Extract Pure-localization covariance/uncertainty through params

[MarcelSoler]

I am trying to assess localization uncertainty using cartographer internal data.
Is this is feasible or just impossible?

My idea was to check:

• Score histogram: Count: 289 Min: 0.500677 Max: 0.892239 Mean: 0.702535.
• Constrains added: n computations resulted in n-2 additional constraints.
• Number of residuals:

Residual blocks                           486                      486
Residuals                                1458                     1458

• Solver Summary from Ceres:

 ...
Termination:                      CONVERGENCE (Function tolerance reached. |cost_change|/cost: 6.656970e-08 <= 1.000000e-06)`

Any answer or help would be appreciated.
Thank you!