Skip to content

KKExtrap: process all requested surfaces in one time-ordered sweep per direction - #2084

Merged
brownd1978 merged 4 commits into
Mu2e:mainfrom
RobMina:kkextrap-sweep-base
Nov 8, 2026
Merged

brownd1978 merged 4 commits into
Mu2e:mainfrom
RobMina:kkextrap-sweep-base

Conversation

@RobMina

@RobMina RobMina commented Oct 30, 2026 •

Copy link
Copy Markdown
Contributor

KKExtrap: process all requested surfaces in one time-ordered sweep per direction

Motivation

KKExtrap runs each extrapolation request (tracker perimeter, upstream IPA/ST/TSDA/OPA, DS shells, passive planes,
CRV, calorimeter disks) as a separate pass, and each pass searches only the trajectory's current end piece. KinKal
can add a material effect only at the trajectory end, so any pass that extends past a surface belonging to a later
pass leaves that surface on an interior piece, where it is never found. The result depends on the order of the
requests: #2059 moved the calorimeter requests last to avoid one such case, and earlier fixes (merging CRV sectors
with their strongbacks, stopping the plane pass before the CRV) addressed others.

The DS shell pass also loses shells on its own. Each new search starts 1e-6 ns after the crossing just added, but
crossing times converge only to IntersectionTolerance, so the search can find the same shell again; the pass then
sees nothing new and stops, skipping every shell further out. On 10,000 nightly off-spill cosmic events
(CentralHelix, field correction on), 103 of 452 legs with a truth barrel traversal miss their outer shells.

Change

  • KKExtrap::extrapolate builds one target list per time direction from all enabled requests and processes them in
    a single sweep (ExtrapolateSweep.hh). One predicate searches every remaining target from the last processed
    crossing, across trajectory pieces in MaxDtStep windows, and returns the earliest crossing; the sweep adds it and
    continues. The same predicate drives Track::extrapolate, so each crossing is on the end piece when it is added.
    Request order therefore cannot change which surfaces are crossed. The sweep's concepts and flow are described at
    the top of ExtrapolateSweep.hh, and KKExtrap::targets lists what to change when adding passive material.
  • A processed target is searched again only once the track has left its intersection tolerance: at least four
    tolerances of path, and for a target with an incidence floor the time to move twice the tolerance along the normal
    at the last crossing's incidence. A plane target (passive plane, CRV sector, stopping-target foil) is also crossed
    again only after its signed distance has left the tolerance. So a shell is never re-found, the remaining shells are
    still searched, and a track that grazes a foil, slab or the IPA gets that material once.
  • Material targets have an incidence floor: a crossing with |cos| to the surface normal below it is not added (DS
    shells and IPA 0.1, stopping-target foils 0.02). A near-tangent contact would otherwise add a path many times the
    thickness; at a foil mirror point that is a momentum change of tens of GeV/c, which sends the extrapolation out of
    the field map. On 5,000 CE events no truth IPA crossing has |cos| below 0.2 and no truth foil crossing below 0.02.
  • Plane targets (passive planes, strongbacks, CRV sectors) are solved by bisection between extrema of the signed
    plane distance, which is monotonic between them for helix and line pieces; shallow excursions and returns are
    found, and a root outside the query window is never taken.
  • A root exactly on a piece boundary is placed one ULP inside the piece that owns it, so the material reference
    piece and Track::extrapolate's end-piece check agree.
  • Upstream: IPA and ST foils allow repeated crossings, TSDA ends the sweep, and after a reflection the sweep either
    returns to the tracker front (BackToTracker) or finishes the reflecting element.
  • t0 is clamped into the fitted range before use, and searches stay inside the trajectory range (the KKExtrap: clamp the calo search time into the trajectory range #2076 class).
  • A problem with one track found by the sweep (a search that cannot converge, a crossing the end piece does not own,
    more than 1000 crossings, an exception from KinKal while adding a crossing) ends that direction only, also when the
    sweep's predicate runs inside KinKal's field-corrected walk: the predicate records the problem and ends the walk, so
    the fit stays usable. The crossings already added are kept, the reason is recorded on the KKTrack, and the saved
    KalSeed gets the new TrkFitFlag bit ExtrapFailed ("ExtrapolationFailed"). When KinKal's own walk fails (for
    example a position outside the field map), KinKal marks the fit unusable as on main; the KalSeed then also gets
    ExtrapFailed, with KinKal's reason. Configuration and geometry-service errors still stop the job.
  • Calorimeter material (Offline#2081, ToCaloMaterial) is part of the downstream sweep: the front-panel layers and
    cooling pipes are plane targets (a pipe-plane crossing between the pipes adds nothing), and each crystal case inner
    support is found where the track enters the case, at its front or from the disk hole through the inner cylinder,
    with the second material placed where the track then reaches the carbon-fibre ring radius (material effects #2081's path-length
    rule). As in material effects #2081, this material starts at the last measurement, layers and rings already in the fit
    (AddCaloMaterial) are skipped, and the supports are entered only forwards in time. Each crossing is now added on
    the end piece as the sweep reaches it; material effects #2081's passes extend first and then intersect, and skip crossings that land
    on an interior piece of their own extension (on 50 reco_03 events, 133 skipped, among them most disk-1 inner-ring
    crossings). The calorimeter material also gets the range-out check.
  • KKFit::createSeed sets ExtrapFailed and RangedOut from the track alone, so a regrown track does not inherit
    them from its seed, and RegrowLoopHelix and RegrowKinematicLine call KKExtrap::checkConfig like the fit modules.
  • Range-out: forwards in time, a material crossing whose mean energy loss reaches the kinetic energy ends the
    direction before that material is added, since the particle stops there. The crossings already added are kept,
    and the saved KalSeed gets the new TrkFitFlag bit RangedOut (bit 29). Without this, KinKal's first-order
    material update turns the momentum around (|p + dmom| along -p̂) and the extrapolation runs back through the
    detector, reaching up to 172 GeV/c on nightly cosmics. In truth, 104 of the 106 matched muons in such legs range
    out, and none turns back in z while keeping its energy (magnetic mirror).
  • KKTrack records that the particle ranged out. The trajectory is not truncated at the stop: when the stop is
    found, KinKal has already extended the end piece by up to one field domain past it, and Track offers no way to
    shorten its fitted trajectory, so the last saved KalSegment can end up to one domain past the stop. A KinKal
    follow-up (trajectory truncation) will remove that; this PR needs no KinKal change beyond v3.9.1. The tracker
    surface sampling likewise runs past the stop (on 10,000 nightly cosmics, 9 sampled crossings on one track).
  • The eight request-specific predicate headers (Extrapolate{CRVRegion,Cylinders,IPA,Planes,ST,Surface, ToTrackerPerimeter,ToZ}.hh), used only by KKExtrap, are removed. KKExtrapConfig is unchanged.
  • New test executable ExtrapolateSweepTest (CMake and scons): tilted-field helix z extrema, rejected-first roots,
    exact piece-boundary material insertion in both directions, coplanar planes (each crossed), grazing planes and foils, and genuine returns, half-open window
    limits, the per-window reach bound, range-out against the material's energy loss, and search budgets. KKShellXing.hh gets the includes and namespace qualifiers it needs to compile alone.

Dependency

Requires KFTrack/KinKal#232, "Extend the end trajectory piece to its outermost field-domain edge before extrapolating"
(Track::coverEndDomain). With field correction on, the fitted trajectory can end before its last field domain,
and the domain walk starts at the domain edge, so crossings in between land on an interior piece where no material
can be added. That fix alone recovers the inner DS shells on main; this PR is validated on top of it.

KinKal v3.9.1 (envset p110, Offline#2100) contains it. Without it, the nightly off-spill validation (KKCHmu with field
correction on) would flag about 7 % of KKCHmu tracks ExtrapFailed (a DS-shell crossing not owned by the end piece).
The branch has main (b24d4163, with Offline#2081 and p110) merged; #2081's calorimeter material passes became sweep
targets (see Change).

Known limitation: overlapping passive-geometry planes (follow-up PR #2109)

Several KinKalGeomMaker passive-material planes overlap each other although the Geant4 volumes they stand for do
not (CI's g4surfaceCheck and rootOverlaps check only the Geant4 geometry). The sweep crosses every target, so in
the overlap bands this PR adds the material of both:

  • adjacent CRV sector envelopes overlap by 122.34 mm, because each spans all four layers, staggered by 42 mm (Run 2
    R, L and T sectors; Run 1A T1/T2). A muon in the band crosses layers of both sectors, so both sector crossings are
    recorded, as they should be; but each adds a full module of material, where the two together should add one;
  • the CRV strongback envelopes overlap the same way, and Run 2 roof concrete types 4 and 9 are two coincident slabs.

The same review of KinKalGeomMaker found further problems that this PR does not change: strongbacks and DS end
walls carry half their thickness, the CRV-U/D strongbacks sit on the tracker side of the module stack, the Run 1A
hatch concrete is misplaced (x and z swapped), and the Run 2 endcap concrete overlaps the CRV-D sectors.
#2109 fixes these, scales each CRV sector's material by the fraction of its layers present at the
crossing, and adds a consistency check of the KinKal geometry against Geant4. In this PR's off-spill validation
(20,000 nightly cosmics, field correction on and off) it affects 2 tracks: one crosses CRV-T1 and T2 at the same
point and gets 64 MeV/c of material from each, one crosses both strongbacks (5 MeV/c each).

