ForwardKalmanFitter

• create new internal per-node code structure: 'predict()' -> 'buildAndAnalyse()' -> 'updateOrSkip()'
• migrate and test MBase interface in new internal structure
• migrate and test PRD interface in new internal structure
• pass Piece-wise DAF through configure
• pass runOutlier through interface
• call piece-wise DAF in 'buildAndAnalyse' part if node is start of drift circle section
• drop momentum-overshooting-stabiliser in 'buildAndAnalyse' (obsolete with 14.4.0 layer properties)
• investigate same-surface error message on min-bias sample (when tags are in nightlies)

DAF-on-Demand-Fitter

• add meaningful EDM screen dump to CompetingROT classes
• write simplified L/R single-hit CompROT creation
• add protection against singular weight matrices to Pixel+SCT CompROTs
• write piecewise DAF acting on a branched Trk::Trajectory
• adjust object flow for both callers (FKF-driftcircles and Seed-recovery)
• integrate simplified single-hit CompROT creation back into official tools
  (current version doesn't go with coding rules)
• replace Tomer's seed recovery with piecewise DAF
• solve paradox of full assignment probabilty but bad track chi2
  
  • solution here are reduced weight-cut values for biased track parameters
  • alternative is unbiased track parameters, but this is currently not enough validated in all DAFs.
• write CompetingMDT_DriftCirclesOnTrack (postpone)

Performance tuning

Outlier logic

• add 5-sigma prediction-on-sensor check to FKF (shown to improve stability)
• decrease priority of sensor check in KOL if chi2 is bad (increases chance of convergence in outlier-logic)
• look closely into TRT-only fits (evt. postpone)

Improvement from Annealing

• annealing at entering of TRT too costly for marginal improvement
  Annealing kicks in in 50-60% of tracks to turn 0.5 tube hits per track to precise drift circles
• annealing as replacement for seed-recovery logic maintains efficiency

Broad vs Precise hits

• enum denoting calibration states in TrkFitterUtils
  Trk::TrackState CalibrationNotKnown, Broad/TubeHit, Nominal, HighPrecision
• enhance IMeasRecalibrator interface
• implement: give all ROT creators to measurement recalibrator,
  and reproduce existing results
• implement: create method for broad-rot making
• ship to and use it in all three(!) places: FKF, KOL and new CompROT->ROT shrinking
• extend its use: SCT in KOL, all in piecewise DAF
• extend its use to TrkDeterministicAnnealingFilter (postpone)

Code clean-up

• do Measurement-ID at trajectory-building state
• export MeasTypeID helper to TrkEventUtils
• export Prediction-within-sensor test to TrkFitterUtils
• move interfaces for competing-ROT tools from tool packages to TrkToolInterfaces
• identify precise/broad calibration at sub-detector level, not with simple assumptions
• split O(1000) lines of flagNewOutlier mess-thod into two: generic removal and indet-specific recovery
• replace all matEffects, doDNA, breakpoint etc parameters by new KalmanMatEffectsController in TrkFitterInterfaces
• migrate entire progressive-filter suite to private use-statements
• structurise piecewise DAF
• apply IDSW cleanup changes

Electrons and DNA

• use new filter-reseeding logic for iterations after denied brempoint
• adopt same ParticleHypothesis interpretation as GSF
• enhance DNA interface to steer multipurpose vs aggressive method
• add DNA to set of tools handled by the DeterministicAnnealingFilter
• update DAF to call FKF with new flag runOutlier=false
• test DAF extensions with DNA on electrons (evt. postpone)

Tests

• ttbar50 tests on 14.2.0. Tests on 14.2.23 impaired (material effects)
• ttbar50 tests on 14.4.0 (fix new crashes, get new numbers)
• valgrind test: no new leaks on 10 ttbar with full recovery+piecewise daf
• full scale RTT/deployment test in 15.0.0

Package bundles:

(1) CompetingROT data model

(2) track fitters ( before author migration)

(3) track fitters ( after author migration and AthAlgTools)