallLevelMap.h (new)	the value→level law as pure functions, plus the two end-point constants	yes	In scope
src/gui/SpectrumWidget.cpp	intensityToWaterfallLevel gathers state and delegates; wfHighThresholdRaw removed	yes	In scope — wfHighThresholdRaw had exactly one caller, so the move leaves nothing behind, and the bit-identity test covers the move
tests/waterfall_manual_black_test.cpp + tests/tests.cmake	registered target, add_test, AETHER_SOURCE_DIR	yes	In scope. Not a socket test — no QTcp*/QUdp*/QLocal*/bind/listen/fake peer; the only I/O is one std::ifstream over a source file. AETHER_SOURCE_DIR has ~8 precedents in tests.cmake
docs/HERMES.md, docs/architecture/radio-capabilities-map.md	record the flag's second consumer	yes	In scope, and the right place

No CHANGELOG.md entry — correct. No capability added, no family named, no settings key added or migrated, no new protocol or public surface, no dialog or a11y surface, no formatting churn, no stray commit. Everything in the diff is explained by the stated fix. The move of the law out of the widget is not scope creep: SpectrumWidget links into no test target, so extraction is the only way the arithmetic becomes testable at all, and the 8.6 M-sample bit-identity check is exactly the right price for it.

3. Blockers

None.

4. Nits (all non-blocking)

• The shipped default still draws black on an HL2 — the practical headline. Default DisplayWfBlackLevel is 15 (SpectrumWidget.cpp:2501) → black point −75 dBm. Against your own seven measured floors the level is 0.000 at every colour gain, and so is slider 50. The first position that lights the floor runs 52 (LNA −12 dB) → 89 (LNA +40 dB) → 98 (48 kHz span). So an operator who flips Black Level to Off without also dragging the slider past halfway still sees the reported symptom. This is consistent with the Flex, where the manual default of 15 → threshold 145 also sits above a 96..120 tile row, so it is not a regression — but it is worth a line in the body so nobody retests at the default and concludes the fix did not take.
• dBm vs dBFS naming. RadioModel.cpp:6661 calls the HL2 payload "DC-centred dBFS bins", and RtlSdrBackend.cpp:105 is explicit that "the axis is dBFS relative to the converter's own full scale." The new rowsAreDbm / kManualBlackAtZeroDbm and both doc paragraphs say dBm. The arithmetic is unaffected, but one of the two comments is stale and this is the name on the new API.
• Pre-existing, same bug class, not yours to fix here: m_autoBlackThresh{145.0f} (SpectrumWidget.h:1706) is also a tile-unit seed. The estimator at SpectrumWidget.cpp:8538 smooths at 0.05, so on a dBm row it needs ~110 rows to walk from +145 down to ~−140 — i.e. SW mode also starts black for several seconds after connect on these three families. Your "SW measures its black point from the row" claim is correct in steady state (I read the estimator at 8515-8538; it genuinely reads binsIntensity), just not at the seed. Worth adding to the follow-up list beside the NB Blank branch.
• Two dBm black-point laws now coexist on one radio. If HL2 native rows stall past waterfallFallbackTimeoutMs(), rows switch to pushRxWaterfallFallbackIfDue → dbmToWaterfallLevel (SpectrumWidget.cpp:12370), whose manual black point is m_wfMinDbm + (125 − level) * 0.4. Same radio, same unit, same slider, a different black point. Pre-existing divergence, but this PR is the first to make both live laws dBm laws, so it becomes visible. (FWIW m_wfMinDbm is never assigned anywhere in src/ — frozen at its −130 default — which is why I am not suggesting it as an anchor below.)
• The source-text pin is, strictly, asserting the implementation back to itself. I would keep it: it is disclosed, it is precedented, and its one real blind spot — MainWindow actually pushing the capability — I checked by hand and it is covered (see below).

5. What I tried to break

Findings below are reasoned from code plus my own local compile-and-run of the new test; nothing was observed in the running app, as you also state.