Validation

References, in order: truth (identity-matched MC trajectories and SurfaceSteps) first, then main (473a2c54) with
KinKal v3.9.0. Samples: 10,000 nightly off-spill cosmic events (CentralHelix and the four LoopHelix fits; arms with
field correction on, calorimeter requests on and off, and field correction off); 5,000 CE events with 1BB pileup
(CeMLeadingLogMix1BB, standard on-spill reco, four LoopHelix fits), against main's continuous-integration build;
5,000 extracted-geometry cosmic events with the field off (KinematicLine). Truth-overlay EventNtuple throughout.
Analyses are crvextrap subcommands (abtransport, abtruth, abdstruth, abtrajtruth, abplanetruth,
abstats, rangeout, dscomplete).

Check Result
Nightly reco_03 (on-spill CE+1BB, all 16,508 events), every KalSeed vs main b24d4163 (p110) no fit loses FitOK; 97 KKUmu and 1 KKUe fits that main leaves unusable are usable; every other fit summary identical, including the momentum and t0; crossings with |dP| > 50 MeV/c 0 (main 275); IPA and foil crossings below the incidence floors 0 (main 108 and 356)
Same, AddCaloMaterial on (the Offline#2103 setting) no fit loses FitOK; 107 KKUmu gain it; 5 momenta change, 2 by more than 1 MeV/c
Same, ToCaloMaterial on no fit loses FitOK; 11 KKDmu, 97 KKUmu, 1 KKUe gain it. Electron front-panel crossings as #2081's passes in number and loss (carbon skins 1.26-1.30, foam 0.22, pipes about 1.4, about 2.8 MeV/c per panel). Inner rings: 2-5 times more crossings (disk 1 carbon-fibre ring 2,408 against 501, hole entries 1,007 against 395) at the same loss per crossing (aluminium 7-10, carbon fibre about 1.6 MeV/c). On KKDmu, #2081's passes reverse slow muons (657 crossings with |dP| > 50 MeV/c, 12 fits lost); here none, and 1,063 RangedOut
Same, both on no fit loses FitOK; no KalSeed carries a calorimeter layer or ring twice
Production off-spill setting (recoMC/OffSpill.fcl, 10,000 nightly cosmic events), every KalSeed vs main no fit loses FitOK; 169 LoopHelix fits that main leaves unusable are usable; the 135 LoopHelix fits that stay unusable (as on main) are exactly those flagged ExtrapFailed, with KinKal's reason; KKCHmu 93 RangedOut
Three arms, 10,000 events each; Offline#2076 event 1430:1652:852 all exit 0, no exceptions
Same tracks (raw-hit identity) and fit summaries vs main 1,175 (on) and 1,367 (off) tracks; every fit-summary field identical except the momentum at t0 on a few tracks (see Momentum at t0)
DS shells per leg with a truth barrel traversal, predicted from reco: complete / outer missing / inner missing on: 441 / 1 / 1 of 460 (main 320 / 103 / 12 of 452); off: 497 / 3 / 0 of 518 (main 466 / 35 / 0 of 519)
All five shells found where the truth muon crosses all five (direction-aligned legs) forward 185/192 (main 100/192), backward 278/375 (main 243/375)
Calorimeter requests on vs off identical DS (5,584), CRV (144), strongback (155) and concrete (1) crossings, all fields
KalSeeds flagged ExtrapFailed / RangedOut, all five fits 0 / 313 (on), 0 / 281 (off), of 42,397 and 42,589
Range-out changes only ranged-out tracks (vs the same build without the stop) every KalSeed without RangedOut identical (all fields and intersections, all five fits); ranged-out CentralHelix tracks lose only crossings after the stop
This revision (range-out bool, no seams) vs 9db6814c: nightly reco_03 events 0-1,999, every KalSeed every fit summary and crossing identical in every collection
Same, production off-spill setting, events 0-3,333, every KalSeed identical except 1 RangedOut KKCHmu, whose last segment now extends past the stop; every crossing identical
Same, off-spill cosmics with CRV extrapolation (10,000 events each, field correction on and off), every ntuple leaf vs the e27cd7f build identical except on 4 tracks: 1 ranged-out track gains 9 tracker-surface crossings after its stop (none in truth); 1 track crosses both CRV-T sectors and 1 both strongbacks at one point in the envelope-overlap band (see Known limitation); 1 field-off track crosses both strongbacks there and so no longer records the second strongback 1.4 m away, where the field-off helix returns to that plane
Reco CPU vs main 0.93 with field correction on (same host); 1.03 off (different hosts); LoopHelix fits 0.96-0.98 (on) and 1.08 (off)
LoopHelix CE: 5,000 events, 20,044 tracks, exit 0 same tracks; 24 muon-hypothesis fits that main leaves unusable (its material update reverses a stopping particle's momentum in the target) are usable and flagged RangedOut; every other fit summary identical; 0 ExtrapFailed
LoopHelix CE: momentum-reversing crossings; RangedOut 0 (main 95 on 54 tracks); 61 KalSeeds, all muon or e⁺ hypotheses, none on the e⁻ fits
KinematicLine, field off: 5,000 events, 3,591 tracks, exit 0 fits identical; tracker, CRV and strongback crossings identical; 0 ExtrapFailed, 0 RangedOut

Crossings added / removed relative to main (field correction on): DS shells +831 / -93, CRV +10 / -18, strongbacks
+13 / -14, calorimeter +67 / -19. Removed DS, CRV and strongback crossings are on low-momentum legs that main carried
through the cryostat only because it skipped shells, or past where the particle stops. Of the 18 removed CRV
crossings, 16 have no truth crossing of that sector; the other 2 are on poorly fitted legs that main placed 5 to 7 m
from the truth crossing (the PR likewise adds 2 such crossings, 6 and 12 m off). The remaining backward-leg misses (97 of 375; 132 on main) are helices that cannot reach the
shells (53), paths along the bore that exit through a DS end (29) or an end wall (9), grazing CryoInner (3, as on
main) and one comparator artefact.

Momentum at t0

A KalSeed reports its momentum at t0 from the saved segment nearest t0. For a cosmic CentralHelix fit t0 can lie
metres from the hits, so that segment can be an extrapolated piece, and its momentum then includes the material the
extrapolation crosses between the hits and t0, which this PR finds more completely (the DS shells). With the
production off-spill setting (10,000 nightly cosmic events), the reported momentum changes on 23 of 1,367 KKCHmu
KalSeeds, on 5 by more than 1 MeV/c (-101 to +72 MeV/c); fit consistency and hits are unchanged. Against truth at
the reported t0 (521 truth-matched tracks) the distribution does not change: median reported - truth 5.3 MeV/c on
main and 5.0 on this PR, the same 68 % width. On the 2 changed tracks with truth, both builds are off by 50 to 190
MeV/c. Neither value is reliable when t0 lies outside the fitted range. On-spill (reco_03) with KinKal v3.9.1 the
LoopHelix momentum and t0 are unchanged.

Range-out against truth

Truth fate after the muon's last truth tracker crossing: it stops (trajectory ends with kinetic energy below 1 MeV),
turns back in z with at least half its energy (magnetic mirror), or leaves. Scored: truth-matched muons whose reco
momentum at the tracker points along the truth momentum (455 tracks with field correction on, 532 off). No scored
muon mirrors.

Field correction on Field correction off
Truth stops / leaves 267 / 188 328 / 204
Flagged RangedOut 50 45
Purity: truth stops, of flagged 48 / 50 = 96 ± 3 % 44 / 45 = 98 ± 2 %
Flagged, of muons that leave 2 / 188 = 1.1 ± 0.8 % 1 / 204 = 0.5 ± 0.5 %
Efficiency: flagged, of truth stops 48 / 267 = 18 ± 2 % 44 / 328 = 13 ± 2 %
Efficiency by where truth stops: DS shell stack 11 / 19 = 58 ± 11 % 6 / 21 = 29 ± 10 %
inside the tracker volume (fitted region) 24 / 114 27 / 152
bore upstream of the tracker / downstream (not calorimeter) 7 / 60, 4 / 11 3 / 59, 6 / 17
calorimeter / outside the cryostat 1 / 50, 1 / 13 2 / 65, 0 / 14
Crossings whose material update would reverse the momentum 0 (149 on 110 tracks without the stop; main 41 on 38) 0 (main 19 on 19)

The flag is reliable when set; most stops are not flagged because they happen where the extrapolation has no
material: inside the tracker after the last hit, in the calorimeter (record-only surfaces), or in bore structures.
In the DS stack, 6 of the 8 missed stops have a reco momentum at the tracker exit 1.3 to 25 times the truth momentum.
The stop position is coarse: the last kept reco crossing lies a median 1.2 m from the truth stopping point.

