windowsill lab · refuted lead

TIC 374861595

For ten weeks this was the lab’s open lead: a ten-percent dip every 1.94 days on a small red star. Then an astronomer explained why it is most likely two stars and not a planet.

Run this search yourself → The survey paper → The windowsill feed →

Depth 0 · no science needed

Start from zero

This page climbs from plain words to the full working file. Already know the field? Skip to the numbers or the dossier.

  1. A star keeps getting a little darker

    Far away there is a small star with a catalogue number instead of a name: TIC 374861595. Every 1.94 days, for about two hours, it loses about a tenth of its light. Then it comes back. It has done this hundreds of times, on schedule.

  2. Something is probably passing in front of it

    When an object circles a star and its path crosses our line of sight, it blocks a little light once per lap. Astronomers call that a transit. Most planets known today were found this way: not seen directly, but by the shadow they cast on their star’s brightness.

    Sketch of a transit: a dark disc clips the lower edge of a glowing star, and the brightness line below dips while it is in front. brightness the dip, ~10 % time →
    A sketch, not data. The real measurements are further down.
  3. The pictures are public

    NASA’s TESS satellite has photographed most of the sky since 2018, a month-long patch at a time, and puts every brightness measurement in a free public archive. NASA’s own software scans that archive automatically. It has flagged this dip 259 times. It has never moved the star onto the short list astronomers actually follow up, called the TOI list.

  4. Two home computers found it too, without being told where to look

    The windowsill lab is two computers in a house that search the same public archive around the clock. They were not pointed at this star. They found the dip on their own, in three separate months of data, and then spent their effort trying to kill it: checking it is on this star and not a neighbour, that it is not two stars eclipsing each other, that no glow from a second star shows up. It survived those checks. But the check for two stars looks for dips that take turns, one deep, one shallow. Two stars that are nearly twins make dips that nearly match. In the three months the lab had checked, the difference was too small to trip that check.

    See the trick for yourself: fold a known planet’s dips into one shape, then search stars nobody pointed you at.

  5. The lab never called it a planet

    A ten-percent dip means something big. Read as a planet, the object came out 2.6 to 3.9 times as wide as Jupiter, which is very large for a planet. The lab had an answer for that: the object seemed to clip only the edge of its star, and an edge-clip makes size very hard to pin down. So the lab said the size was not measured yet. Its mass was not measured at all, and mass is what separates a giant planet from a brown dwarf (an object too small to burn like a star) and from a small star. Settling that takes a large telescope watching the star wobble. For ten weeks the lab’s word for it was open lead: passed the checks, waiting on people and more data, no planet claimed.

  6. It is most likely two stars, not a planet

    On 2026-11-06 an astronomer who checks planet candidates from TESS read the lab’s public file on this star and posted a different explanation: two near-twin stars circling each other every 3.874 days. Each covers part of the other in turn, so the pair dims twice per lap, and the two dips look almost alike. The lab had counted them as one object passing twice as often, every 1.937 days. The astronomer pointed to two clues. A planet that big is hard to believe around a small red star. And the light also rises and falls slowly, once for every two dips. Both were already in the lab’s own file, explained away: the size as an effect of the edge-clip, the rise and fall as the star probably spinning.

    On 2026-11-10 the lab ran its own code on all 24 months of data it had, from 2020 to 2025, not just the three it had checked. In the 16 of those months that fall in 2023 to 2025, one slow rise and fall takes 3.87411 days, give or take 0.00039. Twice the gap between dips is 3.87390 days. The two kept step for 255 cycles, about 2.7 years. That is what two stars locked together by tides do, the way the Moon always keeps one face to the Earth. The lab’s own estimates also find one dip slightly deeper than the other, by 0.2 to 0.4 percentage points, though NASA’s comparison finds no difference. The lab’s word for this star is now refuted: most likely two stars, not a planet. Most likely is as far as it goes, because no mass has been measured.

  7. And there is no coin

    No token, coin or sale is connected to this star or this lab. The data is free, the code is public, and the only thing on offer is the working.

  8. You can run the same search

    Everything the lab did to this star is packed into a free kit: the same pipeline and the same rules, cut down to one person, one star and a laptop. It makes you write your question down before you look, tries hard to kill whatever dip you find, and ends in one of six plain words. None of them is planet.

