{"id":"51298e8f-afab-4f91-8471-b84a71788e3b","arxiv_id":"astro-ph/9601067","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The 4-year COBE DMR maps yield a power-law spectral index n = 1.2 plus or minus 0.3 and a quadrupole normalization of 15.3 microkelvin, consistent with a scale-invariant Gaussian field.","lead":"Four years of COBE satellite microwave data confirm that the early universe's faint temperature ripples follow a scale-invariant, Gaussian pattern, with spectral index 1.2 plus or minus 0.3. The full-sky maps and the foreground model behind that number became the calibration standard for every later cosmic microwave background experiment.","discovery_kind":"extension","skeptic_critique":{"model":"selfhosted/deepseek-v4.1-flash","headline":"The abstract's n = 1.2 ± 0.3 and Q = 15.3 (+3.8/−2.8) are taken from the DIRBE-template rows of Table 2; the free-free-cancelling combination in the same table gives n = 1.0–1.1 with ±0.4 errors, and §3 asserts rather than demonstrates that this spread is benign.","rationale":"The paper's central claim has two halves: the fitted spectral index and normalization (abstract:s3, s4), and consistency with Gaussian statistics (abstract:s2). The reader's weakest_assumption targets the first half through foreground modeling, and that is the right place to push. I agree with the identification and would only sharpen it: the concern is not hypothetical template misestimation but a spread already tabulated in Table 2, where the three Galactic treatments give n = 1.0, 1.1, and 1.2 with errors ranging from ±0.27 to ±0.43. The abstract selects the tightest of these. The paper is commendably transparent about the alternatives — all three treatments appear in the table with their own error bars — and the §3 sentence about data selection versus analysis technique is a fair summary of the authors' belief, just not a demonstrated one within this Letter. Because the discrepancy is at roughly the 1σ level against the larger error, the result survives; what is at stake is the precision claim, not the detection. I considered the Gaussianity half as an alternative load-bearing concern: the 'five times less likely' statement in §3 and summary point (6) is a likelihood ratio of only about 5:1 (Δln L ≈ 1.6) against a one-parameter χ²_N toy family, and the three statistics cited (three-point function, genus, extrema correlations) all probe low-order moments of a field already smoothed to 7°, so they are less independent than the enumeration suggests. The reader's strongest_claim escalates this to 'exclude,' which overstates the paper's own careful wording. But that clause is secondary to the n and Q numbers, and the paper itself hedges appropriately ('it is still possible for interesting signals to be hidden in the data'), so I do not treat it as the single most load-bearing concern. On independent support: the result has been corroborated by WMAP and Planck, the maps and custom cut were released through the NSSDC, and four distinct estimators (two-point function, orthogonal functions on the cut sphere, pixel-temperature likelihood, Hauser-Peebles) agree within errors — that is real, checkable evidence and it is why the verdict stays ACCEPT. The recommended adjustment is none: the concern argues for a modestly wider error bar or a two-value presentation in the abstract, not for changing the verdict. The concrete test above would settle whether the method spread is noise or foreground systematic, and it is a check the authors could have run with the data already in hand.","tokens_in":11500,"tokens_out":4943,"duration_ms":97652,"concrete_test":"Re-fit the joint (Q, n) likelihood on the internal-combination (free-free-cancelling) maps and on the DIRBE-template maps using an identical surviving pixel set, identical pixel weighting, and identical beam and pixel window, so that the only difference between the two fits is the foreground treatment. Then run the same two fits on Monte Carlo realizations of a common input sky plus DMR noise. If the observed n difference exceeds what the noise-only realizations produce, or exceeds the quoted ±0.3, the abstract's error bar understates the foreground systematic and should be widened, or the two values quoted separately in the abstract.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline numbers come from one of three Galactic-correction treatments tabulated in Table 2. Under 'DIRBE-Template Galaxy Correction' the paper reports n = 1.21 (+0.24/−0.28) for orthogonal functions and 1.23 (+0.26/−0.27) for pixel temperatures, with Q = 15.2 (+3.7/−2.6). The abstract quotes n = 1.2 ± 0.3 and Q = 15.3 (+3.8/−2.8), i.e. the DIRBE-template values carrying the tightest errors of the three treatments. The 'Internal Combination Galaxy Correction' rows — the treatment explicitly constructed to cancel free-free emission — instead give n = 1.11 (+0.38/−0.42), 1.00 (+0.40/−0.43), and 1.62 (+0.44/−0.50) for the Hauser-Peebles estimator, with Q = 16.3 (+5.2/−3.7) and 17.2 (+5.6/−4.0). The choice of foreground treatment therefore moves the central n by 0.1–0.2 and widens the 68% interval by roughly 50%. Section 3 disposes of this with a single sentence: 'Differences in the derived values of Q and n between various analyses of DMR data are much more dependent on the detailed data selection effects than on the analysis technique.' That is an assertion, not a test; nothing in this Letter separates 'data selection' from 'foreground treatment' as the cause of the spread. The reader's weakest_assumption identifies the same pressure point (foreground modeling after the custom cut), but frames it conditionally — what would happen if the templates misestimated the sky. The sharper form is that the spread is already visible in the paper's own table, so the live question is whether the quoted ±0.3 is representative of the method-to-method scatter. On the numbers as printed it probably is at roughly 1σ (1.2 versus 1.0 is about half the larger error), so this is a concern about error-bar representativeness, not a refutation of the result.","agreement_with_reader":"agree"},"referee_report":{"model":"selfhosted/deepseek-v4.1-flash","summary":"This Letter summarizes the full 4-year COBE DMR data set: full-sky maps at 31.5, 53, and 90 GHz (Fig. 1), a DIRBE 140 µm-guided custom Galactic cut leaving 3881 of 6144 pixels (Fig. 3), foreground modeling of dust and free-free emission, and a suite of four power-spectrum estimators (2-point function, orthogonal functions on the cut sphere, pixel-temperature likelihood, Hauser-Peebles) applied under three foreground treatments (Table 2). The headline results are a power-law index n = 1.2 ± 0.3 and quadrupole normalization Q_rms-PS = 15.3 (+3.8/−2.8) µK, with Q_rms-PS|n=1 = 18 ± 1.6 µK; a DMR monopole T0 = 2.725 ± 0.020 K; a dipole of 3.353 ± 0.024 mK toward (l,b) = (264.26, 48.22); an observed quadrupole Q_rms = 10.7 ± 3.6 ± 7.1 µK; and Gaussianity tests (3-point function, extrema correlations, genus) that prefer Gaussian models over χ²_N toy models by a factor of about five. The Letter is explicitly a summary of a set of companion papers, and most methodological detail is deferred to them.","tokens_in":11851,"tokens_out":4963,"duration_ms":105699,"significance":"If the results hold, this is the definitive statement of the large-angle CMB anisotropy from the DMR experiment, and the n = 1 consistency is a direct observational anchor for Harrison-Zeldovich initial conditions and inflationary models. The paper's methodological strengths deserve explicit credit: four independent estimators are cross-checked against one another and against three foreground treatments in Table 2, and the values agree within the quoted errors; cosmic variance is included in the error budget rather than treated as an afterthought; the systematic error floor is bounded at 6 µK (95% CL) in the worst channel; the Gaussianity claim is tested with statistics calibrated on independent Monte Carlo realizations rather than asserted from the fit; and the time-ordered data, maps, and custom cut are released publicly through NSSDC, which makes the analysis reproducible. The Gaussianity result is a falsifiable prediction in the sense that the χ²_N family is explicitly excluded at a stated likelihood ratio.","major_comments":[{"comment":"The abstract and the summary quote n = 1.2 ± 0.3 and Q_rms-PS = 15.3 (+3.8/−2.8) µK, which are the DIRBE-template values from Table 2 and carry the tightest errors of the three foreground treatments. The 'Internal Combination Galaxy Correction' rows, constructed specifically to cancel free-free emission, instead give n = 1.11 (+0.38/−0.42), 1.00 (+0.40/−0.43), and 1.62 (+0.44/−0.50) with Q = 16.3 (+5.2/−3.7) and 17.2 (+5.6/−4.0) µK. The qualitative conclusion (n consistent with unity) survives every row, so this is not a challenge to the central claim; but the abstract is the citable artifact, and it currently reports the smallest of three quoted uncertainties without flagging the spread. §3 disposes of the issue in one sentence ('Differences in the derived values of Q and n ... are much more dependent on the detailed data selection effects than on the analysis technique'), which is an assertion rather than a test. I recommend either quoting a foreground-treatment