Extrapolation residuals against truth

Truth at each reco surface is the MC trajectory interpolated where it crosses that surface (crvextrap abtrajtruth;
CRV also from abplanetruth's mid-plane truth). Residuals are |Δpos|, Δt = t_reco - t_truth and
Δp = |p_reco| - p_truth. "Paired" crossings are found once by both builds with the same truth crossing; the change
is the mean change of each absolute residual, with its event-clustered standard error (negative = better). Field
correction on; quality tracks (status > 0, ≥ 20 active hits, fit consistency > 0.01, χ²/ndof < 3):

Surfaces Paired Median |Δpos| main → PR (mm) Δt median / core68 (ns) Δp median (MeV/c) Change in |Δpos| (mm) Change in |Δp| (MeV/c) New in PR: n, median |Δpos| (mm)
Tracker faces 138 7.3 → 7.3 0.12 / 1.17 -0.2 -0.36 ± 0.36 +0.12 ± 0.12 0
DS shells 281 8.2 → 8.2 0.37 / 1.09 -2.3 -0.10 ± 0.05 +0.02 ± 0.02 28, 8.9
Calorimeter faces and edges 69 176.5 → 176.5 0.12 / 1.85 -20.7 +0.001 ± 0.001 0 1
CRV sectors 7 117.1 → 117.4 0.26 / 1.10 -837 -0.36 ± 0.41 -11.7 ± 11.9 0

Without the quality cut the conclusions are the same: DS shells (2,858 paired) |Δpos| change -1.6 ± 0.6 mm, |Δp|
+0.3 ± 0.4 MeV/c, and the 379 crossings only the PR finds have a median |Δpos| of 63 mm against 76 mm for the
paired ones; calorimeter (345) -0.5 ± 0.5 mm; CRV against mid-plane truth (125 crossings) |Δpos| +5 ± 11 mm and
|Δp| -3.7 ± 2.6 MeV/c, median change 0. With field correction off, DS shells (3,493 paired) |Δpos| -0.03 ± 0.02 mm
and |Δp| +0.04 ± 0.01 MeV/c; the DS end walls (66) +0.26 ± 0.15 mm and +0.14 ± 0.06 MeV/c; everything else within
1.5 standard errors of zero.

So no residual gets measurably worse: the only significant increase, |Δp| at the DS shells with field correction off,
is 0.04 MeV/c against a residual width of about 2 GeV/c. Crossings both builds find keep their residuals; DS-shell
positions improve slightly because every crossing now lies on its shell, where main places some off it (field
correction on: ShieldInner 86 crossings up to 19 mm off, Coil 100 up to 8 mm, CryoOuter 16 up to 0.9 mm; off:
ShieldInner 160 up to 39 mm); and the added DS-shell crossings are as accurate as the existing ones.

LoopHelix CE and KinematicLine

CE crossings, main → PR (all tracks): ST foils 87,547 → 87,803, IPA 46,511 → 46,415, TSDA 2,008 → 3,699, ST front and back
sampling surfaces 960 → 1,551, calorimeter 69,693 → 69,804. Every crossing both builds find keeps its residual
against truth; changes are at the 1e-3 mm level (quality tracks, abtrajtruth unless noted):

Surfaces Paired Median |Δpos| main → PR (mm) Change in |Δpos| (mm) Change in |Δt| (ns), |Δp| (MeV/c) New in PR with truth: n, median |Δpos| (mm)
Tracker faces 14,046 4.05 → 4.05 -0.00002 < 1e-6, < 1e-7 0
IPA 8,366 15.5 → 15.5 +0.0002 < 1e-6, < 1e-7 0
ST foils 7,468 56.3 → 56.3 -0.0005 4e-6, 6e-6 0
ST foils, SurfaceStep truth (abtruth) 6,516 22.9 → 22.9 +0.0002 ± 0.0005 < 1e-6, < 1e-6
Calorimeter faces and edges 9,684 11.18 → 11.18 -0.00007 < 1e-6, < 1e-7 1
ST back sampling surface 180 31.2 → 31.2 0 0, 0 82, 35.7

TSDA (the TS downstream absorber, the upstream sweep's end plane at z = -5967 mm): the PR records it on 3,699
tracks, main on 2,008, and every track main records it on the PR records too, at the same pitch (median |pz|/p 0.34).
On all 8 PR-only KKDe tracks in the first 300 events (traced), main's helix reaches the plane as well, but its TSDA
walk ends with the trajectory front 0.4 to 9 mm (one field-domain step) short of it, so its intersection test finds
nothing: the last domain step past the plane is not on the saved trajectory (the KinKal end-domain gap). No truth particle reaches the TSDA: CE electrons start in the target and go
downstream, and those that first go upstream mirror in the rising field before it (deepest truth point -5963 mm).
So on CE, TSDA crossings in both builds are the fitted helix continued upstream past the electron's origin in the
target: a downstream fit extrapolated backwards (main 718, PR 1,372 on truth-matched tracks) or an upstream-hypothesis
fit, the same helix traversed the other way (727, 1,388). The extrapolation itself follows the mirror
(crvextrap mirror): of 1,854 tracks whose truth particle turns back upstream of the target, the reco turns back on
1,746 in both builds, and the PR never turns a reco turn-back into a TSDA crossing; it reaches the TSDA on 67 (main 39;
on the other 28, main's TSDA walk stops short of the plane or misses a near-turnaround crossing). Those 67 agree with truth in the tracker, and their reco |pz|/p exceeds truth by a median 0.01
from the first IPA crossing on (0.001 on the tracks that turn back): scattering in the IPA and target that the
extrapolated mean trajectory cannot follow, and near the mirror point a pitch difference of 0.01-0.04 moves the
turnaround by hundreds of mm. Over all upstream legs the reco pitch tracks truth to a median 0.001 up to the target,
in both builds, and a tenfold finer extrapolation step (0.1 ns, BCorrTolerance 1e-6) changes no crossing.
Upstream generally, equally in both builds: of the ST-foil crossings on truth-matched tracks, 32 % lie before the
particle's first truth point and 13 % have a truth crossing. Range-out on CE: of 9,212 scored truth-matched tracks, 8,779 stop in
the calorimeter, where the extrapolation has no material, and one is flagged (a true stop).

KinematicLine (field off): the PR adds 11 calorimeter crossings with truth (median |Δpos| 5.3 mm against 5.8 mm for
the paired ones) and 2 DS end-wall crossings; paired tracker, calorimeter, CRV and strongback residuals change by
less than 0.001 mm.

CPU: each search window skips any surface farther from the window start than speed x window length, which the
trajectory cannot reach in that window (exact for any field direction). Without this bound the upstream sweep
queries all 37 stopping-target foils in every window and the LoopHelix fits take 53 to 60 % longer; with it the
extrapolation is 11 % of LoopHelixFit::produce instructions (callgrind, 100 events), and the output is
identical with and without the bound (every KalSeed field and intersection, every ntuple crossing, all three arms).

🤖 Generated with Claude Code

@FNALbuild

Copy link
Copy Markdown
Collaborator

Hi @RobMina,
You have proposed changes to files in these packages:

  • Mu2eKinKal
  • RecoDataProducts

which require these tests: build.

@Mu2e/fnalbuild-users, @Mu2e/write have access to CI actions on main.

⌛ The following tests have been triggered for e27cd7f: build (Build queue - API unavailable)

About FNALbuild. Code review on Mu2e/Offline.

@FNALbuild

Copy link
Copy Markdown
Collaborator

☀️ The build tests passed at e27cd7f.

Test Result Details
test with ✅ Command did not list any other PRs to include
merge ✅ Merged e27cd7f at 807d948
build (prof) ✅ Log file. Build time: 09 min 14 sec
ceSimReco ✅ Log file.
g4test_03MT ✅ Log file.
transportOnly ✅ Log file.
POT ✅ Log file.
g4study ✅ Log file.
cosmicSimReco ✅ Log file.
cosmicOffSpill ✅ Log file.
ceSteps ✅ Log file.
ceDigi ✅ Log file.
muDauSteps ✅ Log file.
ceMix ✅ Log file.
rootOverlaps ✅ Log file.
g4surfaceCheck ✅ Log file.
trigger ✅ Log file.
check_cmake ✅ Log file.
FIXME, TODO ➡️ TODO (20) FIXME (5) in 18 files
clang-tidy ➡️ 2 errors 292 warnings
whitespace check ✅ no whitespace errors found

N.B. These results were obtained from a build of this Pull Request at e27cd7f after being merged into the base branch at 807d948.

For more information, please check the job page here.
Build artifacts are deleted after 5 days. If this is not desired, select Keep this build forever on the job page.

@oksuzian oksuzian left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

PR Review Summary — #2084

Reviewed at head e27cd7fa69b50da2cb6833552a44e9550d9f6791. First pass.

Decision

  • 🔴 Request changes, for one S1 (finding 1). The sweep adds the IPA and the stopping-target foils as material at grazing contacts, with no incidence limit. On the nightly reco_03 sample, this turns 5 KKDe fits and 2 KKDmu fits that are good on main into unusable ones. One of them is a 102.1 MeV/c KKDe track with a calorimeter cluster. Apart from that, what I could check matches the description: the crossings both builds find agree, and range-out removes a real pathology that main has. KinKal#232 should be tagged first (see Merge order).

Scope understood

  • KKExtrap::extrapolate now builds one list of target surfaces per time direction and processes them in a single time-ordered sweep (ExtrapolateSweep.hh). It replaces the eight request-specific predicate headers.
  • It adds two TrkFitFlag bits. ExtrapFailed (28) means one direction ended on a problem with the track. RangedOut (29) means the forward direction stopped where the mean energy loss reaches the kinetic energy. Segments and sampling stop at the range-out time.
  • No KKExtrapConfig or fcl change. Production runs this code in the LoopHelix fits (LHDRIFTXTRAP, via recoMC/OnSpill.fcl, which the mdc2025, mdc2030 and mds3a reco.json campaigns use) and in KKCHmu (CHXTRAP, via recoMC/OffSpill.fcl).