• I built and ran the test. macOS, Apple clang, -std=c++20, header-only: 16/16 checks, 0 failed, 8 604 792 samples compared — exactly the figures in your body.
• I re-ran your mutation table independently and it reproduces precisely: M1 (manual branch ignores the unit) → 4 failures, M3 (160→150) → 2, M4 (direction reversed) → 5, M5 (unit leaks into SW) → 2. I added two of my own: swapping bound() for std::clamp → 2 failures, so the NaN-ordering argument in the header comment is real and load-bearing, not decorative; and moving kManualBlackAtZeroDbm −60 → −100 → only 2 failures, both of them the constant-echo and SW-equality checks. So the behavioural checks (lit end, dark end, +30 dB separation) are permissive over a wide band of end points — which is honest given the body calls the constants chosen, but it means the test will not object if the maintainer moves them.
• Both capability push sites check out. MainWindow.cpp:7558-7566 pushes to every SpectrumWidget child on connect and disconnect (connected && caps.panBinsAbsolute(), so it resets to false on disconnect), and MainWindow_Wiring.cpp:3912 covers the case I expected to be the hole — a pane added after connect via Add Panadapter or a layout change. It is already there with a comment explaining why. Hl2Backend.cpp:1767, AnanBackend.cpp:386 and RtlSdrBackend.cpp:117 all set binsAbsolute = true, so the three-family claim is accurate.
• I looked for a second copy of the law, including a GPU/QRhi shader path, since render-path parity is a recurring bug class here. The only other level laws are dbmToWaterfallLevel and kiwiSdrWaterfallLevel; both are untouched and reached from different call sites, and waterfallLevelToRgb/encodeWaterfallLevel are unit-agnostic. Nothing left behind.
• Head freshness and CI. The task checkout was 67999c5c; I fetched pull/6115/head and confirmed it is the same commit and that f5f2b14 is an ancestor, so I reviewed the real head. All five checks are green on it. Linux (ci.yml:282) and macOS (:1147) build full-tree, so the new test compiles under GCC too — but Windows builds only a named target subset, so MSVC never sees this file, and no PR-gating ctest runs this test (every ctest step in ci.yml is -R-filtered to a handful of names). Green CI here does not mean these 16 checks ran; your local full-suite run and mine are the evidence.
• Could not check: anything requiring the running app, an ANAN, or an RTL dongle — same boundary you declared.

6. Recommendation

Approve with nits, plus one maintainer decision you already asked for. The law itself is right and unusually well pinned: a Flex is bit-identical over 8.6 M samples across all three black-point sources including NaN and both infinities, every mutation I could invent is caught, the extraction is necessary rather than cosmetic, and the scope is clean. I tried to break the arithmetic, the capability plumbing, and the sibling render paths, and it held.

The one thing worth the maintainer's attention before merge is the question the body raises: the two end points are tuned to one HL2 and the test only validates them against that radio's floors, while the same switch serves a family whose scale cannot reach them. That is not an argument against merging — it is an argument for either narrowing the switch to the HL2/ANAN path for now, or anchoring the black point to the pan's live display range (m_refLevel − m_dynamicRange) so it self-calibrates and no constant has to be guessed per family. The HL2's auto-floor loop is already driven by this very same panBinsAbsolute flag, so that range is maintained and sits on the noise floor. Concrete next step: pick one of those two, and add the one-line note about the default slider so the fix is retested off the default.

Nice work — the mutation table is accurate to the check, which is rarer than it should be, and the "Not verified" section is exactly the right shape.

---

_{🤖 aethersdr-agent · cost: $9.7718 · model: claude-opus-5}

[on8st]

Bad news first: the default is still dark, and the end points are not changed.

Default. Slider 15 is a black point of −75. The seven HL2 floors read 0.000 at every gain; first lit at 52 … 98. A Flex at 15 (threshold 145, tile 96..120) is also 0.000; first lit at 41 … 65. The default follows the end-point choice, so it is yours.

Two corrections.