    Find your own planet →

Depth 1 · the numbers, each with a plain reading

The object, in brief

Period
1.9369484 done dip every 1 day 22½ hours; most likely half a lap
Depth
10.46 %the star loses about a tenth of its light
Host
mid-M dwarf, 222 pca small red star about 720 light-years away
SPOC detections
259 transits, 23 sectorsNASA’s software saw the dip 259 times over 23 months of watching
Geometry
grazing, b ≈ 0.9–1.05it clips the star’s edge, not its face
Size, if a planet
2.6–3.9 RJup2.6 to 3.9 Jupiters wide, never pinned down; too big to believe around so small a star
Phase-folded TESS light curve of TIC 374861595: a flat baseline with one deep, V-shaped dip of about ten percent at phase zero and no visible secondary eclipse at phase one half.
The lab's own fold of the signal: every dip, 1.94 days apart, stacked in one place. That one deep dip is most likely two stars' eclipses on top of each other, each star covering part of the other in turn, which is why no second dip shows. The 8.41 % in its title is the lab's trapezoid fit, which underestimates a grazing dip; the 10.46 % above is SPOC's limb-darkened fit.

What it is. A 10 % dip every 1.94 days on what the catalogue lists as one small, cool star 222 parsecs away. It is most likely two near-twin stars on a 3.874-day orbit, each covering part of the other once a lap. The lab’s blind search found the dip on its own, in three separate TESS sectors, without being told where to look. The star has no TOI, no community candidate, and no variable-star entry.

What is solid. The dip comes from this star’s position on the sky, not from a neighbouring star. That still holds. So does the lab’s newest measurement. In the 16 sectors from 2023 to 2025, the light between the dips rose and fell in step with twice the dip period for 255 cycles: 3.87411 ± 0.00039 days against 3.87390. In 2020 and 2021 the rise and fall sat at a different point in the cycle and moved between seasons, as starspots do.

What did not hold. Two things the lab had counted as solid. Every other dip nearly matches the ones in between (the odd/even test), and the lab took that to rule out two stars on a lap twice as long. But near-twins make near-twin dips. The lab also expected a second star to make a small extra dip halfway between the main ones, as it passes behind. None showed, and the lab took that to rule out a star. If the true period is double, that dip was never missing. It is every other dip.

What is not. The size was never pinned down. An edge-clip hides how big the object is, and a refit over 24 sectors made the size bigger, not better measured: 2.6 to 3.9 Jupiters wide. Whether the two dips differ is not settled either. The lab’s two estimates say one is deeper by 0.2 to 0.4 percentage points in most sectors, at 3 to 5σ. Three sectors read the other way, and NASA’s own comparison over all 259 dips finds no difference. The lab has not explained that. Read as one star with a planet, the shape of the dip also implies one density for the star, and the star’s brightness and distance imply another, 7.7σ apart. The lab listed three possible causes, one of them more than one star in the light, and followed none. The mass was never measured, so two stars is the most likely reading, not a measured one.

Nothing was submitted. The lab never sent this star to ExoFOP, the community list of planet candidates, and it will not, because this is most likely two stars. Since 2026-08-19 ExoFOP has also taken community candidates only after a peer-reviewed paper, and the survey paper is still a draft.