Findings

  1. 🟠 [S1] Grazing IPA and stopping-target-foil contacts are added as material without an incidence limit, so seven fits that are good on main become unusable

    • Evidence: I ran all 16,508 events of the nightly reco_03 job (Production/Validation/nightly/reco_03.fcl, which is recoMC/OnSpill.fcl) on this head and on its merge base 473a2c54, and compared every KalSeed printout.
      • Five KKDe fits and two KKDmu fits that have FitOK on the merge base lose FitOK here. Each gets ExtrapolationFailed, a fit consistency of -1, and one saved segment.
      • The fits are 1430:967:1931, 1430:967:3331, 1430:1065:4707, 1430:1069:3043 and 1430:1691:683 (KKDe), and 1430:737:2858 and 1430:1691:728 (KKDmu).
      • The result is the same with KinKal#232's header, and again with #231 added.
    • Mechanism, from the crossings the PR saves:
      • IPA. A track that runs tangent to the IPA is given the IPA again every four intersection tolerances while it stays on the surface. IntersectionTolerance is 0.1 mm, so that is about every 1.3 ps.
        • On 1430:1691:728 KKDmu there are 79 IPA crossings between 445.680 and 445.783 ns, all at r = 300.0 mm, with |cos| to the normal below 0.014. Going backwards in time, each one adds energy: the momentum climbs from 130.7 to 6051 MeV/c, and one step has dP = -2405.9 MeV/c.
        • The other four KKDe cases have 24 to 44 such IPA crossings, and 1430:737:2858 KKDmu has 14.
      • Stopping-target foil. On 1430:1691:683 KKDe (p = 102.1 MeV/c, fit consistency 0.183 on the merge base), the track's magnetic mirror point lies 0.02 mm from foil 4 (z = -4611.13 mm). That contact is added as a foil crossing with pz = -0.000 and dP = -77,318 MeV/c.
      • In all seven, the runaway momentum makes KinKal's field-corrected walk leave the field map. I reran each event with the extrapolation debug print on, and each logs "extrapolation failed backwards: Extrapolation error: Position out of range". KinKal then marks the fit outsidemap, and createSeed clears FitOK.
      • Over the sample, crossings with |dP| > 50 MeV/c on KKDe and KKDmu go from 1 on the merge base to 59 here. Eight of these are on fits that stay FitOK.
    • Cause in the code:
      • The DS shells get an incidence floor: mincos_ = max(MinV, dsShellMinCos_ = 0.1), at KKExtrap.hh:221 and :54.
      • The IPA target (:153-163) and the foil targets (:167-181) keep the default mincos_ = 0 (ExtrapolateSweep.hh:54), so the check at :404 never rejects them.
      • The IPA also has maxcross_ = 0, and it is not covered by the plane-only "leave the tolerance first" rule (ExtrapolateSweep.hh:157). So it is found again as soon as the 4-tolerance guard (:371-372) has passed.
    • Reproduce on one event (about one minute):
      #include "Production/Validation/nightly/reco_03.fcl"
      source.fileNames : [ <the two reco_03 dig files> ]
      source.firstRun : 1430
      source.firstSubRun : 1691
      source.firstEvent : 683
      physics.producers.KKDe.ExtrapolationSettings.Debug : 1
      
      Run it with mu2e -c ev683.fcl -n 1.
      • On the merge base, KKDe has fit consistency 0.183 and FitOK, with 21 intersections.
      • At this head the log shows KKExtrap: backwards ( 104: ST_Foils ):4 ... at time 420.501 position (-12.5731,-57.7294,-4611.13), then KKExtrap: extrapolation failed backwards: Extrapolation error: Position out of range. The KalSeed reads fit consistency -1.000, one segment and ExtrapolationFailed, without FitOK.
    • Impact:
      • In the production on-spill reconstruction, about 3 in 10,000 KKDe fits are lost, one of them at signal momentum.
      • The lost fits' saved intersections carry momenta of up to 77 GeV/c.
      • The description reports 0 ExtrapFailed on 5,000 CE events. At the rate here, about 1.5 KKDe losses would be expected in that sample, so finding none is plausible by chance.
    • Suggested fix:
      • Give the IPA and foil targets an incidence floor, as the DS shells have.
      • Do not add the IPA again while the track has not left its surface tolerance, which is the rule the plane targets already follow.
      • I tried the DS-shell value 0.1 on both (mincos_ = std::max(minv_,dsShellMinCos_) on the IPA and foil targets). It removes all seven losses and every |dP| > 50 MeV/c crossing, and keeps most range-out flags (258 become 244). But it also drops ST-foil crossings from 285,034 to 273,836 (-3.9%), so the foil value needs choosing rather than copying.
  2. 🟡 [S2] A failure inside KinKal's field-corrected walk makes the whole fit unusable, not just one direction

    • Evidence:
      • The description says a problem with one track "ends that direction only".
      • With field correction on, Track::extrapolate calls the predicate (needsExtrapolation) inside a try block. Its catch records Status::outsidemap (KinKal v3.9.0 Fit/Track.hh:975, :1000-1002, :1025, :1034-1036), and outsidemap is not usable() (Fit/Status.hh:18). createSeed then clears FitOK (KKFit.hh:686-688).
      • The same thing happens for KinKal's own map exceptions, which is what finding 1 hits. It also happens for any of the sweep's own runtime_errors when they fire during the walk. A throw that happens in KKExtrap::sweep itself, outside the walk, only sets ExtrapFailed (KKExtrap.hh:94-101, :314).
      • On off-spill cosmics with the production settings, 70 LoopHelix fits carry ExtrapolationFailed. All 70 are already unusable on the merge base, so this is the existing KinKal behaviour now made visible, not new damage.
    • Suggested fix: either keep the sweep's own errors out of the walk (catch them in needsExtrapolation, record them, and return false), or correct the description.
  3. 🟡 [S2] The Regrow modules can carry stale new flags and swallow a configuration error

    • Stale flags. createSeed starts from the flag it is given and only ever sets ExtrapFailed and RangedOut (KKFit.hh:680, :693-694). RegrowLoopHelix (:307-309) and RegrowKinematicLine (:252-254) pass in the original seed's status(). So a regrown track keeps RangedOut or ExtrapFailed from the original fit even when its own extrapolation no longer sets them. A fix is to clear the two bits before merging, as createSeed does for FitOK.
    • Swallowed configuration error. The fit modules and KKTruthSeed call KKExtrap::checkConfig, but neither Regrow module does. KinKal's "minimum domain step exceeds the extrapolation maximum step" throw (v3.9.0 Fit/Track.hh:950) is now caught per track and becomes ExtrapFailed, where on main it stopped the job. A fix is to call checkConfig in the two Regrow constructors.
    • Regrow is configured in Production (JobConfig/reco/regrowLH.fcl, regrowKL.fcl), but no prodtools campaign config runs it, so this is S2. I did not run a Regrow job.
  4. 🟡 [S2] With the production setting, the KKCHmu momentum at t0 changes for 7% of tracks, and the description does not mention it

    • Evidence: I ran 9,314 off-spill cosmic events with the recoMC/OffSpill.fcl KKCHmu settings (field correction off). 382 of 5,384 KKCHmu KalSeeds report a different t0-segment momentum than the merge base: 122 change by more than 1 MeV/c, ranging from -214 to +242 MeV/c. Fit consistency and hit counts are unchanged.
    • Example, 1430:0:317: 1404.195 becomes 1315.191 MeV/c. The difference, 89.0 MeV/c, equals the summed dP of the five DS crossings at 46180.3-46181.7 ns. t0 (46224.3 ns) lies after the track has left the DS there, so the nearest saved segment now includes those losses.
    • Reporting a momentum at a t0 outside the fitted range is existing KalSeed behaviour. Still, this changes production KKCHmu output, and the description's "every fit-summary field identical" does not cover it. Please state it, or say which of the two values matches truth better.
  5. ⚪ [S3] Small items

    • ExtrapolateSweep::isActive (ExtrapolateSweep.hh:87) has no caller.
    • The new header adds five namespace-scope using declarations (:26-30).
    • The comment explaining why the IPA and ST thicknesses are literals, with its TODO (merge base KKExtrap.hh:65-69), was shortened to one line (:50-52) that no longer says where the numbers come from.

Merge order

  • KinKal#232. Its header changes nothing for this PR on CE (all 16,508 reco_03 printouts are identical with and without it) or with the production off-spill setting. With field correction on in KKCHmu (Validation/offSpill.fcl), though, 367 of 4,876 KKCHmu tracks get ExtrapolationFailed without it ("material crossing ( 85: DS_CryoInner ):0 at 9996.62 is not owned by the end piece Range [9995.52,9996.52]", on 1430:0:310), and 0 with it. Merge this PR only once a KinKal tag with #232 is in the envset, or the nightly off-spill validation will show this.
  • Offline#2081 (calorimeter material) conflicts with this PR in KKExtrap.hh (git merge-tree). Its extrapolateCaloMaterial pass will have to become sweep targets, in whichever PR merges second.