• The pan range tracks the floor only with the auto floor on, and that defaults off (SpectrumWidget.cpp:2527).
• Your own table puts both bands inside one span: RTL floors light at 11 … 61, HL2's at 52 … 98. The RTL gap is the other end: nothing above −60 dBFS can be blacked.

option	cost
keep −60 … −160	RTL signals above −60 always lit
narrow to HL2/ANAN	new capability field; RTL stays black in Off; ANAN is unmeasured too (AnanBackend.cpp:378)
anchor to the pan range	a scale move recolours new rows only; fixed while auto floor is off
end points in PanAmplitudeModel	a record field, three backends; no ANAN figure
anchor to the row's measured floor	Off becomes SW with a second slider

Done in a54dff9c: the API says rowsAreAbsoluteDb (dBFS under a dBm label), and the test states its floors are one HL2's.

The SW seed and the fallback law are follow-ups. Agreed: green CI did not run this test.

73, ON8ST

[ten9876]

Second adversarial pass at a54dff9c

This follows the bot review on 67999c5c and the author's replies, and doesn't repeat them. It covers four things that pass didn't: the fix running in the app, a fourth producer of dB rows, the sibling #6117, and settings meaning.

Issue fit

Yes, for HL2, ANAN and RTL-SDR. I have now seen the fix work in the running app, which the body lists as its biggest unverified claim. The diagnosis is right: RadioModel::onBackendSpectrumFrame (src/models/RadioModel.cpp:6659-6750) sends the pan frame on as the waterfall row, so the row is in dB and the tile threshold 160 - level blanks it. But the predicate the PR uses to detect a dB row, panBinsAbsolute(), misses a fourth producer that reaches the same function: the built-in Demo. On the Demo, Off is still black at every slider position on this head (evidence below).

Scope

File / group	What it changes	Claimed?	Verdict
src/gui/WaterfallLevelMap.h (new)	value-to-level law as pure functions, two end-point constants	yes	In scope
src/gui/SpectrumWidget.cpp	intensityToWaterfallLevel delegates; wfHighThresholdRaw moved	yes	In scope
tests/waterfall_manual_black_test.cpp, tests/tests.cmake	new registered no-Qt test	yes	In scope. Not a socket test (only I/O is one std::ifstream over a source file), so preflight is clean
docs/HERMES.md, docs/architecture/radio-capabilities-map.md	record the flag's second consumer	yes	In scope (see nit 3 on wording)

Everything in the diff is explained by the stated fix. No CHANGELOG entry, no new settings key, no capability field, no protocol surface.

Blockers

None.

Should-fix (non-blocking; maintainer call on the shape)

1. The Demo makes dB rows but doesn't declare binsAbsolute, so the fix can't reach it. The predicate is a proxy, not the row's unit.

• src/core/backends/sim/SimSignalSource.cpp:136-151 emits host-computed float rows with kFloorDbm = -120.0.
• SimBackend.cpp:47-51 relays them as spectrumFrameReady, so they go through the same onBackendSpectrumFrame → panFeedWaterfallRowReady path as the HL2.
• SimBackend::capabilities() (SimBackend.cpp:248-379) declares no panAmplitude record, so panBinsAbsolute() is false and the manual branch keeps 160 - level on a dB row.

Bridge evidence, PR head, Demo DEMO-0001. The run was offscreen with isolated settings and a -DAETHER_GPU_SPECTRUM=OFF build, so the widget's pixels can be grabbed offscreen. Each value is the mean luma of the newest ~1.3 s of waterfall rows, cropped clear of the Display panel. get display read back wfAutoBlack:false and each wfBlackLevel:

state	head as posted	head + Demo declares binsAbsolute=true (scratch mutation, not PR code)
SW, offset 50	4.04 (lit)	4.04
Off, 0	0.00	0.10 (max 59: the carrier only)
Off, 15 (default)	0.00	0.41
Off, 50	0.00	12.4
Off, 75	0.00	145.3
Off, 100	0.00	203.7

floors read noiseFloorDbm: -103.97 on both runs. So the reported defect reproduces on the Demo at this head. With the flag declared, the PR's law behaves as designed on a dB row:

• It rises monotonically with the slider.
• The floor first lights near slider 44, from the threshold -60 - L against a -104 floor.
• Slider 0 leaves only the carrier.