Words on this page, in plain language
TIC
The TESS Input Catalog: a numbered list of more than a billion stars. TIC 374861595 is just this star’s entry number.
TESS
NASA’s Transiting Exoplanet Survey Satellite, launched 2018. It measures the brightness of most of the sky, one patch at a time.
Sector
One of TESS’s patches, watched for about 27 days. “S27” is sector 27.
SPOC
NASA’s automatic processing pipeline for TESS. Its Data Validation report (DV) is the official fit of this signal.
TOI, CTOI, ExoFOP
A TOI is a TESS Object of Interest, the official follow-up list. A CTOI is a Community TOI, one proposed from outside. ExoFOP is the website where both live.
Transit, depth, period
A transit is one dip. Depth is how much light goes missing. Period is the time between dips. For a planet that is one full orbit. For two near-twin stars on a circular orbit it is half of one, because each star covers the other once per orbit.
P, 2P
P is the 1.94 days between dips. 2P is twice that, 3.874 days: most likely the true orbit, with two near-twin stars and two dips per lap.
M dwarf
A small, cool, reddish star, the most common kind in the galaxy. This one is about 0.6 of the Sun’s width and mass, about 3,200 °C at the surface against the Sun’s 5,500.
pc (parsec)
An astronomer’s distance unit, about 3.26 light-years. 222 pc is roughly 720 light-years.
Grazing, b
Grazing means the object clips the star’s edge instead of crossing its face. b, the impact parameter, says how far off-centre the crossing is: 0 is dead centre, 1 is the very edge.
RJup, R⊕, R☉
Sizes measured in Jupiters, Earths and Suns.
Secondary eclipse
A second, shallower dip half an orbit later, when the companion goes behind the star. A glowing companion such as a star would show one. Folded at 1.94 days, none shows halfway between the dips. If the orbit is twice that, as is most likely, the missing dip is every other main dip.
Odd/even, eclipsing binary
An eclipsing binary is two stars taking turns blocking each other. If the dips alternate deep and shallow, the real period is double. These dips nearly match, which near-twin stars also produce. The lab’s estimates find them 0.2 to 0.4 percentage points apart over 24 sectors; NASA’s own comparison finds no difference.
Brown dwarf, radial velocity
A brown dwarf is heavier than a planet but too light to shine like a star. Radial velocity measures the star’s back-and-forth wobble, which gives the companion’s mass.
σ (sigma)
How surprising a difference is, measured against the noise. 1σ is ordinary; 7.7σ is far too large to be chance.
BLS, SDE
Box Least Squares is the search method: slide a box-shaped dip along the data at every trial period. SDE scores how much a peak stands out.
Open lead
The lab’s word for a signal that passed every automatic check and waits on people and more data. It claims nothing. This star was one from 2026-09-01 to 2026-11-10.
Refuted
The lab’s word for a lead it has ruled out. For this star it means most likely two stars, not a planet. It does not mean a second star has been measured.
PC
In the ExoFOP line in §1: the standard code for “planet candidate”, the entry-level label for a transit worth following up. Candidate means worth checking, not confirmed.

Depth 2 · the working dossier

What follows is the lab’s own file on this object, unedited, including its own retractions. It now opens with the 2026-11-10 refutation, and everything below that is kept as the record, not as a current claim. It is written for astronomers. A good order for everyone else: the refutation first, then §1, the one table, then §5, where the lab tried to break its own case and did not, then §0, what it had already got wrong and fixed.

Prepared 2026-09-14, adversarially re-checked 2026-09-14, corrected and re-scoped 2026-09-18. Nothing has been sent.

Refuted on 2026-11-10 — read this before anything below

This star is most likely an eclipsing binary with a period of 3.874 d, twice the 1.937 d this package assumes. It is not a planet candidate. Nothing was sent to ExoFOP and nothing will be. The upload line in §1 is dead. Everything below is kept as the record of how the lab got here; none of it is a current claim.

The reasons came from outside. A TOI vetter read some of the lab's analysis of this star and opened issue #189 on 2026-11-06: a planet this size is implausible around an M dwarf, and the light varies coherently at twice the period, which the issue reads as a close-to-equal-mass binary at 2P. Both signals were already in this package and were explained away. On 2026-11-10 the lab ran its own gates and one new measurement on all 24 light curves it had on disk, and its data agree with the issue. The full account is in docs/investigations/2026-11-10-tic374861595-eb-at-2p-refutation.md.

How the package's main claims read now:

claim below where how it reads now
"Odd/even eclipses equal" and "Not an EB at 2P." §2, §5d A close-to-equal-mass pair has near-equal eclipses, so near-equal depths do not exclude 2P. Over all 24 sectors the two dips at 2P differ by 0.2 to 0.4 percentage points, the same one deeper in 18 to 21 sectors, at 3σ to 5σ.
"No secondary eclipse" as the load-bearing measurement; "a stellar companion is ruled out" §2, §5c If the period is 2P, the dips at P are the primary and the secondary, and the bound tests nothing. This lab wrote that down on 2026-08-18 for TIC 287328866.
Size: "The grazing degeneracy is the likely explanation." §2, §5f SPOC's fit (2.59 ± 0.46 R_Jup) sits at the lab's own 2.5 R_Jup planet ceiling and the refit (2.65 to 3.92 R_Jup) is over it. The simpler reading is that the companion is a star.
Out-of-transit modulation: "Probably the host's rotation near a 2:1 spin–orbit commensurability." §2 Measured over all 24 sectors on 2026-11-10: the modulation has matched 2P to about one part in ten thousand for 255 cycles (3.87411 ± 0.00039 d against 3.87390 d). That is rotation locked to a 3.874 d orbit, which is what tides in a close binary produce.
"Planet candidate, unchanged." and, of the refit, "it hardens it." §5e Withdrawn.
The paper's thesis: human triage left this signal in a gap §3, §5a Withdrawn. The triage was most likely right.