Checked, no action needed

  • 🟢 The #2076 class is handled.
    • t0 is clamped into the fitted range before use (KKExtrap.hh:83-84), and search clamps its window into the trajectory range (ExtrapolateSweep.hh:417-418).
    • Every new TimeRange is built from ordered times: ExtrapolateSweep.hh:311, :377 (guarded at :375) and :423, CentralHelixFit_module.cc:447, and KinematicLineFit_module.cc:382 and :401.
    • Event 1430:1652:852 runs, and all 16,508 events finish with exit code 0 on this head, with #232, and with #231 and #232 together.
  • 🟢 Range-out does what the description says.
    • On the merge base, 274 KKUmu crossings have |dP| > 50 MeV/c, up to -1.8×10⁷ MeV/c (the material update reverses a stopping muon). Here there are none on KKUmu. 252 KKUmu and 6 KKUe fits carry RangedOut, and 97 KKUmu fits that the merge base leaves unusable become usable.
    • All 164 OPA crossings this PR no longer records are on KKUmu tracks that have such a crossing on the merge base, for example 1430:1612:284, whose OPA crossing there has a momentum of 1.8×10⁷ MeV/c.
  • 🟢 The TrkFitFlag bits are added safely. Bits 28 and 29 were unused, the values are explicit, and MCSeed stays at 31. The name map adds "ExtrapolationFailed" and "RangedOut" (TrkFitFlag.cc:52-53).
  • 🟢 Build files and removed headers.
    • ExtrapolateSweepTest is in both the SConscript and CMakeLists.txt, and FNALbuild's check_cmake passed at this head. I ran the binary, and it reports "all checks passed". CI does not run it, which is the same as StereoLineTest.
    • Nothing outside KKExtrap.hh includes the eight removed headers: I checked Offline main 807d948f and ran a Mu2e organization code search.
  • 🟢 Crossings both builds find agree. Calorimeter, TT_Front, TT_Mid and TT_Back crossings that both builds record match to within 0.003 ns.

Validation check

  • Build/tests run:
    • I built three muse workspaces from the merge base's Production 4479f3fa and mu2e-trig-config a664b74b: this head with release KinKal v3.9.0, this head with KinKal#232's Fit/Track.hh, and this head with #231 and #232 together. All three gave rc 0 and no warnings. For the overlay builds I confirmed that the overlaid header is the one compiled.
    • On each, I ran the full 16,508-event reco_03 job and compared it with the merge-base printouts kept from the #2076 work.
    • I ran 9,314 off-spill cosmic events with the validation and the production KKCHmu settings, on the merge base, this head, and this head with #232.
    • I ran the single-event debug jobs quoted above, and ExtrapolateSweepTest.
  • Config contract check: pass. There is no fcl change, and KKExtrapConfig is unchanged.
  • Cross-repo consistency: needs follow-up, for KinKal#232 and Offline#2081 (see Merge order). EventNtuple#436 reads the new bits.

Residual risk

  • I did not run the KinematicLine (field off, extracted) or truth-matched comparisons. The truth residuals in the description are taken as reported.
  • The CE comparison covers one dataset (CeMLeadingLogMix1BB MDC2025au, the nightly inputs).

Author follow-ups

  1. Bound the IPA and ST-foil grazing contacts (finding 1), and rerun the CE comparison checking for fits that lose FitOK.
  2. Make the description's "ends that direction only" true, or correct it (finding 2).
  3. Clear the two bits and call checkConfig in the Regrow modules (finding 3).
  4. Describe the KKCHmu t0-momentum change (finding 4).
  5. Merge after a KinKal release with #232, and agree the order with #2081.

@RobMina

RobMina commented Oct 30, 2026

Copy link
Copy Markdown
Contributor Author

Thanks for the careful review. The new commit addresses all five findings; the description is updated.

  1. Grazing IPA and foil contacts. The IPA and the stopping-target foils now have incidence floors, as the DS shells do: IPA 0.1 (the DS-shell value) and foils 0.02. I chose the foil value from truth: on 5,000 CE events no truth-matched foil crossing has |cos| below 0.02 (and no IPA crossing below 0.2), while 0.05 and 0.1 would drop truth-matched crossings (5 and 43). After a crossing, a target is also searched again only once the track can have left the intersection tolerance at that crossing's incidence, so a track running along the IPA gets it once. ExtrapolateSweepTest has a new case for this (a cylinder crossed at |cos| 0.003 to 0.5).

Full nightly reco_03, all 16,508 events, every KalSeed against the merge base 473a2c5:

No fit loses FitOK, in any collection. Your seven fits are FitOK with the merge base's momentum and fit consistency.
Crossings with |dP| > 50 MeV/c: 0 (merge base 275, all but one on KKUmu).
IPA and foil crossings below the floors: 0 (merge base 108 and 356).
ST-foil crossings 284,653 (merge base 284,229); RangedOut 236, all KKUmu.
97 KKUmu and 1 KKUe fits that the merge base leaves unusable are usable.
2. Failures inside KinKal's walk. The sweep's own errors are now caught in needsExtrapolation: it records the reason and returns false, so KinKal ends the walk normally, the fit stays usable and the KalSeed gets ExtrapFailed. KinKal's own failures (such as leaving the field map) still mark the fit unusable, as on main; the description now says so. With the production off-spill setting, 135 LoopHelix fits are unusable on both builds, and they are exactly the ones flagged ExtrapFailed; 169 fits that the merge base leaves unusable are usable.

  1. Regrow. createSeed clears ExtrapFailed and RangedOut before setting them from the track, and both Regrow modules call checkConfig. I have not run a Regrow job.

  2. KKCHmu momentum at t0. Added to the description (Momentum at t0). With the production setting on the nightly off-spill input (10,000 events, 1,367 KKCHmu), the reported momentum changes on 23 KalSeeds, 5 by more than 1 MeV/c. Against truth at the reported t0 (521 truth-matched tracks), the distribution is unchanged: median reported - truth 5.3 MeV/c on the merge base, 5.0 here, same width. On the 2 changed tracks with truth, both values are 50-190 MeV/c off: when t0 lies outside the fitted range neither is reliable. My sample has fewer KKCHmu than yours (5,384 in 9,314 events), so the counts are not directly comparable.

  3. Small items. isActive is removed, the new header uses qualified names, and the thickness comment says again where the numbers come from, with the TODO.

Merge order. Agreed: the description now asks for a KinKal tag with KFTrack/KinKal#232 first, and notes the KKExtrap.hh conflict with #2081.

🤖 Generated with Claude Code

@oksuzian oksuzian left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

PR Review Summary — #2084

Reviewed at head 4afa41b1415c687728df129a68cd0e3f31180917. Re-review of e27cd7f.

Decision

  • 🟢 Approvable. The new commit fixes all five findings from my first review, and I verified each one. On the full nightly reco_03 sample no fit loses FitOK, and the seven fits that e27cd7f made unusable are back to their main values. The KinKal tag with KFTrack/KinKal#232 stays a merge-order condition, which the description now states. FNALbuild has not built this head yet (mu2e/buildtest: "This test has not been triggered yet"), so I am posting this as a comment and have asked for a build. I will approve once it is green. My two local builds of this head compile without warnings.

What changed since e27cd7f

  • One new commit, "Fix grazing intersections in extrapolation", on top of e27cd7f. The branch was not rebased, and the merge base is still 473a2c54. It touches ExtrapolateSweep.hh, KKExtrap.hh, KKFit.hh, the two Regrow modules and ExtrapolateSweepTest_main.cc.

Prior findings

  • 🟢 [was 🟠 S1] Grazing IPA and stopping-target-foil contacts added as material. FIXED in 4afa41b, verified.

    • The IPA target now has an incidence floor of 0.1 and the foil targets 0.02 (KKExtrap.hh:55-56, :164, :182).
    • I ran all 16,508 events of the nightly reco_03 job on this head and compared every KalSeed printout with the merge base. The result:
      • No fit in any collection loses FitOK. 97 KKUmu fits and 1 KKUe fit that the merge base leaves unusable become usable, and 236 KKUmu fits carry RangedOut.
      • All seven fits from my first review have FitOK again, with the merge base's fit consistency, momentum and t0. These are 1430:967:1931, 1430:967:3331, 1430:1065:4707, 1430:1069:3043 and 1430:1691:683 (KKDe), and 1430:737:2858 and 1430:1691:728 (KKDmu).
      • On 1430:1691:683 with the extrapolation debug print on, there is no foil-4 contact and no failed extrapolation. KKDe has fit consistency 0.183 and 21 intersections, as on the merge base.
      • Crossings with |dP| > 50 MeV/c go from 275 on the merge base to 0. IPA crossings with |cos| below 0.1 go from 108 to 0, and foil crossings below 0.02 from 356 to 0.
      • Foil crossings total 284,653, against 284,229 on the merge base. The 0.1 floor I tried on the first pass cost 3.9% of them (273,836), so 0.02 is the better value.
    • The new re-search guard (ExtrapolateSweep.hh:389-390, :426) does not drop separate traversals of a surface; it removes double counts. On tracks whose fit status is unchanged and that have no grazing crossings, e27cd7f added 505 IPA crossings that re-found a traversal it had just added, 1 to 3 ps later and above the floor. On 1430:737:204 KKDe, for example, there were crossings at 765.957 and 765.955 ns, each with dP of about -0.31 MeV/c.
      • On 219 of the 247 tracks with such re-finds, this head has exactly that many fewer IPA crossings.
      • On the other 28, the count differs by a few in either direction, because removing the extra energy loss changes the later loops. For example, 1430:967:3925 KKUmu keeps 3.6 MeV/c more and reaches the TSDA without a third IPA loop.
      • The merge base itself has 541 IPA crossings that exactly duplicate another one, and this PR removes those too.
  • 🟢 [was 🟡 S2] A failure inside KinKal's walk made the whole fit unusable. FIXED in 4afa41b, verified.