That second column is the first in-app sighting of this fix. Under the PR's own law it also matches what the HL2 should do.

PanAmplitudeModel::binsAbsolute is documented (RadioCapabilities.h:114-122) as "a bin does not move when the display reference level moves". That property is about the auto floor. The PR now also reads it as "the waterfall row is in dB". Those coincide on HL2, ANAN and RTL, but not on the Demo, and nothing keeps them coinciding for the next backend that routes through onBackendSpectrumFrame. The new paragraph in radio-capabilities-map.md:541 says "A backend with absolute bins has no waterfall plane of its own", but the inference the code needs runs the other way: no waterfall plane, so the row is the pan frame, so the row is in dB. Two options:

• (a) Make the Demo declare binsAbsolute = true. That is truthful under the field's own definition, since its bins are host-computed and fixed against the reference level. It also turns on the Demo's auto-floor loop, which is a behaviour change, so it is better as its own small PR.
• (b) Key the unit off where the row came from: rows that arrive via panFeedWaterfallRowReady from onBackendSpectrumFrame are pan frames by construction. That is closer to the real invariant, but it is a larger change.

Either way, the PR shouldn't read as complete while the one backend every reviewer and new user drives still shows the reported symptom. At minimum, the body and the capabilities-map paragraph should say the Demo is not covered.

Sibling #6117 (NB Blank on a dB row)

• No conflict. git merge-tree of a54dff9c with 63a73a35 is clean, and so is a54dff9c with current main (579efd73f). The hunks don't overlap: updateWaterfallRow's blanker versus intensityToWaterfallLevel. The tests.cmake blocks land in different places.
• Same root cause, treated consistently: both take the row's unit from m_panBinsAbsolute, never from the sign of the data, and both treat 1 dB as one tile unit of colour-range width. So both inherit the Demo gap in item 1. Whatever is decided there should land once and cover both.
• Naming drift: this PR renamed to rowsAreAbsoluteDb / "dBFS under a dBm label" after review. #6117 still says RowKind::Dbm and "dBm row". Worth aligning before either merges, so the codebase doesn't carry two names for one concept.
• Neither PR touches the third consumer of the same unit mismatch, the SW seed m_autoBlackThresh{145.0f} (SpectrumWidget.h:1706). I saw it in the Demo run: SW drew roughly the first 8 s of rows black before the estimator walked down from +145 (visible in the SW grab). The bot and the author already listed this as a follow-up, and I agree.

Settings meaning and migration

DisplayWfBlackLevel keeps its key, its 0..100 clamp (SpectrumWidget.cpp:4259) and its default of 15. On a Flex the law is bit-identical: I re-ran waterfall_manual_black_test at this head on Linux/GCC and got 16/16 with 8,604,792 samples compared. On a flag-true radio the old meaning of any saved value was "always black", so there is no meaning to preserve, and no migration is needed. The key is per-pan but not per-radio, so one saved number now means a tile threshold on a Flex and a dB threshold on an HL2. That is inherent to the design and not a defect, but worth knowing when someone switches radios.

Nits (non-blocking)

1. CI is stale against main: runs were created at 18:34Z and main moved at 19:25Z (579efd73f). The trial merge is clean. No CI ctest -R filter runs waterfall_manual_black_test, so green CI does not mean these 16 checks ran.
2. Pre-existing and not yours: SpectrumWidget.h:1642 still comments m_wfBlackLevel as "0-125" while the setter clamps to 0..100.
3. docs/HERMES.md "Off needs the row's unit, and for a while it did not have it. … Off drew nothing at any slider position — 12 of 12 frames exactly black on the bench" narrates the bug's history. The capabilities-map paragraph's "Before that the tile threshold was applied…" does the same. These are design docs, so this is softer than the comments rule, but the current-invariant sentence would carry it alone and the bench story fits the PR body.