What the lab has still not done: read Gaia's radial-velocity error for itself, or explained why SPOC's own odd/even comparison reads no difference where the lab's estimators read one. Both are listed in the investigation note.

0 · Read this first: what changed on 2026-09-18

Three numbers in the previous revision were wrong and one framing decision was open. All four are settled here. The density row was corrected twice on the same day — once wrongly — and the row records both passes, because a package that hides its own retractions is not a package anyone should trust.

was is why
The ExoFOP upload line in §1 carried this lab's trapezoid values carries SPOC's limb-darkened values §5 already concluded SPOC's ephemeris supersedes ours, but the paste-ready line was never updated. It was the one artifact in the package meant to be pasted, and it was the one carrying the retracted numbers.
Stellar density — both sides were in solar units, and only one carried the label "ρ★ from fit 1.64; catalogue 2.57", the second tagged g/cm³ fit 1.644 ρ☉ = 2.32 g/cm³; catalogue 2.571 ρ☉ = 3.62 g/cm³ 0.601 / 0.616³ = 2.571, so the catalogue's "2.57 g/cm³" is M/R³ in SOLAR units with a cgs label on it. SPOC's starDensitySolarDensity = 1.64423 is solar too. The original comparison was therefore like-for-like and the factor-1.56 discrepancy is real; only the unit label was wrong. A 2026-09-18 revision of this file briefly claimed the two agree to 11 % — that was an error, made by converting one side to cgs and not the other, and it is retracted here.
Odd/even statistic "1.13σ" 1.288 (significance 0.256) oddEvenTransitDepthComparisonStatistic value="1.2880648" in the committed DV XML.
Companion exclusion "every stellar and L-type companion excluded by two orders of magnitude"; survivors "cooler than ~1,000 K" stars excluded by 10–48×, early-L marginal, survivors cooler than ≈1,800 K Recomputed by Planck integration over the TESS band in scripts/tic374861595_secondary_limit.py. The conclusion holds; the margin was overstated.
Baseline and sector count "259 over ~2,300 d (980 expected; 25+ sectors)"; sector list given as ranges "35–39, 61–69, 87–97" 259 over 1,897 d; 23 sectors, the first being S27; the sectors enumerated individually The DV root's own startCadence/endCadence span 1,365,963 two-minute cadences = 1,897.2 d = 979.5 cycles, which is the DV's own expectedTransitCount of 980 to 0.05 %. "~2,300 d" implies 1,187 cycles and contradicts that number by 207. The sectorsObserved bitmap has exactly 23 bits set and none below 27, so the range forms named six sectors the DV does not cover: S39, S64, S91, S92, S95 and S97. Five of those six are simply not observations of this target; S97 is different — the refit reached it (§5f), so it is real and merely postdates this DV. See §5f's coverage note. Found by the survey-paper thread; verified here against the committed XML.

Plus: the AAVSO VSX check that was outstanding is now closed (§5d), and the venue question is answered (§3) — Research Notes of the AAS do not satisfy ExoFOP's rule, because the AAS itself says RNAAS is non-peer-reviewed.

And later the same day, two things that had been waiting on someone else's machine: the Mann+2019 0.72 M☉ line is retracted (§5d, checked against the authors' own posterior), and the refit has been run (§5f) — over 24 sectors rather than three. It derives R★ instead of copying it, explains the O−C drift as the trapezoid's period error, and leaves the radius less determined than before, not more. Read §5f before quoting any radius from this package.

0.1 · The gate

ExoFOP-TESS changed its community-candidate policy on 2026-08-19 (read on exofop.ipac.caltech.edu/tess/candidate_help.php, 2026-09-14):

"Candidates must be published in a peer-reviewed journal that offers online access before being uploaded to ExoFOP." … "approval to upload candidates is required" (Published Candidate Upload Request form) … "Paper URL (required)".

So "submit a CTOI" is no longer a form. It is a peer-reviewed paper, then an approval request, then the upload. That page disallows automated fetching, so this quote is a human transcription from 2026-09-14 and has not been re-read since — check it against the live page before acting on it.

Confirmed: no existing TOI or CTOI for this TIC (full CTOI table downloaded 2026-09-14, 5,137 rows, zero matches; ExoFOP target page reads "No TOI").

1 · The candidate, in one table

Stellar parameters from TIC v8. Transit parameters are SPOC's multi-sector Data Validation fit (sectors 1–96, generated 2025-12-24, spoc-5.0.125), not this lab's — see §5 for why, and §2 for what this lab's independent pass adds.

value source
Host TIC 374861595 · Gaia DR3 4756470194913503104 · 2MASS J05542154-6407473 TIC v8
RA, Dec (ICRS) 88.5897756°, −64.1297956° (05:54:21.55 −64:07:47.3) TIC v8
Tmag / G 14.076 / 15.25 TIC v8 / Gaia DR3
Distance 222.2 ± 1.6 pc (plx 4.47 ± 0.03 mas) TIC v8 (Bailer-Jones 2018)
Teff, R★, M★, ρ★ 3445 ± 157 K · 0.616 ± 0.019 R☉ · 0.601 ± 0.020 M☉ · 2.571 ρ☉ = 3.62 g/cm³ TIC v8 (cool-dwarf relations)
Gaia RUWE 1.05 (no astrometric binary signal) Gaia DR3
TIC contamination ratio 0.070 TIC v8
Period 1.9369484 ± 0.0000002 d SPOC DV s1–s96, 259 transits
Epoch (BJD_TDB) 2459036.54794 ± 0.00005 (BTJD 2036.54794) SPOC DV s1–s96
Depth 104603 ± 405 ppm (10.46 %) SPOC limb-darkened Mandel-Agol fit
Duration T14 2.1533 ± 0.0063 h (ingress 1.0767 h) SPOC DV
Transits observed 259 over 1,897 d (980 expected; 23 sectors, first is S27) SPOC DV
Rp/R★ 0.42988 ± 0.07582 SPOC DV
Rp 28.90 ± 5.17 R⊕ = 2.59 ± 0.46 R_Jup SPOC DV
Impact parameter b 0.9000 ± 0.1120 — grazing SPOC DV
Inclination 83.300 ± 0.799° SPOC DV
a/R★ 7.71473 ± 0.04768 → ρ★ = 1.644 ρ☉ = 2.32 g/cm³ SPOC DV; ρ★ re-derived here
T_eq (A=0, full redistribution) 744.5 ± 34.3 K SPOC DV
MES / model SNR / bootstrap FAP 508.60 / 507.39 / 0 SPOC DV

ExoFOP upload line

Pipe-delimited, in the public export's column order for the fields we have. Dead as of 2026-11-10: the star is refuted (see the top of this file). Do not send this. It had been held until the paper existed and the upload request was approved.

target|flag|disposition|period|period_err|epoch|epoch_err|depth_ppm|depth_ppm_err|duration_hrs|duration_hrs_err|rp_rearth|rp_rearth_err|incl|incl_err|b|b_err|rprs|rprs_err|ars|ars_err|notes
374861595|newctoi|PC|1.9369484|0.0000002|2459036.54794|0.00005|104603|405|2.1533|0.0063|28.90|5.17|83.300|0.799|0.900|0.112|0.42988|0.07582|7.71473|0.04768|Blind BLS recovery in windowsill-lab survey; parameters adopted from SPOC multi-sector DV s1-s96 (2025-12-24); grazing, b=0.90; no secondary eclipse, stellar companion excluded; mass unknown

Tag (ExoFOP convention YYYYMMDD_username_description_nnnn): <YYYYMMDD>_<exofop-username>_windowsill-tic374861595_0001.

One thing still needs a human eye: the exact upload-form column order lives behind the login wall (add_new.php?param=bulkparams). The line above follows the public export's order. Open the form once and check the header before pasting.

2 · Why it is a candidate, and what this lab independently contributes

This lab did not discover the signal — SPOC has it as TCE 1 in every 2-minute sector. What a blind search added is an independent recovery and, more usefully, one measurement SPOC's pipeline does not make: the exclusion of a stellar companion from the absent secondary eclipse (§5c). That is the load-bearing thing.

For.

Against, or unresolved.

Verdict: planet candidate (PC) — a grazing transit by an object cooler than ≈1,800 K on a 1.937-day orbit around an M2–M3 dwarf at 222 pc. Stellar companion excluded by the secondary-eclipse limit. Radius degenerate, mass unknown.

3 · Venue — the question is settled, and RNAAS loses

The previous revision left this open: "a Research Note of the AAS if RNAAS counts as peer-reviewed under ExoFOP's rule." It does not.

The AAS says so itself, on journals.aas.org/research-notes/ (read 2026-09-18):

"Research Notes of the AAS is non-peer reviewed, indexed and secure record of works in progress" … "Research Notes are moderated but not edited."

So RNAAS cannot clear a gate whose text is "published in a peer-reviewed journal." Do not spend the writing effort there expecting a CTOI upload at the end of it. RNAAS remains a legitimate way to put a citable, ADS-indexed marker down in days, and it is worth knowing that option exists — but it is a separate decision with a separate purpose, and taking it does not advance the ExoFOP path by one step.

What the paper should be

A survey-methods paper in AJ (or PASP), with this candidate as its worked case — not a single-object note. Reasons, in order of weight:

  1. A single-candidate paper has thin new content and a hostile referee. SPOC found this signal 259 times and published the DV reports. A paper whose claim is "we independently recovered a signal the pipeline already has, and we think it deserved promotion" invites exactly one question: what is new? The honest answer — one archival measurement (the secondary bound) and a disagreement with a triage decision — is a paragraph, not a paper.
  2. The survey is real, is already done, and now has a draft. Re-derived from the receipts on 2026-09-18 by scripts/survey_paper_numbers.py: 12,898 target-searches over a sample of 12,151–12,702 distinct stars, 116 above threshold, every one dispositioned from a vocabulary of 18 terms, 14 of them used, 9 known planets recovered blind, 7 leads minted and 5 killed by the survey's own gates (of 7 refutation rulings), 0 planets claimed — plus 5 hunt files refused outright by their own controls (uniformity-failed, budget-over-share). A pipeline that logs its own refusals is the unusual part, and it is the part that is publishable. Quote figures from that reader, never from memory or from this paragraph — "12,898 targets", "closed 14-term vocabulary" and a bare "5 refutations" are each wrong in a way a referee would catch. The survey-paper draft itself is on claude/project-thread-alqen7 (1c8a7f1).
  3. The two halves need each other. A survey that claims no occurrence rate and no detection completeness (the repo's claim_boundary explicitly disclaims both) has a weak positive result on its own. A candidate with no survey around it has no denominator. Together they make one defensible claim: an automated disposition pass over 12,898 targets surfaced a high-MES signal that human triage had set aside across 23 sectors, and the measurement that discriminates it was sitting in the archive the whole time.
  4. It clears the ExoFOP gate on its own terms. AJ and PASP are peer-reviewed without argument, so the paper that makes the science case is also the paper that unlocks the upload. One writing effort, both outcomes.

Frame it as a gap, not an accusation. The claim is not "the TESS Science Office erred." It is: high-MES, grazing, large-inferred-radius signals on faint M dwarfs fall into a triage gap, because the cheap archival test that resolves them — the secondary-eclipse surface-brightness bound — is not part of the standard promotion path. That is constructive, checkable, and generalises past one target. An accusation does not survive review; a named gap with a proposed test does.

4 · What "submit" requires, in order

  1. Two analysis upgrades (scripts/tic374861595_refit.py, written to run on Ben's machine because this environment cannot reach MAST): a limb-darkened Mandel-Agol fit with the grazing degeneracy properly explored, and a Gaia-anchored R★ with its own errors. Without these the radius is indefensible.
  2. Write the paper per §3. AJ or PASP.
  3. ExoFOP account — "REQUEST AN ACCOUNT" on exofop.ipac.caltech.edu/tess (its own registration, not the Exoplanet Archive login). Independent of everything above; can be done today.
  4. Published Candidate Upload Request (pub_candidate_upload_request.php) with the paper URL.
  5. Upload via add_new.php?param=bulkparams with the line in §1 and the tag.

5 · Adversarial re-check — SPOC has seen this 259 times

5a · Someone has looked, and did not promote it

MAST holds SPOC Data-Validation products for TIC 374861595 in every 2-minute sector it has — 23 of them, read off the DV bitmap: 27, 28, 30, 31, 35, 36, 37, 38, 61, 62, 63, 65, 66, 67, 68, 69, 87, 88, 89, 90, 93, 94, 96 — plus multi-sector runs. The latest multi-sector DV (sectors 1–96, 2025-12-24, spoc-5.0.125; XML committed beside this file, every number in §1 re-extracted from it 2026-09-18) finds our signal as TCE 1 at MES 508.6, bootstrap FAP 0, suspectedEclipsingBinary: false, centroid on target, ghost diagnostic clean.

The pipeline vetted it clean and the TESS Science Office never promoted it. Note the coverage starts at sector 27 (July 2020), not at launch — TESS did not observe this target in its first two years, so the unpromoted span is about five years, not seven. The likeliest reading is human triage: a V-shaped 10 % transit giving Rp > 2 R_Jup on a faint M dwarf is what gets set aside as a probable eclipsing binary without further work. That is the objection the paper must answer head-on, and §5c is the answer.

Two further TCEs (268 d and 291 d, 3 and 2 events, MES ~12) are multi-sector false alarms of the usual kind and are not part of this candidate.

5b · What the re-check withdrew from earlier revisions

5c · What stands, and why the grazing geometry does not spoil it

No secondary eclipse. This lab: −159 ± 628 ppm over 34 events on undetrended flux with per-event local baselines. SPOC: maximum weak-secondary MES 3.3, 602 ± 168 ppm, at phase 0.328 d, not at 0.5 P = 0.968 d. Two routes, same null. Adopted 3σ ceiling on a phase-0.5 event: 1,105 ppm.

The obvious worry is that b = 0.90 breaks the usual secondary/primary relation, because only part of the companion's disc ever crosses the star. It does not, and the cancellation is exact:

primary depth   = A_overlap · I_star      / F_total
secondary depth = A_overlap · I_companion / F_total

The same overlap area appears in both — at secondary the star covers exactly the region the companion covered at primary — so

secondary / primary = I_companion / I_star, the surface-brightness ratio, independent of b, of k, and of how grazing the configuration is.

Against the 10.46 % primary, the 1,105 ppm ceiling bounds the companion's TESS-band surface brightness at < 1.06 % of the host's. Computed by Planck integration over the TESS band (scripts/tic374861595_secondary_limit.py, two different bandpass stand-ins agreeing to 1 %):

companion Teff I₂/I★ predicted secondary verdict
M5V star 3050 K 5.1e−1 52,800 ppm excluded, 48× over
M7V star 2650 K 2.1e−1 21,800 ppm excluded, 20× over
M9V / BD boundary 2400 K 1.0e−1 11,000 ppm excluded, 10× over
L2 brown dwarf 1900 K 1.6e−2 1,670 ppm excluded, 1.5× over
L8 brown dwarf 1400 K 7.2e−4 75 ppm allowed
T4 brown dwarf 1100 K 3.2e−5 3 ppm allowed
planet at T_eq 745 K 4.0e−8 ~0 ppm allowed

Cut-off: T_companion ≲ 1,800 K. Every stellar companion is excluded by 10–48×; early-L is excluded marginally; mid-L and cooler survive. The previous revision said "excluded by two orders of magnitude" with a 1,000 K cut-off — both were overstated, and the correction is in the direction that matters (the true bound is looser than claimed). The conclusion is unchanged: a stellar companion is ruled out; a giant planet or a cool brown dwarf is not.

Two caveats, both stated rather than buried. Reflected light and thermal redistribution are ignored, which can only add secondary flux and so makes the exclusions conservative. Blackbodies are a poor model for L and T dwarfs, which are fainter than blackbodies in the TESS band — so the real cut-off sits somewhat above 1,800 K, again in the permissive direction for the objects being kept.