    • needsExtrapolation now catches the sweep's own exceptions (rethrowing std::bad_alloc), records the reason and returns false (ExtrapolateSweep.hh:176-188). KKExtrap checks failed() after the limited pass and after each extension, and records the reason as an extrapolation failure (KKExtrap.hh:307-310, :322-330). ExtrapolateSweepTest checks this path, and it reports "all checks passed" at this head.
    • The description now says that KinKal's own failures still make the fit unusable, as on main. On 9,314 off-spill cosmic events with the production KKCHmu setting, I checked:
      • The 70 LoopHelix fits that are unusable on both builds are exactly the ones flagged ExtrapFailed.
      • Rerunning one of them (1430:0:750 KKDmu) with the debug print logs "Extrapolation error: Position out of range", which is KinKal's field-map exit.
      • 22 fits that the merge base leaves unusable become usable, and none loses FitOK.
    • These samples did not produce a sweep error inside the walk, so the catch itself is covered only by the unit test.
  • 🟢 [was 🟡 S2] Stale flags in the Regrow modules and a swallowed configuration error. FIXED in 4afa41b, verified by reading the code.

    • createSeed now clears ExtrapFailed and RangedOut before setting them from the track (KKFit.hh:694-697).
    • Both Regrow constructors call checkConfig once extrap_ exists. config_ is built in the member initializer list, so it is ready by then (RegrowLoopHelix_module.cc:171, :184; RegrowKinematicLine_module.cc:157, :167).
    • I did not run a Regrow job.
  • 🟢 [was 🟡 S2] The KKCHmu momentum at t0 changes without being mentioned. FIXED: the new "Momentum at t0" section describes the effect and compares it with truth.

    • On my off-spill sample (9,314 events), 382 of 5,384 KKCHmu KalSeeds still report a different t0 momentum, the same as at e27cd7f. The description's 23 of 1,367 comes from a different input, so the rate depends on the sample.
    • On reco_03, the description's "13 low-momentum fits" matches what I see: 12 LoopHelix fits of 36 to 68 MeV/c whose t0 moves by 0.1 to 0.4 ns, and 1430:737:1470 KKDe, whose momentum moves from 36.193 to 36.224 MeV/c.
  • 🟢 [was ⚪ S3] Small items. FIXED in 4afa41b, verified.

    • isActive is gone.
    • The namespace-scope using declarations are replaced by qualified names.
    • The thickness comment again says where the numbers come from (KinKalGeomMaker::makeDS/makeTarget), with the TODO.

Merge order

  • The description now asks for a KinKal tag with KFTrack/KinKal#232 first. With the production settings it makes no difference: reco_03 printouts are identical with and without the #232 header, as are the production off-spill outputs.
  • With KKCHmu field correction on (Validation/offSpill.fcl) and without #232, 367 of 4,876 KKCHmu tracks are flagged ExtrapFailed but stay usable. With #232, none is flagged.
  • Offline#2081 still conflicts in KKExtrap.hh (git merge-tree), as the description notes. This head merges cleanly with current main (858d5f4d).

Validation check

  • Builds: two muse workspaces of this head, from the merge base's Production 4479f3fa and mu2e-trig-config a664b74b. One uses release KinKal v3.9.0 and one has KinKal#232's Fit/Track.hh overlaid; every recompiled unit carries the overlay. Both built with exit status 0 and no warnings.
  • Jobs:
    • The full 16,508-event reco_03 job on both builds (all 12 chunks exit 0). Every KalSeed printout, including crossings, is identical between the two builds.
    • 9,314 off-spill cosmic events with the production and the validation KKCHmu settings, on both builds.
    • Single-event debug reruns of the seven fits and of 1430:0:750.
    • ExtrapolateSweepTest.
  • Not run: a Regrow job, the KinematicLine (field off) comparison, and truth-matched residuals, which are taken from the description.

@oksuzian

Copy link
Copy Markdown
Collaborator

@FNALbuild run build test

@FNALbuild

Copy link
Copy Markdown
Collaborator

⌛ The following tests have been triggered for 4afa41b: build (Build queue - API unavailable)

@FNALbuild

Copy link
Copy Markdown
Collaborator

☀️ The build tests passed at 4afa41b.

Test Result Details
test with ✅ Command did not list any other PRs to include
merge ✅ Merged 4afa41b at 858d5f4
build (prof) ✅ Log file. Build time: 04 min 30 sec
ceSimReco ✅ Log file.
g4test_03MT ✅ Log file.
transportOnly ✅ Log file.
POT ✅ Log file.
g4study ✅ Log file.
cosmicSimReco ✅ Log file.
cosmicOffSpill ✅ Log file.
ceSteps ✅ Log file.
ceDigi ✅ Log file.
muDauSteps ✅ Log file.
ceMix ✅ Log file.
rootOverlaps ✅ Log file.
g4surfaceCheck ✅ Log file.
trigger ✅ Log file.
check_cmake ✅ Log file.
FIXME, TODO ➡️ TODO (21) FIXME (5) in 20 files
clang-tidy ➡️ 4 errors 341 warnings
whitespace check ✅ no whitespace errors found

N.B. These results were obtained from a build of this Pull Request at 4afa41b after being merged into the base branch at 858d5f4.

For more information, please check the job page here.
Build artifacts are deleted after 5 days. If this is not desired, select Keep this build forever on the job page.

@oksuzian oksuzian left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

PR Review Summary — #2084

Reviewed at head 4afa41b1415c687728df129a68cd0e3f31180917. Approval following my re-review at this head.

Decision

  • 🟢 approve. FNALbuild passed at this head when merged with main 858d5f4d: the build, every validation job, trigger, g4surfaceCheck, rootOverlaps, check_cmake and the whitespace check are green. My re-review at this head (review 5718680937) found all five earlier findings fixed, so this approval replaces my earlier request for changes.

Merge order, unchanged

  • Please merge only once a KinKal tag with KFTrack/KinKal#232 is in the envset, as the description says. #2081 still conflicts with this PR in KKExtrap.hh.

@oksuzian oksuzian mentioned this pull request Nov 2, 2026
@oksuzian
oksuzian requested a review from brownd1978 November 3, 2026 23:50
@RobMina

RobMina commented Nov 5, 2026

Copy link
Copy Markdown
Contributor Author

I merged main (b24d4163: Offline#2081 and envset p110 with KinKal v3.9.1) into this branch. The only conflict was
KKExtrap.hh, where #2081 added the calorimeter material passes. They are now targets of the downstream sweep:

  • The front-panel layersa and cooling pipes are plane targets; a crossing of the pipe plane between the pipes adds nothing.
  • Each crystal case inner support is found where the track enters the case, at its front or from the disk hole through
    the inner cylinder; the second material goes where the track then reaches the carbon-fibre ring radius, with
    material effects #2081's path lengths.
  • As in material effects #2081: the material starts at the last measurement, layers and rings already in the fit (AddCaloMaterial)
    are skipped, the supports are entered only forwards in time, and both options stay off by default.
  • Differences: each crossing is added on the end piece as the sweep reaches it, and gets the range-out check.
    material effects #2081's passes extend first and then intersect, so they skip crossings that land on an interior piece of their own
    extension (133 in 50 reco_03 events, from their debug print), most disk-1 inner-ring crossings among them.

ExtrapolateSweepTest has a new case, and passes. Full nightly reco_03 (16,508 events) on this merge and on the CI
build of b24d4163, every KalSeed compared:

Setting Result
default no fit loses FitOK; every other fit summary identical apart from the 97 KKUmu and 1 KKUe fits that main leaves unusable, including momentum and t0 (with v3.9.1 the small t0-segment shifts are gone); |dP| > 50 MeV/c 0 (main 275)
AddCaloMaterial (#2103) no fit loses FitOK
ToCaloMaterial no fit loses FitOK. Electron front-panel crossings as main in number and loss (carbon skins ~1.3, foam 0.22, pipes ~1.4 MeV/c, about 2.8 per panel, as in DocDB 58830). Inner rings: 2-5x more crossings at the same loss per crossing. On KKDmu, main's passes turn slow muons around (657 crossings with |dP| > 50 MeV/c, 12 fits lost); here none, and 1,063 RangedOut
both no fit loses FitOK; no KalSeed carries a calorimeter layer or ring twice

The production off-spill setting (10,000 cosmic events) is as before: no fit loses FitOK.

One thing for #2103, not caused by this PR: with AddCaloMaterial on, 17 KKDmu fits on reco_03 carry an inner-ring
crossing inside the fit whose material update reverses the momentum (p + dP < 0, dP down to -133 MeV/c), identically
on main. The in-fit material has no range-out check.

🤖 Generated with Claude Code

@RobMina

RobMina commented Nov 5, 2026

Copy link
Copy Markdown
Contributor Author

@FNALbuild run build test

@FNALbuild

Copy link
Copy Markdown
Collaborator

⌛ The following tests have been triggered for 9db6814: build (Build queue - API unavailable)

@oksuzian oksuzian left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

PR Review Summary — #2084

Reviewed at head 9db6814cc607f586eed31db9ad6f334acc1c8b57. Re-review of 4afa41b.

Decision

  • 🟢 approve. The merge of main keeps everything Offline#2081 added, and the calorimeter material now runs inside the sweep. I compared this head with the CI build of main b24d4163 on all 16,508 reco_03 events, with the calorimeter options off and with them on. No fit loses FitOK in any setting. With the options off, every other fit summary is identical, momentum and t0 included. The merge-order condition from my last review is met: KinKal v3.9.1 is in the envset, and the off-spill validation no longer flags any KKCHmu track. FNALbuild build 3557 passed every test at this head.