What I tried to break

• Built the head on Linux (GCC, Qt 6.12) and ran the new test: 16/16. I didn't redo the bot's M1–M5 mutation table. Its numbers already match the body.
• Drove the Demo through the bridge on my own socket, with isolated settings, offscreen, AETHER_AUTOMATION_NO_TX=1. I checked that get radio returned AetherSDR Demo / DEMO-0001. I toggled Black Level to Off via displayAutoBlackBtn, stepped displayBlackLevelSlider 0/15/50/75/100, read each back with get display, and grabbed pan-visible 0 each time. The table above is the result. The GPU build can't be grabbed offscreen ("Spectrum renderer unavailable"), so I rebuilt with AETHER_GPU_SPECTRUM=OFF. That changes the renderer, not the law, which both paths share via intensityToWaterfallLevel.
• A/B without a base build: with the flag false, the head's manual branch runs main's exact law (proven bit-identical by the test). So head-as-posted on the Demo is the "before", and the scratch flag flip is the "after". The mutation was reverted and is not part of anything pushed.
• Flex was not driven (no hardware in a review). Flex invariance rests on the bit-identity test, which I re-ran.
• Not tested: ANAN and RTL hardware. The bot's RTL-saturation table is consistent with what I saw at slider 100 on the Demo: a -104 floor reads 204/255 mean luma, close to white.

Recommendation

Approve with nits, plus one maintainer decision. The law is correct, Flex is provably unchanged, scope is clean, and the fix now has in-app evidence. Before merge, decide item 1, the Demo/predicate gap that #6117 shares:

• either a follow-up making SimBackend declare binsAbsolute,
• or a statement in both PR bodies that the Demo is not covered.

The end-point constants are still the open maintainer question from the earlier round.

[on8st]

Bad news first, two items.

1. You are right: the Demo is not covered, and this push only says so. Confirmed from the source, not run: SimSignalSource.cpp:137, SimBackend.cpp:47, and no panAmplitude record in SimBackend::capabilities().
2. The merged tree is not green here: 614 of 616 (macOS, Qt 6.12.0). hl2_pan_create_async_test (fix(hl2): drop spectrum and meter output from a closed receiver's DSP. Principle VIII. #6113) and peripheral_auth_dialog_test (feat(peripherals): authenticate direct TGXL, PGXL, and Antenna Genius connections #6008) fail, identically on clean main at d9de044c. a54dff9c was 611 of 611.

Pushed 700c2d7e, a merge of main (d9de044c) and one docs commit, no rebase:

• radio-capabilities-map.md states the inference your way round, calls binsAbsolute a proxy, and lists the Demo as not covered.
• Both doc paragraphs keep the invariant and drop the history (nit 3).
• CI re-runs on current main (nit 1).

Yours to pick: (a) or (b). I would take (a), as its own PR, naming the Demo's auto-floor change. I will not open it unasked. (b) is not sized.

#6117. git merge-tree against its head is still clean. It takes this PR's name, RowKind::AbsoluteDb.

Left. SpectrumWidget.h:1642: touching it pulls that header into the a11y scan (two unrelated warnings). The SW seed stays a follow-up. Nothing new on the end points.

On those two failures: the HL2 one fails here on clean main every time (60 of 60), and its cause is in the test, not in the guard; I posted it on #6113. The other one I have not traced beyond a layout that needs 680 px in a 640 px viewport on this Mac.

73, ON8ST

[ten9876]

Maintainer ruling on #6115 and #6117: merge both as they stand. They're correct for HL2, ANAN and RTL-SDR, and the bodies now say Icom and the demo aren't covered.

For the gap, I agree with you: option (a), a row-unit signal set where onBackendSpectrumFrame emits the row, as its own follow-up PR. If it changes the demo's auto-floor, name that in the PR. Please go ahead and build it.

73, Jeremy KK7GWY

[ten9876]

We're updating this branch against main and landing it (maintainer pass); please don't push to it meanwhile. New commits go on top, with no rebase or force-push.

[ten9876]

Maintainer pass. Answers to the two reviews' points (Jeremy's 5402532542 and aethersdr-agent's 5401871688, both body-only). The head is a3758582: on8st's 700c2d7e, plus signed merges of main up to 52b6f7ad (which includes #6117, no conflicts), plus one docs commit.