range in the abstract or adding a sentence stating explicitly that the quoted errors are the tightest of the three treatments tabulated, and that the free-free-cancelling combination gives a larger uncertainty on n.","section":"Abstract; §4 item (5); Table 2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The quadrupole is quoted as Q_rms = 10.7 ± 3.6 ± 7.1 µK, where the second error is Galactic modeling and dominates. The stated 68% interval, 6 µK ≤ Q_rms ≤ 17 µK, is narrower than the quadrature sum of the two quoted errors (8.0 µK, which would give roughly 3–19 µK), and §4 item (4) gives a 95% interval of 4 µK ≤ Q_rms ≤ 28 µK. Please explain how the intervals were derived from the likelihood (the shape of the distribution, and how the Galactic modeling error was propagated), so that the reader can see that the ± 3.6 ± 7.1 decomposition and the two intervals are mutually consistent. Since the consistency of the observed quadrupole with Q_rms-PS is one of the six numbered summary claims, this internal check is load-bearing for item (4).","section":"§3 (Quadrupole ℓ = 2); §4 item (4)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The custom cut is described as maximizing the number of surviving pixels while minimizing Galactic contamination, but no objective function or threshold criterion is stated, and the sensitivity of the derived parameters to the cut is acknowledged only qualitatively: 'Moderate changes to this custom cut will cause derived CMB parameters to change somewhat.' The authors have the machinery in hand (Fig. 3 already shows residual Galactic signal versus surviving pixel count), so a quantitative statement would materially strengthen the paper: for two or three alternative cut thresholds bracketing the adopted one, report the resulting n and Q_rms-PS. This would also let the reader separate the 'data selection' explanation offered in §3 from the foreground-treatment explanation, which is the live question raised by Table 2.","section":"§2 (Galactic cut); Fig. 3","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"Footnote j cites 'Kogut et al. 1996' without a letter suffix, while the reference list contains 1996a, 1996b, and 1996c. Please disambiguate.","section":"Table 2, footnote j","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The Hauser-Peebles entry under 'DIRBE-Template Galaxy Correction' is blank ('—'). A one-line note explaining why that estimator was not applied under that treatment would prevent the reader from reading the blank as a withheld result.","section":"Table 2, DIRBE-Template rows","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The text gives mean signal-to-noise ratios of ~0.5, 1.5, and 1.0 for the 31, 53, and 90 GHz 10°-smoothed maps, while the abstract quotes ~2 per 10° patch for the multi-frequency coadded map. Please state explicitly whether these are per-pixel or per-beam-patch values and how the coadded figure is obtained from the per-channel ones.","section":"§2, signal-to-noise statement","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The quantity n_eff is introduced ('COBE DMR should find neff ≈ 1.1 for a Peebles-Harrison-Zeldovich n = 1 universe') without definition. A clause identifying it as the effective spectral index after radiative transfer in the standard CDM model would help the non-specialist reader.","section":"§1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The dipole is quoted only as an amplitude in mK and a direction. Giving the corresponding Solar System velocity (or the equivalent temperature in km s⁻¹) would allow direct comparison with the FIRAS value and with the pre-COBE dipole literature.","section":"§3 (Dipole ℓ = 1)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The companion papers are listed inconsistently as 'to be submitted' (Banday et al. 1996a) and 'submitted' (the others). Please standardize the status designations, since several load-bearing analyses in this Letter (systematic error budget, Galactic template fitting, Gaussianity tests) are deferred to these manuscripts.","section":"References","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"minor_revision","confidential_remarks":"This is a summary Letter in a coordinated series, and the load-bearing methodological analyses (mapmaking and systematic error budget in Kogut et al. 1996a, Galactic template fitting in Kogut et al. 1996b, Gaussianity tests in Kogut et al. 1996c, power-spectrum estimators in Górski et al. 1996, Hinshaw et al. 1996a,b, Wright et al. 1996) all reside in manuscripts listed as submitted. That is normal practice for a Letter series, but the editor may wish to coordinate the review of the companions so that the systematic-error and foreground-modeling claims can be assessed at the same time. My major comments concern how the headline uncertainties are quoted and one internal consistency check on the quadrupole error budget; neither challenges the central result, which is robust across all three foreground treatments tabulated."