What changed since 4afa41b

  • One merge commit, 9db6814c, of main b24d4163, which brings Offline#2081, #2090 and envset p110 (KinKal v3.9.1).
  • Git merges KKFit.hh, KKTrack.hh and KKShellXing.hh without conflict. This PR's own changes to those three files are line for line the same as at 4afa41b.
  • Beyond the conflict in KKExtrap.hh, the merge commit adds new code:
    • #2081's two calorimeter passes become sweep targets (KKExtrap.hh:266-305, :348-361, :492-551);
    • ExtrapolateSweep gains notBefore (ExtrapolateSweep.hh:94, :403);
    • ExtrapolateSweepTest gains a case for notBefore.

The KKExtrap.hh resolution, checked against #2081 on main

  • 🟢 Everything #2081 asked of the extrapolation survives:
    • The material starts at the last measurement (:350-355, as main's tstart).
    • Front-panel layers already in the fit are skipped by full surface Id, and inner supports by support Id in both calorimeter containers (:357-360). These are the same two checks as main's extrapolateCaloMaterial and extrapolateCaloSupport.
    • The supports are entered only forwards in time and only moving downstream (:356, :534).
    • The thicknesses are main's: CaloMaterialLayer::thickness for the pipes, and path length times |cos| for the supports, as in CaloSupportXing::thickness.
    • The crossings go into the same generic containers main uses (addMaterialPlaneXing, addMaterialCylXing). So SaveMaterialXings and KKFit's in-fit checks, which the merge leaves untouched, behave as on main.
    • With ToCaloMaterial off, no calorimeter target is built. notBefore is NaN, and therefore inactive, for every other target.
  • 🟢 The behaviour with ToCaloMaterial on differs from main in one way, as the description says. main extends the fit first and then intersects, so it skips crossings that land on an interior piece of its own extension. In a 50-event debug run of main (KKDe and KKDmu), it printed "not at the end of the fit" 49 times: 29 inner-ring and 20 front-panel crossings. Here the sweep stops at each crossing and adds it on the end piece. The same 50 events on this head report no extrapolation failure.

Measured on reco_03 (16,508 events), this head against main b24d4163

  • 🟢 Options off. No fit loses FitOK.
    • 97 KKUmu and 1 KKUe fits that main leaves unusable become usable. 236 KKUmu fits carry RangedOut, and one KKDmu track carries ExtrapolationFailed. These are the same flag counts as in my 4afa41b run.
    • Every other fit summary is identical, including momentum and t0, apart from one KKUe track (1430:737:3070), which saves 6 more segments.
    • At 4afa41b, 13 low-momentum fits moved in t0 by 0.1-0.4 ns. With KinKal v3.9.1 on both sides, they no longer change.
    • Crossings with |dP| > 50 MeV/c go from 275 to 0. Foil crossings go from 284,229 to 284,653 and IPA crossings from 151,807 to 150,147, exactly as at 4afa41b.
  • 🟢 ToCaloMaterial on.
    • No fit loses FitOK. 11 KKDmu, 97 KKUmu and 1 KKUe fits that main leaves unusable become usable.
    • Against this head's own defaults, the option loses no fit either. It adds RangedOut to 1,063 KKDmu tracks and changes 6 other fit summaries slightly: the saved segments on 3, and t0 on 3.
    • I reran both builds with SaveMaterialXings on to see the calorimeter crossings. On KKDe, the front-panel crossings agree with main to within 1.4% per layer, with the same mean loss: 1.30, 0.22, 1.29 and 1.41 MeV/c on disk 0.
    • The inner-ring crossings are 1.3 to 4.8 times more frequent, depending on the ring and the entry, at a similar mean loss per crossing. On disk 1, the carbon-fibre ring goes from 501 to 2,408 crossings at 1.60 and 1.62 MeV/c. The hole entries go from 395 to 1,007 at 6.98 and 7.08 MeV/c.
    • On KKDmu, main has 657 calorimeter crossings with |dP| > 50 MeV/c, the worst at −8.2×10⁹ MeV/c. This head has none.
  • 🟢 Both options on.
    • No fit loses FitOK. 8 KKDmu and 107 KKUmu fits that main leaves unusable become usable, and 4 fit summaries change in momentum or t0 (see residual risk).
    • With SaveMaterialXings, on 2 of the 6 chunks, no KalSeed carries any calorimeter layer or ring twice. No inner-support ring is crossed more than twice. The same holds on all 16,508 events with ToCaloMaterial alone.

Prior findings

  • 🟢 All five findings from my first review stay FIXED, as verified at 4afa41b. The merge does not touch ExtrapolateSweep.hh's grazing and failure handling, KKFit::createSeed or the Regrow modules.
  • 🟢 [was the merge-order condition] FIXED.
    • KinKal v3.9.1 contains KFTrack/KinKal#232. On 9,314 off-spill cosmic events with Validation/offSpill.fcl, this head flags no KKCHmu track ExtrapolationFailed. At 4afa41b without #232, it flagged 367 of 4,876.
    • With the production off-spill setting, no fit loses FitOK and 22 become usable. The 382 KKCHmu momenta that differ from main are the same count as at 4afa41b.
    • The #2081 conflict is resolved. The head merges cleanly with current main, which is still b24d4163.

Findings

  1. ⚪ [S3] The calorimeter rework, about 200 lines, lives only in the merge commit. git log -p and GitHub's commit view show a merge without its diff, so a later reader has to know to look there. The PR comment and the description explain it well, which is enough here. Next time, a separate commit on top of the merge would keep it visible.

Validation check

  • Build/tests run: yes.
    • A full muse build of this head (Production 94a4d81f, mu2e-trig-config 5999ffa2, p110) finished with exit status 0 and no warnings, and ExtrapolateSweepTest reports "all checks passed".
    • reco_03 ran in five settings on both builds: defaults, ToCaloMaterial, both options, and the last two again with SaveMaterialXings (both options on 2 of the 6 chunks only). That is 52 chunks, all with exit status 0. The off-spill runs used both the validation and the production settings, on both builds. There were also two 50-event debug runs.
  • Config contract check: pass. No fcl or parameter changes in the increment.
  • Cross-repo consistency: pass.

Residual risk

  • Not run: AddCaloMaterial alone (the Offline#2103 setting), a Regrow job, the KinematicLine comparison, and resolution against truth with the calorimeter options on.
  • With both options on, one KKDe fit (1430:967:368, fit consistency 0.28) reports a t0 137 ns later than on main, with the opposite direction and 0.3 MeV/c less momentum. The sweep changes the extrapolated trajectory, and the reported t0 seems to move to another piece. I did not trace it.

Author follow-ups

  • None required.

@brownd1978 brownd1978 left a comment •

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This PR replaces the individual component extrapolation, tested in an hard-coded order, with a general sweep, which is a clear improvement. It also adds the physical concept of ranging out, and of extrapolation algorithmic failure. It also adds a diagnostic extrapolation test. It is impossible to manually review the code correctness given the refactoring of 8 separate algorithms/files into 1, that must be done through validation tests.
My main feedback is that the newly introduced concepts (limit, seam, ...) and the overall structure and flow of the sweep need to be described in appropriate comments. I suggest also a comment on how, in future, what functions would need to be updated when introducing new passive material.
Finally, I note that, outside the scope of this PR, extrapolation is still using some hard-coded numbers (foil thickness, ipa thickness) that should come from geometry service.

Comment thread Mu2eKinKal/inc/KKTrack.hh Resolved
Comment thread Mu2eKinKal/inc/KKTrack.hh Resolved
Comment thread Mu2eKinKal/inc/KKFit.hh Resolved
Comment thread Mu2eKinKal/inc/ExtrapolateSweep.hh Resolved
Comment thread Mu2eKinKal/inc/ExtrapolateSweep.hh Resolved
Comment thread Mu2eKinKal/inc/ExtrapolateSweep.hh Outdated
bool withinReach(size_t itg, KinKal::VEC3 const& pos, double reach) const;
template <class KTRAJ> bool search(KinKal::ParticleTrajectory<KTRAJ> const& ptraj, double tend) const;
template <class KTRAJ> bool firstCrossing(KinKal::ParticleTrajectory<KTRAJ> const& ptraj, KTRAJ const& piece, size_t itg, KinKal::TimeRange win, KinKal::Intersection& inter) const;
template <class KTRAJ, class VISIT> bool visitPlaneExtrema(KTRAJ const& piece, KinKal::Plane const& plane, KinKal::TimeRange const& range, VISIT&& visit) const;

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I am curious why these functions exist only for planes: are they not needed as well for cylinders and other shapes?

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In the case of planes, there is a closed form for the extrema of the normal velocity component of a helix (or line). So it is possible to test cheaply for intersections of the trajectories with planes. I found in testing that KinKal's generic intersection with planes sometimes returned a root outside the requested time window, which is solved using these cheap tests. So far in testing that same issue has not occurred with the cylinder intersections, and if it were to come up the best place for the fix would be in KinKal's intersection, for both shapes.

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the explanation. A plane intersection claiming to be in time when it is not is an error. Please add an issue to KinKal and comment that these functions are temporary patches.

@RobMina

RobMina commented Nov 6, 2026

Copy link
Copy Markdown
Contributor Author

Thanks Dave! The new commit addresses your inline concerns and adds some clarifying comments.

It also removes the "seams" logic, which will be addressed in a followup KinKalGeomMaker PR. And the trajectory truncation will be addressed in a separate KinKal PR.