Should-fix 1: Demo and Icom, the predicate as proxy. Ruled 2026-10-03: merge as it stands, correct for HL2, ANAN and RTL-SDR. Option (a), a row-unit signal set where onBackendSpectrumFrame emits the row, comes as on8st's follow-up PR. a3758582 adds Icom (IcomCivBackend, scope rows through toDbm) next to the Demo in radio-capabilities-map.md's not-covered paragraph, and the PR body now opens with the same statement.

Sibling #6117. It merged as 52b6f7ad. The two paragraphs in radio-capabilities-map.md are reconciled in a3758582:

• fix(gui): NB Blank works on a dBm waterfall row. Principle VIII. #6117's paragraph names NB Blank.
• This PR's opens "The manual Black Level reads it as the row's unit too", where it used to say "second consumer".
• The not-covered paragraph now covers both readers.

The names agree: rowsAreAbsoluteDb here and RowKind::AbsoluteDb there.

End points and default. These are the maintainer's call, ruled "as they stand". No change.

Nits

1. Stale CI: fixed. CI re-runs on a3758582, which contains current main.
2. SpectrumWidget.h:1642 "0-125": not changed. It is pre-existing, and the ruling made nits optional. on8st's reason holds: touching that header pulls unrelated a11y warnings into this PR's scan.
3. History in the docs: already dropped by on8st in 700c2d7e. Verified.

The SW seed (m_autoBlackThresh{145}) and the fallback dbmToWaterfallLevel law stay as listed follow-ups. They are outside this fix.

Verification on Linux/GCC (Qt 6, RelWithDebInfo) at a3758582

• Full build is clean.
• waterfall_manual_black_test: 16 checks, 0 failed. Mutation M1 gives 4 failed, as the body states.
• waterfall_impulse_blanker_test, from fix(gui): NB Blank works on a dBm waterfall row. Principle VIII. #6117, passes in the merged tree.
• ctest excluding socket-owning tests: 100% of 481 passed. 124 tests were left out because they are socket-owning by name or by a grep of their sources.
• Every Static checks script exits 0. The colour ratchet is +0, and a11y reports 0 findings.
• The fix was not observed in the app in this pass. The in-app Demo evidence is in Jeremy's review: the flag flip makes Off light the floor.

Maintainer pass by Claude for @ten9876.

[ten9876]

Approving at a3758582 (maintainer pass, pr-land).

Maintainer ruling (2026-10-03): merge as it stands. It is correct for HL2, ANAN and RTL-SDR. Icom and the demo are not covered, and the body and radio-capabilities-map.md both say so. Option (a), a row-unit signal emitted from onBackendSpectrumFrame, follows as on8st's own PR. The end points (-60 to -160 dB) and the default of 15 stay as they are.

Review findings, all addressed:

• Naming (rowsAreAbsoluteDb): on8st, a54dff9c.
• Test-scope comments: on8st, a54dff9c.
• Docs state the invariant, not the history: on8st, 700c2d7e.
• Icom added to the not-covered paragraph, and the paragraphs reconciled with #6117's (now on main as 52b6f7ad): this pass, a3758582.
• SpectrumWidget.h:1642 comment nit: left as it is, since it is pre-existing and the ruling made nits optional.

All 4 threads were answered with evidence and resolved. Both body-only reviews were answered in a top-level comment.

Verified empirically (Linux/GCC, RelWithDebInfo, at a3758582):

• Full build is clean.
• waterfall_manual_black_test: 16 checks, 0 failed, and mutation M1 gives 4 failed.
• waterfall_impulse_blanker_test passes in the merged tree.
• ctest excluding socket-owning tests: 481 of 481 passed. 124 were excluded by name or by a source grep.
• All Static checks scripts exit 0. The colour ratchet is +0, and a11y reports 0 findings.

Reasoned, not observed: behaviour on HL2, ANAN and RTL hardware.

Freshness: the CI run on this head was created at 2026-10-03T23:53:29Z. main's tip 52b6f7ad dates from 23:46:21Z and is in the merged tree, so there is no drift.