},"author_rebuttal":null,"desk_editor":{"model":"selfhosted/deepseek-v4.1-flash","letter":"Bottom line: this is the paper you cite for the pre-WMAP large-angle numbers, and it earns that. The 4-year maps, the DIRBE 140 µm custom cut (3881 of 6144 pixels, with the trade-off curve shown in Fig. 3), the fitted dust (2.7 ± 1.3 µK) and free-free (7.1 ± 1.7 µK) template amplitudes, and the Gaussianity limits are all new relative to the 1- and 2-year analyses the paper cites. Table 2 is the strongest part: four independent estimators crossed against three foreground treatments, agreeing within errors, with cosmic variance explicitly included and systematics bounded at 6 µK (95%). The quadrupole's Galactic modeling error (7.1 µK, dominating its 3.6 µK statistical error) is disclosed rather than buried, which is the right instinct. The Gaussianity claim is tested against the 3-point function, genus, and extrema correlations calibrated on Monte Carlo realizations, so it is not forced by the fit — the reader is right that circularity is low here, and the n_eff ≈ 1.1 expectation is imported from cited standard CDM, not fitted. Where I'd push: the abstract's n = 1.2 ± 0.3 and Q = 15.3 (+3.8/−2.8) sit in the DIRBE-template rows, and the internal-combination rows — the ones built to cancel free-free — give n = 1.00–1.11 with errors near ±0.4, plus a 1.62 outlier from Hauser-Peebles. Foreground treatment moves the central n by 0.1–0.2 and widens the interval by roughly half. §3's one-liner that this reflects 'data selection' rather than analysis technique is an assertion, not a test. That said, the stress-test framing overstates it: the abstract's Q errors (+3.8/−2.8) are actually a touch wider than the tightest DIRBE-template row (+3.7/−2.6), so the quoted bar is not cherry-picked from the best case, and 1.2 versus 1.0 is well inside 1σ on the printed numbers. This is an error-bar representativeness question, not a refutation. The other honest soft spot is structural: the load-bearing analyses live in eight companions, most listed as 'submitted' or 'to be submitted' — Banday 1996a is 'to be submitted.' That is the Letter genre, and the maps, time-ordered data, and custom cut are public through NSSDC, which is more than most 1996 papers offer. Who it's for: anyone who needs the canonical large-angle normalization and Gaussianity reference, and anyone teaching how to present a foreground systematic honestly. Worth a serious referee — Table 2 is transparent enough that a referee can actually audit the spread rather than take the sentence in §3 on faith.","headline":"The canonical pre-WMAP large-angle numbers, backed by a genuinely transparent four-estimator, three-foreground-treatment table — the only real soft spot is that the quoted error bar is a bit tighter than the method-to-method spread the paper itself tabulates.","tokens_in":12720,"tokens_out":2347,"would_cite":true,"duration_ms":46531,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"pith_extraction":{"msc":[],"pacs":["98.70.Vc","98.80.Es"],"model":"selfhosted/deepseek-v4.1-flash","headline":"The four-year COBE maps turn the cosmic microwave background into a measured spectrum: a nearly scale-invariant, Gaussian field of primordial ripples with amplitude about 30 microkelvin.","keywords":["cosmic microwave background","COBE DMR","angular power spectrum","spectral index","quadrupole normalization","Gaussian statistics","Galactic foreground removal","full-sky maps"],"falsifier":"A foreground template that is wrong in shape rather than amplitude would show up as a spatial cross-correlation between the surviving DMR pixels and an independent synchrotron or dust map at higher sensitivity; detecting such a residual at a level comparable to the quoted $10 pm 4~mu$K foreground budget, or finding that a different defensible Galactic cut moves $n$ or $Q_{rms-PS}$ outside the quoted errors, would undercut the fit.","tokens_in":12793,"feed_emoji":"🌌","tokens_out":2903,"duration_ms":70793,"temperature":0.7,"pith_summary":"This Letter asks whether the faint temperature ripples in the cosmic microwave background, measured over four full years by the COBE DMR instruments, look like the kind of primordial fluctuations that inflationary and Harrison-Zeldovich models predict. The