Validation of this commit against 9db6814 (same inputs as before):

  • nightly reco_03, 2,000 events: every fit summary in every collection identical, no fit loses FitOK;
  • production off-spill setting, 3,334 events: every KalSeed identical except one ranged-out KKCHmu, whose last segment now extends past the stop (no clipping); every crossing identical;
  • off-spill cosmics with CRV extrapolation, 10,000 events each with field correction on and off, against the PR's earlier build: every ntuple leaf identical except on 4 tracks. One ranged-out track gains 9 tracker-surface crossings after its stop, none in truth (the removed clipping is what dropped them). Two tracks cross both CRV-T sectors, or both strongbacks, at one point in the band where the envelopes overlap, and get both materials (see the seam reply). On one field-off track the second strongback is used up there, so a crossing 1.4 m away, where the field-off helix comes back to the strongback plane, is no longer recorded.

@RobMina

RobMina commented Nov 6, 2026

Copy link
Copy Markdown
Contributor Author

@FNALbuild run build test

@oksuzian oksuzian left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

PR Review Summary — #2084

Reviewed at head 445cd49a915e429d71d59dc8d84e87721ada4872. Re-review of 9db6814.

Decision

  • 🟡 comment now, approve when green. I would approve this code. The new commit answers all of Dave's points, and his main request, comments on the sweep's concepts and flow and on how to add new material, is met. On all 16,508 nightly reco_03 events every printed KalSeed is identical to 9db6814c. The two behaviours the commit gives up, the seam de-duplication and the clipping at the range-out point, change 9 of 4,876 KKCHmu tracks on my off-spill sample with the validation setting (6 seam pairs, 3 post-stop) and 3 of 5,384 with the production setting, and the description documents both as known limitations (finding 1). FNALbuild has not built this head: at 03:34 UTC mu2e/buildtest still said "This test has not been triggered yet", and the 03:00 trigger had no reply. I am triggering it again and will approve once the build at 445cd49a is green.

What changed since 9db6814

  • One commit, "Remove seams, ranged-out track trimming, add clarifying comments" (7 files, +121 −165). It merges cleanly with main, which is still b24d4163.
  • KKTrack records range-out as a bool, and nothing clips segments, sampling or the trajectory at the stop any more (KKFit.hh:873, :1044, CentralHelixFit_module.cc:442, KinematicLineFit_module.cc:381, :395).
  • The seam logic is gone. A plane, CRV sector or foil is now guarded only against re-crossing itself until the track leaves its tolerance slab (ExtrapolateSweep.hh:179-180, :426). Coplanar neighbours are each crossed.

Dave's review

  • 🟢 His main request is met. ExtrapolateSweep.hh:1-13 now describes the targets, the cursor, the search, the accept step, the stop conditions, limit/unlimit, the re-crossing guard and the closed-form plane roots. KKExtrap.hh:41-43 says what to add for new passive material. Every new member and method has a one-line comment.
  • 🟢 The inline points are answered in the code or on the thread:
    • extrapfails_ is diagnostic text. The physical condition reaches the KalSeed as ExtrapFailed, set in KKFit::createSeed (KKFit.hh:770-773), and the comment now says so (KKTrack.hh:199).
    • The range-out time is replaced by a bool, as he suggested (KKTrack.hh:158, :175, :200).
    • The clipping he asked about is removed. The trajectory is not truncated at the stop, because KinKal's Track cannot shorten it; see finding 1.
    • limit/unlimit and the high-water mark are explained in the header and at :102-103.
    • Seams are removed, with the overlapping geometry left to a follow-up; see finding 1.
    • The plane-only extremum functions are explained in the reply and at :12.
  • His note on the hard-coded foil and IPA thicknesses he placed outside this PR, and the commit does not touch it.

Measured, this head against 9db6814 and against main b24d4163

  • 🟢 Nightly reco_03, all 16,508 events, defaults. Every printed KalSeed in every collection matches 9db6814c: track counts, fit summaries, flags (236 KKUmu RangedOut, one KKDmu ExtrapolationFailed) and every intersection. So the comparison with main is unchanged from my last review: 97 KKUmu and 1 KKUe fits become usable and none loses FitOK. I did not compare segment time ranges, which the print does not show.
  • 🟢 Off-spill cosmics, 9,314 events, validation and production settings. No fit summary changes except cos on ranged-out KKCHmu tracks, by at most 0.002 (2 tracks in validation, 5 in production). Flags and track counts are the same, and no crossing is lost. The changes are the two cases in finding 1.
  • 🟢 ExtrapolateSweepTest reports "all checks passed". A full muse build of this head (Production 94a4d81f, mu2e-trig-config 5999ffa2, p110) finished with exit status 0 and no warnings.

Findings

  1. 🟡 [S2] The seam double counting and the post-stop crossings are documented limitations, but the follow-ups are not yet tracked, and the double counting is more frequent on my sample than the description says.

    • Seams. With the validation setting, 6 of 4,876 KKCHmu tracks record two coplanar CRV crossings at the same time and place, each with its full material: 5 strongback pairs at 4.3-5.3 MeV/c each, and one CRV-T1/T2 pair at 71.5 MeV/c each (1430:0:90791, 2.1 GeV/c track). With the production setting, it is 3 of 5,384: 2 strongback pairs and one CRV-T1/T2 pair at 113 MeV/c each (1430:0:50204, 1.1 GeV/c). main and 9db6814c have none. Only the momentum from the doubled crossing onwards changes: the CRV-T positions move by at most 0.003 mm compared with 9db6814c. The description quotes 2 tracks in 20,000 events and 64 MeV/c per sector.
    • Post-stop crossings. With the validation setting, 3 of the 52 ranged-out KKCHmu tracks gain 5, 10 and 12 sampled tracker crossings after the stop (1430:0:3926, 19224, 24052), and the KalSeed's last segment runs past the stop. The production setting adds none. The RangedOut flag marks these tracks, and main has no range-out at all, so this is not a regression against main.
    • Suggested fix: the description has <follow-up PR link> as a placeholder, and neither follow-up exists yet. Please open the KinKalGeomMaker issue or PR, and the KinKal truncation one, and link them, so the limitations stay tracked after the merge.
  2. ⚪ [S3, carried over] The calorimeter rework from the merge of main still lives only in the merge commit 9db6814c.

Validation check

  • Build/tests run: yes. One muse build of this head, ExtrapolateSweepTest, reco_03 on all 16,508 events (6 chunks) and the 9,314-event off-spill sample in both settings. All jobs exited with status 0. The 9db6814c and main outputs are from my last review, with the same job files and inputs.
  • Config contract check: pass. No fcl or parameter changes.
  • Cross-repo consistency: pass.

Residual risk

  • Not checked at this head: the calorimeter options, a Regrow job, the KinematicLine comparison and resolution against truth. The commit changes the KinematicLine and CentralHelix sampling only for ranged-out tracks.
  • The field-off case in the description, where a doubled strongback uses up a later real crossing, does not occur on my sample.

Author follow-ups

  1. Open and link the two follow-ups (finding 1).

@oksuzian

oksuzian commented Nov 6, 2026

Copy link
Copy Markdown
Collaborator

@FNALbuild run build test

@oksuzian

oksuzian commented Nov 6, 2026

Copy link
Copy Markdown
Collaborator

@FNALbuild run build test

@FNALbuild

Copy link
Copy Markdown
Collaborator

⌛ The following tests have been triggered for 445cd49: build (already triggered: build) (Build queue - API unavailable)

@FNALbuild

Copy link
Copy Markdown
Collaborator

☀️ The build tests passed at 445cd49.

Test Result Details
test with ✅ Command did not list any other PRs to include
merge ✅ Merged 445cd49 at b24d416
build (prof) ✅ Log file. Build time: 04 min 33 sec
ceSimReco ✅ Log file.
g4test_03MT ✅ Log file.
transportOnly ✅ Log file.
POT ✅ Log file.
g4study ✅ Log file.
cosmicSimReco ✅ Log file.
cosmicOffSpill ✅ Log file.
ceSteps ✅ Log file.
ceDigi ✅ Log file.
muDauSteps ✅ Log file.
ceMix ✅ Log file.
rootOverlaps ✅ Log file.
g4surfaceCheck ✅ Log file.
trigger ✅ Log file.
check_cmake ✅ Log file.
FIXME, TODO ➡️ TODO (18) FIXME (4) in 18 files
clang-tidy ➡️ 2 errors 281 warnings
whitespace check ✅ no whitespace errors found

N.B. These results were obtained from a build of this Pull Request at 445cd49 after being merged into the base branch at b24d416.

For more information, please check the job page here.
Build artifacts are deleted after 5 days. If this is not desired, select Keep this build forever on the job page.

@oksuzian oksuzian left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

PR Review Summary — #2084

Reviewed at head 445cd49a915e429d71d59dc8d84e87721ada4872. Follow-up to my comment review at the same head.

Decision

  • 🟢 approve. FNALbuild build 3561 has now built this head, merged on main b24d4163, and every test passed: the prof build, all eleven validation jobs, rootOverlaps, g4surfaceCheck, trigger, check_cmake and the whitespace check. The FIXME/TODO count and clang-tidy rows are informational. This is the approval my comment review promised once the build was green. The two findings in that review still stand and do not block: please open and link the seam and truncation follow-ups, and note that the calorimeter rework lives only in the merge commit.

@brownd1978 brownd1978 left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for making the updates! I'm approving this, with the request that the plane intersection failure be recorded as a KinKal issue. Perhaps the traj truncation mentioned as well.

@RobMina

RobMina commented Nov 6, 2026

Copy link
Copy Markdown
Contributor Author

Here is the issue for the KinKal plane intersection bug: KFTrack/KinKal#233

And the truncation KinKal PR: KFTrack/KinKal#234

And followup Offline PR: #2111

@brownd1978 brownd1978 left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for addressing my comments.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants