Planck NPIPE: Final Unified Planck Data Release

• Planck NPIPE is a unified reprocessing framework that combines LFI and HFI data to produce calibrated full‐sky maps with reduced noise and systematics.
• It employs advanced time-domain modeling and short-baseline destriping to efficiently correct for instrumental effects in temperature and polarization measurements.
• The PR4 release enhances analyses of CMB lensing, SZ effects, and parity tests, providing a robust benchmark for ΛCDM and extended cosmological models.

Planck NPIPE, also referred to as Planck Public Release 4 (PR4), is the final joint reprocessing of the Planck mission time-ordered data from the Low Frequency Instrument and High Frequency Instrument into calibrated frequency maps in temperature and polarization, together with detector-set products, low-resolution data products, and 600 end-to-end simulations (Collaboration et al., 2020). It was designed as a unified LFI+HFI framework that combines features of the earlier instrument-specific pipelines while lowering noise and systematics at essentially all angular scales and improving internal consistency between frequency channels (Collaboration et al., 2020). In current Planck-based cosmological practice, NPIPE functions both as a data release and as a methodological benchmark for tests of $\Lambda$CDM, CMB lensing, secondary-anisotropy reconstruction, and parity-violating or modified-gravity extensions.

1. Release context and defining characteristics

Planck PR4/NPIPE is the latest Planck data release, following PR1 (2013), PR2 (2015), and the PR3 “legacy” release of 2018. In contrast to the earlier public releases, which relied on separate LFI and HFI processing chains, NPIPE reprocesses all Planck time-ordered data in a single coherent pipeline and produces calibrated full-sky maps across the nine Planck frequency channels from 30 to 857 GHz (Collaboration et al., 2020).

The stated design goal was not merely another map release, but a re-implementation of most of the DPC Level-2 processing with improved calibration, noise removal, and systematic corrections, executed on high-performance computing systems (Collaboration et al., 2020). The resulting maps and simulations are intended to support both direct cosmological inference and downstream component separation. NPIPE maps also retain the Solar dipole in mapmaking, provide single-detector products for line-emission work, and include high-resolution HFI temperature maps at ${N}_{\rm side}=4096$ for 217–857 GHz (Collaboration et al., 2020).

A notable operational difference from PR3 is that NPIPE incorporates data acquired during satellite repointing manoeuvres, corresponding to about 8% more effective observation time. This directly improves map noise, and in lensing applications it is one contributor to the higher signal-to-noise ratio of PR4 products relative to PR3 (Carron et al., 2022).

2. Processing architecture and mapmaking formalism

NPIPE processing is organized into local preprocessing and global reprocessing. Local preprocessing comprises single-detector, single-pointing-period operations such as ADC corrections, glitch handling, removal of 1 Hz spikes and harmonics in LFI, sky-load differencing, thermal decorrelation, 4 K line removal, and transfer-function deconvolution for HFI optical bolometers (Collaboration et al., 2020). Global reprocessing then fits a large template set jointly across detectors and frequencies.

The core time-domain model is

$$\vec d = \mathbf{P}\,\vec m + \mathbf{F}\,\vec a + \vec n,$$

where $\vec d$ is the multi-detector time-ordered data, $\mathbf{P}$ is the pointing matrix, $\vec m$ is the sky map, $\mathbf{F}$ is the template matrix, $\vec a$ are template amplitudes, and $\vec n$ is the residual noise. In the generalized destriper formulation used by NPIPE, the template amplitudes are solved from

$$\vec a = \left(\mathbf{F}^{\mathrm T}\mathbf{N}^{-1}\mathbf{Z}\mathbf{F}\right)^{-1}\mathbf{F}^{\mathrm T}\mathbf{N}^{-1}\mathbf{Z}\vec d,$$

with

${N}_{\rm side}=4096$0

This template space includes short destriping baselines, gain fluctuation templates, ADCNL distortion templates, orbital dipole, far-sidelobe pickup, transfer-function residuals, zodiacal components, bandpass mismatch templates, and foreground polarization templates at CMB frequencies (Collaboration et al., 2020).

For mapmaking, NPIPE uses Madam destriping with 167 ms baselines. This very short-baseline destriping is central to the suppression of striping and low-frequency noise and is part of the reason NPIPE maps show reduced large-scale striations and improved detector-set consistency relative to earlier Planck products (Collaboration et al., 2020). For polarized horns, NPIPE also enforces horn symmetrization, effectively solving polarization from detector differences and thereby reducing temperature-to-polarization leakage from beam mismatch and sub-pixel structure (Collaboration et al., 2020).

At CMB frequencies, calibration is partially degenerate with large-scale polarization. NPIPE breaks this degeneracy with polarized foreground priors built from 30, 217, and 353 GHz information, depending on channel, during calibration iterations (Collaboration et al., 2020). A direct consequence is a non-trivial large-scale polarization transfer function. For low multipoles, the CMB polarization signal is partially filtered, and low-${N}_{\rm side}=4096$1 analyses must therefore correct either with the supplied transfer functions or with matched simulations (Collaboration et al., 2020).

3. Data products, splits, and simulation suite

NPIPE releases full-mission maps of Stokes ${N}_{\rm side}=4096$2, ${N}_{\rm side}=4096$3, and ${N}_{\rm side}=4096$4 at all nine Planck frequencies, together with half-ring maps, detector-set “A/B” splits, single-detector maps, single-horn maps, low-resolution maps, and pixel-pixel noise covariance matrices (Collaboration et al., 2020). In practice, the A/B detector-set splits are especially important because they furnish nearly independent noise realizations for cross-spectra and null tests; this split structure is used extensively in PR4 lensing and birefringence analyses (Diego-Palazuelos et al., 2022).

The simulation suite comprises 600 complete signal, noise, and systematics realizations of the full-frequency and detector-set maps, and, as a Planck first, these simulations include full time-domain processing of beam-convolved CMB anisotropies (Collaboration et al., 2020). They propagate CMB, foregrounds, scanning, beam convolution, instrumental noise, gain fluctuations, ADCNL, transfer-function residuals, and bandpass mismatch through the NPIPE processing chain (Collaboration et al., 2020). This simulation set underpins transfer-function estimation, covariance construction, low-${N}_{\rm side}=4096$5 likelihood building, and systematics validation across later PR4 studies.

For polarization, the large-scale transfer function is a defining feature. In the NPIPE low-${N}_{\rm side}=4096$6 polarization analysis, the quadrupole and octupole are substantially suppressed, while the transfer function approaches unity by ${N}_{\rm side}=4096$7 (Collaboration et al., 2020). This filtering does not invalidate the data; rather, it makes simulation-based calibration obligatory for analyses of the largest angular scales.

The NPIPE design also improved map-level usability for component separation and secondary-anisotropy reconstruction. In thermal Sunyaev–Zeldovich work, PR4 provides nine full-mission sky maps plus half-ring splits for each channel, enabling NILC-based ${N}_{\rm side}=4096$8-map reconstruction over 98% of the sky after masking only the brightest 2% of 857 GHz pixels along the Galactic ridge (Chandran et al., 2023).

4. Likelihoods and cosmological parameter inference

In later cosmological analyses, the standard PR4 high-${N}_{\rm side}=4096$9 choice is the CamSpec likelihood built from NPIPE maps. In that usage, “PR4 reference cosmology” typically means NPIPE maps analyzed with CamSpec, whereas the legacy PR3 reference cosmology is based on legacy maps analyzed with Plik (Jense et al., 10 Oct 2025). A nuisance-marginalized CamSpec-NPIPE-lite likelihood was subsequently derived and released for joint analyses with ACT, SPT, and related datasets (Jense et al., 10 Oct 2025).

A central result of the PR3-to-PR4 comparison is that the additional constraining power of NPIPE comes from polarization at all scales and from temperature at multipoles above $$\vec d = \mathbf{P}\,\vec m + \mathbf{F}\,\vec a + \vec n,$$0 (Jense et al., 10 Oct 2025). For Planck-only analyses that use TE/EE across the full multipole range or TT at high $$\vec d = \mathbf{P}\,\vec m + \mathbf{F}\,\vec a + \vec n,$$1, PR4 therefore constitutes the natural endpoint of Planck primary-anisotropy inference. By contrast, once Planck is truncated to the large-scale range used in combination with ground-based data, PR3-versus-PR4 differences become small and extended-model constraints are effectively insensitive to the choice of Planck maps and likelihood (Jense et al., 10 Oct 2025).

Several extension studies illustrate how NPIPE altered the interpretation of earlier Planck anomalies. In parametric tests of gravity, reanalysis with PR4 anisotropy spectra from NPIPE maps shows that the previous PR3 preference for nonzero $$\vec d = \mathbf{P}\,\vec m + \mathbf{F}\,\vec a + \vec n,$$2 and $$\vec d = \mathbf{P}\,\vec m + \mathbf{F}\,\vec a + \vec n,$$3 is significantly reduced, with updated CamSpec and HiLLiPoP likelihoods bringing the data into much better agreement with General Relativity and a $$\vec d = \mathbf{P}\,\vec m + \mathbf{F}\,\vec a + \vec n,$$4CDM cosmology; the shift is closely tied to the mitigation of the Planck lensing anomaly (Specogna et al., 2024). In axion-like early dark energy analyses, a new high-signal-to-noise NPIPE likelihood yields $$\vec d = \mathbf{P}\,\vec m + \mathbf{F}\,\vec a + \vec n,$$5 without SH0ES and no evidence in favour of a significant EDE component, leaving a residual tension of $$\vec d = \mathbf{P}\,\vec m + \mathbf{F}\,\vec a + \vec n,$$6 with the SH0ES Cepheid-based $$\vec d = \mathbf{P}\,\vec m + \mathbf{F}\,\vec a + \vec n,$$7 measurement (Efstathiou et al., 2023).