authors combine all six radiometer channels at 31.5, 53, and 90 GHz, cut away the Galactic plane using a dust map as a guide, and fit a power law to the surviving sky. They find a spectral index $n = 1.2$ with an uncertainty of $0.3$, so the data are consistent with the scale-invariant case $n = 1$, and a quadrupole normalization $Q_{rms-PS} = 15.3^{+3.8}_{-2.8}$ microkelvin, or $18.0$ microkelvin if $n$ is fixed at 1. They also report that the pattern passes several tests for Gaussian random statistics. If this holds, the largest observable scales in the universe carry a spectrum and a probability distribution matching the simplest early-universe pictures, which is what makes the result worth a public page: it is the first full-sky, multi-frequency measurement of the shape of the primordial seed field, not just a detection that ripples exist.","feed_headline":"Four years of COBE data fit a scale-invariant early universe","feed_subtitle":"The full multi-frequency sky gives a spectral index of 1.2 and a quadrupole near 15 microkelvin, with Gaussian statistics holding.","key_machinery":"The load-bearing mechanism is a multi-frequency likelihood analysis on a cut sky. Six differential radiometers at three frequencies observe the same sky, so foreground emission with a non-thermal or dust-like frequency dependence can be separated from a thermal CMB signal by its spectrum; the DIRBE $140~mu$m map supplies an independent spatial template for where the Galaxy is worst, allowing a cut that keeps more sky than a blunt $|b| < 30^circ$ cut would. On the surviving pixels the team evaluates a Gaussian likelihood for the spherical-harmonic amplitudes, either through the two-point correlation function, through orthonormal modes constructed on the cut sphere, or through a full pixel-pixel covariance matrix. Cosmic variance, $sigma(C_ell)/C_ell approx sqrt{2/(2ell+1)}$, is folded in alongside instrument noise, which is why the low multipoles carry large error bars and why the quadrupole can sit below the full-spectrum normalization without contradiction.","core_discovery":"The central claim is that the 4-year COBE DMR sky, once the Galaxy is removed, is described by a scale-invariant power-law angular power spectrum with Gaussian statistics. Using a custom Galactic cut guided by the DIRBE $140~$micron map, which leaves 3881 of 6144 pixels, and correcting for dust and free-free emission traced by the same data, the authors fit a spectral index $n = 1.2 pm 0.3$ and a quadrupole normalization $Q_{rms-PS} = 15.3^{+3.8}_{-2.8}~mu$K; at fixed $n = 1$ the normalization is $18 pm 1.6~mu$K. The quadrupole alone is measured at $Q_{rms} = 10.7 pm 3.6 pm 7.1~mu$K, lower than the full-spectrum fit but consistent with it given cosmic variance. Three-point correlations, correlations of temperature extrema, and the topological genus of the maps all agree with a random-phase Gaussian field, and non-Gaussian toy models with $1 < N < 60$ are about five times less likely. The same data yield a monopole temperature $T_0 = 2.725 pm 0.020$ K and a dipole of $3.353 pm 0.024$ mK toward $(l,b) = (264.26^circ pm 0.33^circ, 48.22^circ pm 0.13^circ)$.","pith_inferences":["Because the quadrupole's Galactic modelling error, $7.1~mu$K, is twice its statistical error, the single most informative improvement to this analysis is a better dust and free-free template rather than more integration time; a lower-frequency foreground survey would attack the same number directly.","The stated consistency between the 1-year, 2-year, and 4-year normalizations is a weak test of systematics, since all three use overlapping time-ordered data; an independent instrument at the same frequencies would be the sharper check.","If the low observed quadrupole is a real cosmological feature rather than cosmic variance, it points to suppressed large-angle power, which later full-sky polarization and temperature experiments could confirm or retire.","The five-to-one preference for Gaussian statistics is a modest likelihood ratio on a single sky; it constrains but does not close the space of mildly non-Gaussian models, and the paper's own framing treats it as a consistency check rather than a detection."],"forward_implications":["The Harrison-Zeldovich case $n = 1$ survives: the measured index is within one sigma of scale invariance, so inflationary models that predict a nearly flat spectrum are not excluded by the largest angular scales.","The normalization $Q_{rms-PS}|_{n=1} = 18 pm 1.6~mu$K becomes the anchor that ties the amplitude of the CMB ripples to the amplitude of matter clustering in structure-formation models.","The Gaussianity tests place a quantitative bound on exotic alternatives: random-phase chi-squared models with $1 < N < 60$ are disfavoured by roughly five to one relative to the Gaussian case.","The maps and the custom Galactic cut are released publicly, so the same sky cut and noise properties can be reused by later experiments and reanalyses.","The monopole and dipole values from DMR agree with the independent FIRAS measurements, cross-checking the absolute temperature scale and the Solar System's motion."],"supporting_citations":[{"why":"The original 1-year detection of CMB anisotropy that this 4-year analysis extends and re-tests.","marker":"Smoot et al. 1992"},{"why":"Supplies the DMR calibration and its uncertainty, which sets the absolute scale of every reported amplitude.","marker":"Bennett et al. 1991"},{"why":"Provides the upper limits on residual systematic errors that the 4-year map-making inherits and tightens.","marker":"Kogut et al. 1992"},{"why":"The map-making, data-editing, and monopole-temperature paper that the sky maps and $T_0$ value rest on.","marker":"Kogut et al. 1996a"},{"why":"Derives the dust and free-free Galactic templates and the quadrupole fit, which carry the foreground-removal argument.","marker":"Kogut et al. 1996b"},{"why":"Performs the three-point function, extrema-correlation, and genus tests behind the Gaussian-statistics claim.","marker":"Kogut et al. 1996c"},{"why":"The pixel-pixel covariance likelihood analysis, one of the independent power-spectrum routes summarized in Table 2.","marker":"Hinshaw et al. 1996a"},{"why":"Constructs orthonormal functions on the cut sphere and evaluates the exact Gaussian likelihood in that basis.","marker":"Gorski et al. 1996"},{"why":"The Hauser-Peebles style angular power spectrum on the cut sphere, the third independent fit listed in Table 2.","marker":"Wright et al. 1996"},{"why":"The two-point correlation function analysis, which supplies the $n = 1$ normalization from that statistic.","marker":"Hinshaw et al. 1996b"}],"fun_headline_variants":["COBE's four-year sky fits scale-invariant Gaussian statistics","COBE DMR measures a spectral index of 1.2 and a 15.3 μK quadrupole","The COBE quadrupole is lower than the full fit but consistent","Low signal-to-noise COBE patches still support a Gaussian sky"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The argument depends on the surviving high-latitude sky being contaminated at well under the level of the cosmic signal after the DIRBE-guided cut and the fitted dust and free-free templates are applied, and the authors themselves note that moderate changes to the cut shift the derived parameters.","fun_headline_variants_meta":{"raw":{"variants":["COBE's four-year sky fits scale-invariant Gaussian statistics","COBE DMR measures a spectral index of 1.2 and a 15.3 μK quadrupole","The COBE quadrupole is lower than the full fit but consistent","Low signal-to-noise COBE patches still support a Gaussian sky"]},"model":"selfhosted/deepseek-v4.1-flash","effort":"low","attempts":1,"cost_usd":0.0,"raw_usage":{"total_tokens":3175,"prompt_tokens":1134,"completion_tokens":2041,"prompt_tokens_details":null,"completion_tokens_details":{"reasoning_tokens":1967}},"tokens_in":1134,"tokens_out":2041,"duration_ms":32482,"image_pages":[],"temperature":1.0,"reasoning_budget":50,"reasoning_tokens":1967,"transport_provider":"selfhosted","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-10-29T19:21:42.591579+00:00","model_set":{"reader":"selfhosted/deepseek-v4.1-flash"},"falsifier":"A foreground template that is wrong in shape rather than amplitude would show up as a spatial cross-correlation between the surviving DMR pixels and an independent synchrotron or dust map at higher sensitivity; detecting such a residual at a level comparable to the quoted $10 pm 4~mu$K foreground budget, or finding that a different defensible Galactic cut moves $n$ or $Q_{rms-PS}$ outside the quoted errors, would undercut the fit.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The original 1-year detection of CMB anisotropy that this 4-year analysis extends and re-tests."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the DMR calibration and its uncertainty, which sets the absolute scale of every reported amplitude."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the upper limits on residual systematic errors that the 4-year map-making inherits and tightens."},{"cited_title":"L., et al.\\ 1996, ApJ Letters, submitted","cited_arxiv_id":null,"evidence_quote":"The Hauser-Peebles style angular power spectrum on the cut sphere, the third independent fit listed in Table 2."}],"review_version":1}