IMSN

The master register currently contains 120 event rows. Of those, 115 completed SuperQuick, 12 were promoted and completed QUICK4, 2 have completed DEEP4, and 1 has completed DEEP16. I think the scientifically sensible interpretation of “each event” here is therefore the 12 events that have a full QUICK4-or-deeper waveform analysis; the other 108 are still screening-only rather than comparable full analyses. The table below uses the current shared register plus the individual V30 reports.

Current V30 analysed-event table

Candidate basin is diagnostic only — it is not a claimed measured ringdown frequency. “Structure” and “Info” are the scores from the furthest completed stage, so DEEP values can be stricter than QUICK values.

Event	SNR	Furthest stage	Candidate basin*	Structure / Info	Evolution	Two-decay	Detector support	Current V30 position
GW231226_101520	34.7	DEEP16	166.9 Hz	66 / 58	CANDIDATE	SUPPORTED	H1 + L1	4k→16k SURVIVES
GW250114_082203	78.6	DEEP4	224.6 Hz	56 / 64	SUPPORTED	CANDIDATE	H1 + L1	DEEP16 next
GW240615_113620	26.4	QUICK4	176.5 Hz	70 / 45	SUPPORTED	SUPPORTED	H1 + L1; V1 no	High-priority DEEP4
GW240621_195059	28.2	QUICK4	142.8 Hz	69 / 47	CANDIDATE	NOT SUPPORTED	H1 only; L1/V1 no	Important DEEP4 stress test
GW230927_153832	20.3	QUICK4	386.1 Hz	49 / 44	CANDIDATE	CANDIDATE	H1 + L1	DEEP4
GW241102_124058	21.7	QUICK4	435.0 Hz	25 / 45	CANDIDATE	CANDIDATE	H1 + L1; V1 no	DEEP4
GW190521_074359	25.9	QUICK4	55.7 Hz	24 / 53	CANDIDATE	NOT SUPPORTED	H1 + L1	DEEP4
GW231123_135430	21.8	QUICK4	474.8 Hz	8 / 49	NOT SUPPORTED	NOT SUPPORTED	H1 + L1	DEEP4 for stable high-f structure
GW240920_124024	37.4	QUICK4	516.8 Hz	24 / 41	NOT SUPPORTED	NOT SUPPORTED	H1 only; L1 non-informative	Low-interest/control DEEP4
GW231028_153006	22.4	QUICK4	408.7 Hz	14 / 38	CANDIDATE	NOT SUPPORTED	H1 only; L1 no	Lower-priority DEEP4
GW230627_015337	28.7	QUICK4	438.3 Hz	8 / 46	CANDIDATE	NOT SUPPORTED	L1 strong; H1 weak	Lower-priority DEEP4
GW231206_233901	21.9	QUICK4	165.2 Hz	8 / 38	NOT SUPPORTED	NOT SUPPORTED	H1 only; L1 no	SCREEN ONLY / control

Event-by-event assessment

GW231226_101520 — strongest validated waveform survivor so far.
This is currently the cleanest demonstration that the staged V30 architecture works. DEEP4 found a recurrent damped structure near 170 Hz and supported early two-decay morphology; DEEP16 then recovered essentially the same behaviour from data sampled four times more finely. At 16 kHz the network candidate is 166.9 Hz, the independent free damped-mode search is again 170.0 Hz with $\tau=7.86$ ms and seven recurrent starts, and the two-decay result remains SUPPORTED. V30 therefore automatically records resolution_survival_status = SURVIVES. The precise stationary frequency still fails the measurement gate because H1 and L1 prefer somewhat different positions within the basin, but the waveform morphology itself is now convincingly resolution-stable.

GW250114_082203 — strongest evolving waveform candidate.
This is the most important contrast with GW231226. DEEP4 identifies the primary channel as EARLY EVOLVING, with $\Delta\mathrm{BIC}=48.0$, both H1 and L1 supporting the behaviour, and four recurrent windows. Later in the waveform the fits settle towards roughly 220–225 Hz: the network diagnostic is 224.6 Hz, the publication-style free damped mode is 217.5 Hz with $\tau\sim5.1$ ms, and the late-settling track reaches 225.8 Hz. The two-decay result remains only CANDIDATE. This looks less like “one ringdown note” and more like strong early evolution followed by a later settling regime. DEEP16 is now the obvious test: if that evolving→settling morphology survives increased resolution, GW250114 and GW231226 may represent genuinely different waveform families.

