QUBIC · Agent2 · 28 September 2026

From sky to timelines.
From timelines to maps.

Independent implementation, actual telescope data, and explicit numerical checks.

QUBIC_JUN26 left unchanged

Two hidden problems explained the earlier results.

Noise reproducibility

QUBIC resets the random seed inside get_noise(). The earlier external seed did not produce matching noise.

Electrical units

The installed reader used microamps in an amperes formula. Agent2 converts to SI locally.

Fair comparisons

The original and supplied split notebooks used different frequencies and sizes. Controlled runs now match settings.

Honest scope

Real-data outputs are intensity diagnostics with calibration limits and recorded convergence.

Identical final maps in controlled runs.

FI220 and FI150 at nside64; TD150 at nside32. Each split run independently regenerated TOD, then reloaded its full pointing from disk.

0 µK maximum FI220 cross-pipeline difference

All three comparisons: bit-for-bit identical maps. TD solver converged: True. Equality alone is not a sky-recovery test.

FI220 I Q U cross-pipeline residuals are zero

Matching pipelines can still have reconstruction residuals.

FI220 simulation: input, reconstructed map and residual, for I/Q/U. Residuals use a separate, tighter color scale.

Input sky versus reconstructed sky and residuals

248 detectors. Every sample equals 1.

565,186 native samples per TES at 156.996 Hz, represented exactly. Groups of 40 are averaged for the map. PCG residual: 9.67e-05.

A deliberately unrealistic 1 W plumbing test. A flat detector signal need not reconstruct to a uniform sky.

Map reconstructed from one-hour constant TOD

A real TD intensity quick-look.

Real TD uncalibrated intensity map

24 March 2026, 07:25 scan. 10,231 averaged samples; PCG 460 iterations, residual 9.18e-05.

Not a calibrated sky measurement. About 98% of offset-removed TOD remains unexplained by this model. HWP is absent; only I is fitted under Q=U=0.

Open pipeline_2_tod_to_map_example.ipynb, or run python real_data_to_map.py DATASET. Maps, PNGs, coverage and JSON are saved together.