Pre-registration: a Landauer principle for spacetime's information production, and the dark-energy history it fixes

A timestamped prediction, made before the next DESI, Euclid and Roman cosmology releases. THE PRINCIPLE. The negative pressure of the dark sector does work on the expansion exactly equal to the Landauer cost of the spacetime information produced by that expansion: −p dV = T dS_info, with dS_info = k_B ln2 · Γ · dV₄. It rests on four inputs: (i) Landauer's bound, k_B T ln2 per bit, paid to the surroundings (measured); (ii) the open-system first law of Prigogine et al. (1989), with a creation enthalpy of k_B T ln2 per created bit; (iii) one postulate: production at a constant rate per unit four-volume. This is the counting law of causal-set theory, which supplies the FORM ONLY; the required Γ is 2.8×10⁻¹⁵² of one element per Planck four-volume; (iv) one choice: the price is paid at the temperature of a relic thermal bath, T ∝ 1/a. Paying it at the horizon temperature instead reproduces ΛCDM exactly. There is one normalisation (fixed by Ω_DE0) and no shape parameter. THIS IS NOT A NEW CLASS OF COSMOLOGY. At the background level the model is a member of the CCDM / open-system particle-creation family. It is also equivalent to a decaying vacuum Λ(t) whose density is proportional to the comoving Hubble radius 1/(aH); this was verified numerically to 10⁻⁷. The horizon variant is the Lima–Jesus–Oliveira "dark degeneracy" model. Only the rate law, Q ∝ T_CMB(a), is new: no published use of it was found. THE REGISTERED PREDICTION (Ω_DE0 = 0.6835). w₀ = −0.9354. CPL (w₀, w_a) = (−0.935, −0.326). A w = −1 crossing at z = 0.276. Matter-era w = −7/6. The full w(z) and ρ(z) tables are in the document. FALSIFICATION CRITERIA (symmetric, fixed now). A, CPL basis: refuted if Δχ² ≥ 11.83; passed if Δχ² ≤ 6.18. B, crossing: refuted at ≥ 3σ if a crossing in 0.15 < z < 0.45 is excluded, or if the crossing is measured outside that window. C, matter era: refuted if a binned w at 1 < z < 2.5 is inconsistent with −1.10 to −1.16 at ≥ 3σ. A ΛCDM-consistent measurement refutes the model. No switch of bath, exponents or normalisation will be made after the fact. PROVENANCE, NOT OPTIONAL. The model was built on 8 October 2026, after DESI DR2, the supernova samples and Planck were known. Its choices were made knowing the data's preferred direction. Everything in section 4 of the document is therefore recorded as what was known, NOT as evidence. Δχ² against ΛCDM, with no extra parameter, for Pantheon+ / DES-Dovekie / Union3: compressed CMB (θ*, R): −8.42 / −9.58 / −8.72; full Planck 2018 TT,TE,EE (plik-lite, no lensing): −6.65 / −7.95 / −6.86. On the full spectrum the model trails the two-parameter CPL fit by 0.94σ / 1.61σ / 2.48σ. S₈ = 0.820, which is 0.012 above ΛCDM; the model does not relieve S₈. A post-hoc measurement of the two exponents finds that the energy source scales as a^−(1.05 to 1.10) ± 0.08–0.10 in the matter era, consistent with T ∝ 1/a. A weak blind test on SDSS DR16 BAO came out in the predicted direction: Δχ² = −1.27, likelihood ratio 1.9. The prediction was sealed before the data were used (SHA-256 87e4888107ae562299b908010a8937615229714cbdf240d84a7b03f44ef5ab6a). RELATION TO EARLIER RECORDS. Record 10.5281/zenodo.22972538 registered λ = √3/2 and, as post-hoc, "mechanism B" (w₀ = −0.937, crossing at z = 0.348). This record replaces neither. H* shares mechanism B's counting law and is nearly degenerate with it: the two are 0.8σ apart at Roman/Euclid precision. A result in that region therefore supports the shared counting law, not H*'s pricing. LABORATORY. The principle places the cost of information in the surroundings. It therefore predicts a NULL result for the single-electron Penning comparison registered in 10.5281/zenodo.22972538, and the two records are registered rivals for that measurement. FILES. PREREGISTRATION_Hstar_2026-10-09.md: the registered document (SHA-256 447766143a5b841bbed1b23f096b2ed306142dd60b60f990459104734d2c77b9). hstar_core.py and hstar_registered_table.py: these reproduce HSTAR_REGISTERED_OUTPUT.txt and HSTAR_REGISTERED_TABLE.csv. known_data_comparison.zip: scripts and outputs for sections 1.1 and 4, including the sealed SDSS prediction and data-provenance notes. SHA256SUMS.txt: checksums of all of the above.