GW240615_113620 — probably the most interesting event still waiting for DEEP4.
QUICK4 gives this event a high 70/100 structure score. More importantly, both the evolving channel and the early two-decay channel are already SUPPORTED. The evolving fit has $\Delta\mathrm{BIC}=26.3$, is supported by two of the three detectors and recurs in three windows. H1 and L1 agree reasonably around the ~176 Hz region, whereas Virgo does not support it; the leave-one-detector-out test nevertheless passes with only a 7.5% maximum shift. An independent free damped-mode analysis lands at 175 Hz, very close to the 176.5-Hz network basin, and recurs at three starts. This is exactly the kind of promising QUICK4 result where DEEP4 should determine whether the apparent complexity is robust or simply a successful screening fit.

GW240621_195059 — high structure, but deliberately suspicious.
The event scores a high 69/100 for waveform structure, has strong event localisation and produces a very large apparent evolving-model gain: the early/evolving test reaches $\Delta\mathrm{BIC}=98$. The problem is that only one of three detectors supports that evolving solution, and the joint candidate around 142.8 Hz is effectively H1-dominated; removing H1 moves the result by nearly 24%. There is also one residual follow-up feature, unlike most events in the set. The independent free-mode search instead prefers roughly 165 Hz. This makes GW240621 extremely useful not because we should believe the strongest fit, but because it is a test of V30’s ability to recognise strong but detector-inconsistent post-merger structure without over-interpreting it.

GW230927_153832 — clearly complex, but with competing frequency families.
QUICK4 finds a substantial event-localised post-merger signal, with blind $\Delta\mathrm{BIC}=48.7$ and localisation $\Delta\mathrm{BIC}=57.2$. The main stationary diagnostic is around 386 Hz and both detectors support the local joint basin, but their independent strongest peaks are dramatically different. The analysis also preserves a strong short-lived competitor around 187 Hz, while the independent free damped-mode search lands near 185 Hz with three recurrent starts. That correspondence is notable. The evolving channel itself is only CANDIDATE because it recurs once. My interpretation is that this event may contain more than one useful timescale/frequency regime, making it a good DEEP4 target rather than a candidate for a single-frequency interpretation.

GW241102_124058 — excellent local frequency agreement, but weak overall evidence.
This event is interesting because H1 and L1 agree very well around the ~435-Hz basin: the reported joint spread is only 1.5%, and the three-detector jackknife is exceptionally stable at 0.9%. Virgo, however, does not support the structure. The broad blind search is only modest, $\Delta\mathrm{BIC}=5.7$, the evolving channel is merely CANDIDATE and has only one supporting detector, while the two-decay test is also CANDIDATE. The damping fit piles up against an upper search bound, which is another reason not to interpret it physically yet. DEEP4 is worthwhile mainly to discover whether the very good H1/L1 frequency consistency survives stronger null tests.

GW190521_074359 — one of the cleanest simple-frequency candidates.
This event is different from the high-complexity cases. QUICK4 finds a very low-frequency basin around 55.7 Hz, activated over six starts, with H1 and L1 reasonably consistent and a joint spread of 9.5%. The localisation statistic is strong, but waveform-structure score is only 24/100: two-decay is not supported and the apparent evolution is only a CANDIDATE because it is driven primarily by one detector and one recurrent window. That makes GW190521 useful as a potential comparatively simple, predominantly stationary post-merger waveform. DEEP4 can establish whether the stable ~56-Hz family is genuinely recoverable without needing to invoke extra morphology.

GW231123_135430 — unusual stable high-frequency structure rather than complex morphology.
Its 8/100 structure score looks uninteresting at first, but this event has one striking feature: H1 and L1 independently agree very closely around 475 Hz, with only 1.3% joint-basin spread. The free damped-mode analysis also finds 475 Hz and it persists through five start times. Yet the broad blind statistic is actually poor, $\Delta\mathrm{BIC}=-5.9$, and neither evolution nor two-decay structure is supported. This discrepancy is exactly why DEEP4 is justified: it needs to decide whether the stable ~475-Hz feature is genuine event-centred morphology, a narrow instrumental structure, or a peculiarity of that fitting model. It is an unusual event, but not because it looks complicated.

GW240920_124024 — useful negative/control case.
Despite a fairly high catalogue SNR of 37.4, V30 does not recover convincing post-merger complexity. The blind ringdown gain is only $\Delta\mathrm{BIC}=0.4$ with an empirical null p of 0.294, evolution is NOT SUPPORTED, two-decay is NOT SUPPORTED, and only H1 meaningfully supports the selected high-frequency basin. The publication-style free-QNM result is also WEAK. Although the automatic register still sends it to DEEP4, scientifically I would regard it mainly as a control event: if DEEP4 continues to find little, that is valuable evidence that V30 isn’t automatically producing elaborate morphology from every loud event.

GW231028_153006 — H1-dominated high-frequency structure.
The ~409-Hz network candidate is supported strongly by H1 but essentially not by L1, producing a huge 77.5% detector disagreement. The free damped-mode search independently lands near 415 Hz, which is why the event remains worth investigating, but it recurs only once. Evolution is also just a CANDIDATE, supported by one detector with no neighbouring-window recurrence, and the two-decay test is negative. I therefore regard this as interesting but fragile high-frequency structure: DEEP4 should test it, but it should rank below events where the morphology is visibly network-supported.

GW230627_015337 — weak network support despite a coherent-looking frequency region.
QUICK4 identifies a diagnostic basin near 438 Hz, but L1 provides most of the actual evidence; H1 supplies only weak support. The early/evolving test reaches $\Delta\mathrm{BIC}=12.3$ but is supported by only one detector and has no recurrent windows, while two-decay is not supported. A separate free damped-mode search finds a short-lived candidate near 495 Hz, again with only one start. The event is worth DEEP4 because there is event-localised transient structure, but its present fingerprint is too detector-dependent and non-recurrent to rank highly.

GW231206_233901 — best retained as a low-priority comparison waveform.
This is the only one of the 12 that V30 itself does not promote to DEEP4. The diagnostic basin is around 165 Hz, but H1 supports it while L1 does not, giving a 31.3% detector spread. Evolution and two-decay structure are both NOT SUPPORTED, the overall waveform-structure score is only 8/100 and the broad off-source test is not especially compelling. There is an independent free damped-mode candidate near 175 Hz, but the overall network evidence is insufficient. Its real value is as a control population member against which the more complicated events can be compared.

What the collection is starting to show

The most interesting result is that the 12 events do not collapse into one generic waveform pattern.

At the moment I would divide them roughly into:

Resolution-validated complex: GW231226
Strong evolving: GW250114
Strong evolving + multi-decay candidate: GW240615
Strong but detector-inconsistent / possibly contaminated: GW240621
Competing-regime morphology: GW230927
Mostly stationary/simple: GW190521
Stable narrow high-frequency structure: GW231123
Weak or detector-dependent candidates: GW241102, GW231028, GW230627
Controls / little extra structure: GW240920, GW231206

That is a much more useful league table than ranking them simply by SNR. In fact, the highest-SNR event, GW250114, and the 34.7-SNR GW231226 already have quite different post-merger fingerprints, while several ~20–30 SNR events show their own distinctive behaviour.

My next DEEP4 priority order from the Drive as it stands would be:

GW240615 → GW240621 → GW230927 → GW190521 → GW231123 → GW241102 → GW231028 → GW230627 → GW240920, while GW250114 should go directly to DEEP16 and GW231206 stays as a control.

The V30 architecture is now starting to do what we intended: use the low-resolution survey to discover waveform families, then spend the expensive high-resolution analysis only on events whose structure survives increasingly severe tests.