5d · Independent checks that came back clean

5e · Where the package stands

Planet candidate, unchanged. A strong, on-target, grazing transit that SPOC has seen 259 times and never promoted, whose one discriminating measurement is the missing secondary eclipse — which rules out a stellar companion and leaves a giant planet or a cool brown dwarf. The paper writes itself around that measurement, the survey that found it blind, and an RV request. It does not write itself around a radius, because the radius is degenerate.

§5f reports the refit, which has now been run. It does not change this verdict; it hardens it.

5f · The refit, run — 2026-09-18

scripts/tic374861595_refit.py has been run against MAST. Full output in TIC374861595-refit-2026-09-18.log, machine-readable in the matching .json:

py -3.11 scripts/tic374861595_refit.py --backend numpy \
  --kmag 11.710 --kmag-err 0.026 \
  --parallax 4.512610 --parallax-err 0.023275 \
  --errors mcmc --json docs/submissions/TIC374861595-refit-2026-09-18.json

The archive is much larger than this package assumed. 24 SPOC 2-minute sectors (27–97), 378,654 good cadences, BTJD 2036.3 to 3988.3 — a 1,952-day (5.3-year) baseline, 289 transit windows, 105,227 cadences in the fit. The blind search used three sectors.

Why this is 24 sectors and §1 is 23. They count different things and both are right. §1 is the DV run, generated 2025-12-24 over sectors 1–96: its sectorsObserved bitmap has 23 bits, the highest S96, spanning 1,897 d. This refit queried MAST on 2026-09-18, nine months later, and got one sector more and 55 days more. The extra sector's identity is not actually pinned down by either artifact. The 55-day tail is 2.00 sector lengths past the DV's last cadence while the count rises by only one, which fits one skipped sector followed by one observed one — and that would make the highest sector S98, not the S97 written above. Neither the log nor the JSON enumerates the sectors, so the "27–97" range label is a typed number, not a derived one. Do not quote it in the paper until the script prints the list.

Both gaps are closed in code as of 2026-09-19, and both need a re-run on a machine with MAST access to produce their output. The script now derives the sector list from each product's own SECTOR keyword, prints it, counts any product it could not identify, and ships it in the JSON as coverage; and --chain <file.npz> writes the joint MCMC samples, with the JSON saying so when it is not passed. Five tests in tests/test_tic374861595_refit.py gate the coverage helper, including that it will not emit a range across a sector that was never observed.

The stellar inputs are now measured rather than back-derived. 2MASS Ks = 11.710 ± 0.026 (PSC 05542154−6407473; Qflg AAA, Cflg 000, Xflg 0 — clean) and Gaia DR3 parallax 4.512610 ± 0.023275 mas (RUWE 1.05, non_single_star = 0). TIC v8's 4.47187 mas is DR2. The script's built-in default K is back-derived from the package's own M_K and must never be used for anything published; pass the real one.

What the refit settled

row was is
R★ catalogue — "agrees with TIC rather than checking it" 0.6148 ± 0.0186 R☉, derived here from the measured Ks and the DR3 parallax, M_K = 4.982 ± 0.028. It reproduces TIC to 0.2 %, which is now a check instead of a copy.
M★ disputed; one line said 0.72 M☉ 0.5998 ± 0.0185 M☉
O−C "+5.2, +3.2, +0.8 min over 73 cycles, monotone; almost certainly trapezoid mid-time bias" Explained — and the guess was right in spirit, wrong in detail. It is the trapezoid's period, not its mid-times. The trapezoid period sits 3.382 s/cycle short of SPOC's (16 σ on its own error bar); over 73 cycles that predicts −4.12 min of drift against the −4.40 min observed, agreeing to 7 %. No TTV is required, and none is claimed.

What it did not settle — and one thing it made worse

One operational note. --backend auto picks batman whenever it is installed, and that path rebuilds its model inside every likelihood call: 348 ms per evaluation against 3.9 ms for the numpy integrator at the same cadence count. That is ~90× and it is the difference between a roughly three-hour MCMC and one on the order of ten days. The two backends agree to 1.54 ppm in --selftest. Pass --backend numpy until the model object is hoisted out of the likelihood.

6 · Files