NPIPE also sharpened Planck’s large-scale polarization constraints on primordial tensors. Using PR4 maps, the BB spectrum over $$\vec d = \mathbf{P}\,\vec m + \mathbf{F}\,\vec a + \vec n,$$8–150 yields $$\vec d = \mathbf{P}\,\vec m + \mathbf{F}\,\vec a + \vec n,$$9 at 95% confidence on more than 50% of the sky; combining EE, BB, and EB gives $\vec d$0; and adding Planck TT further tightens the Planck-only bound to $\vec d$1, with Planck plus BICEP2/Keck 2015 giving $\vec d$2 (Tristram et al., 2020).

5. CMB lensing, thermal SZ, and other derived products

PR4/NPIPE enabled a new generation of Planck-derived secondary-anisotropy products. In CMB lensing, quadratic estimators applied to NPIPE maps with more optimal filtering increase the reconstruction signal-to-noise by almost 20% relative to the 2018 PR3 release (Carron et al., 2022). The amplitude of the CMB-marginalized lensing power spectrum is then constrained to $\vec d$3 in units of the Planck 2018 best fit, and lensing alone gives $\vec d$4 with weak priors and element-abundance observations (Carron et al., 2022). With BAO, the same PR4 lensing reconstruction yields $\vec d$5, $\vec d$6, and $\vec d$7 (Carron et al., 2022).

The improved PR4 lensing map has also been used in cross-correlation cosmology. A tomographic cross-correlation between Planck PR4 lensing and DESI DR1 quasars detects the cross-spectrum at signal-to-noise ratio 21.7 and the quasar auto-correlation at 27.2, and, with a DESI DR1 BAO prior, yields $\vec d$8 and $\vec d$9 (Belsunce et al., 27 Jun 2025). When combined with ACT DR6 lensing auto-spectra, the same framework furnishes a sound-horizon-free estimate $\mathbf{P}$0 (Belsunce et al., 27 Jun 2025).

In thermal SZ science, NPIPE underlies an improved all-sky Compton-$\mathbf{P}$1 map built from the nine PR4 frequency maps with a tailored NILC pipeline. The PR4 $\mathbf{P}$2-map covers 98% of the sky and shows reduced large-scale striations, lower residual Galactic dust near the plane, about 7% lower residual thermal-noise contamination, and about 34% lower residual CIB contamination at small angular scales relative to the previous PR2 product (Chandran et al., 2023). The same PR4 dataset was also used to recalibrate Planck SZ observable–mass relations in a Bayesian framework including relativistic SZ effects; those corrections shift inferred integrated Compton-$\mathbf{P}$3 at the 5–15% level and can matter at up to approximately $\mathbf{P}$4 for the hottest clusters (Perrott, 2024).

6. Parity tests, anomalies, and scientific legacy

NPIPE’s improved polarization maps and end-to-end simulations made PR4 a natural dataset for parity-violation searches. In harmonic-space birefringence analyses of nearly full-sky PR4 polarization, the initial result is a birefringence angle $\mathbf{P}$5, but the measured value decreases as the Galactic mask is enlarged, consistent with contamination from polarized foreground EB (Diego-Palazuelos et al., 2022). Two independent foreground-mitigation strategies produce broadly consistent corrections, yet the conclusion remains cautious: cosmological significance is not assigned to the measured value of $\mathbf{P}$6 until foreground polarization is better understood (Diego-Palazuelos et al., 2022). A later map-space PR4 analysis finds $\mathbf{P}$7 for SEVEM maps and $\mathbf{P}$8 for Commander maps, but the dominant systematic remains the Planck polarimeter-angle calibration, and there is no evidence for a birefringence dipole (Sullivan et al., 11 Feb 2025).

Temperature-only anomaly studies using NPIPE likewise support a nuanced interpretation. A multiscale topological analysis of NPIPE temperature fluctuations with relative homology finds specific deviations in the number of loops at large smoothing scales, including a nominal $\mathbf{P}$9 excess at $\vec m$0, FWHM $\vec m$1 and $\vec m$2, although the relevant simulation distribution is manifestly non-Gaussian and not Poissonian (Pranav, 2021). The broader conclusion is conservative: notwithstanding persistent large-scale anomalies similar to those seen in WMAP and earlier Planck analyses, observations of the CMB are largely consistent with the standard cosmological model within $\vec m$3 (Pranav, 2021).

Taken together, these applications define the legacy of Planck NPIPE. It is the final Planck mapmaking and simulation framework, the basis of the PR4 cosmology, and the benchmark against which the stability of late Planck anomalies and extensions is now judged. A plausible implication is that NPIPE’s main scientific impact has been twofold: first, to reduce the dependence of Planck results on known mapmaking systematics through a unified LFI+HFI treatment and extensive end-to-end simulations; second, to show that several PR3-era hints for beyond-$\vec m$4CDM physics weaken when the same tests are repeated on PR4 products (Collaboration et al., 2020).