Issue Report

# Issue Report

(used GPT-5.4 and Opus 4.7)

**Scope:** Full pipeline audit across `EventDisplay_v4` (C++), `Eventdisplay_AnalysisScripts_VTS`
(shell orchestration), `Eventdisplay_AnalysisFiles_VTS` (auxiliary files), and `Eventdisplay-ML`
(XGB/Python ML). Log files and release-test artifacts were also reviewed.

**Severity labels:** `HIGH` — incorrect physics output or silent job failure; `MEDIUM` —
degraded reliability, reproducibility, or correctness in edge cases; `LOW` — cosmetic,
misleading diagnostics, or minor fragility.

---

## Pipeline overview (reference)

| Stage | IRF chain entry | Data-analysis entry |
|---|---|---|
| Image reconstruction | `IRF.evndisp_MC.sh` → `evndisp` | `ANALYSIS.evndisp.sh` → `evndisp` |
| Stereo / table lookup | `IRF.mscw_energy_MC.sh` → `mscw_energy` | `ANALYSIS.mscw_energy.sh` → `mscw_energy` |
| Angular reco (ML) | DispBDT / XGB training (`IRF.train*`) | DispBDT XMLs consumed by `mscw_energy` |
| Gamma/hadron (ML) | BDT / XGB training, optimisation, effective areas | TMVA/XGB friend trees consumed by `anasum` |
| Science analysis | — | `ANALYSIS.anasum*.sh` → `anasum` |
| DL3 export | — | `ANALYSIS.v2dl3.sh` → `v2dl3` |

C++ class chains:
- **`evndisp`:** `VReadRunParameter` → `VEventLoop` → `VTraceHandler`/`VFitTraceHandler`, `VCalibrator`, `VImageAnalyzer`, `VArrayAnalyzer`, `VPedestalCalculator`
- **`mscw_energy`:** `VTableLookupRunParameter` → `VTableLookup` → `CData`, `VDispAnalyzer`, `VTMVADispAnalyzer`, `VEmissionHeightCalculator`
- **`anasum`:** `VAnaSumRunParameter` → `VAnaSum` → `VStereoAnalysis` → `CData`, `VGammaHadronCuts`, `VTMVAEvaluator`
- **`makeEffectiveArea`:** `VGammaHadronCuts` → `CData` → `VInstrumentResponseFunction` → `VInstrumentResponseFunctionData`

---

## A — C++ algorithm defects

### A1 · HIGH · Out-of-bounds read in pulse-timing extraction
**Location:** `EventDisplay_v4/src/VTraceHandler.cpp:346-358`  
`getPulseTiming()` interpolates on the rising edge using `fpTrace[i+1]`. When the pulse maximum
lands on the last sample (`maxpos = size-1`), `i+1` is out of bounds.  
**Impact:** Truncated or late pulses produce undefined timing values in the signal characterisation
path.

### A2 · HIGH · Sliding-window charge branch returns `FADC[1]` instead of integrated charge
**Location:** `EventDisplay_v4/src/VTraceHandler.cpp:745-757`  
`calculateTraceSum_slidingWindow()` computes `charge` and updates the integration-window metadata
but then returns `FADC[1]`.  
**Impact:** When the raw sliding-window branch is active, charge extraction is silently wrong.

### A3 · MEDIUM · Fitted-trace T0 mixes pedestal-subtracted and absolute amplitude conventions
**Location:** `EventDisplay_v4/src/VFitTraceHandler.cpp:433-442`  
`getTraceMax()` returns a pedestal-shifted peak, but `getTraceTZero()` solves for
`-(imax/2 + fPed)`, re-adding the pedestal to the half-height target.  
**Impact:** Fitted pulse-arrival times are biased whenever the pedestal is non-zero.

### A4 · HIGH · Low-gain average-T0 assignment always writes the high-gain field
**Location:** `EventDisplay_v4/src/VCalibrationData.cpp:365-373`  
`setAverageTZero(bool iLowGain)` unconditionally assigns `fAverageTZero_highgain`, even when
`iLowGain == true`.  
**Impact:** The low-gain timing calibration path corrupts the high-gain average-T0 used later in
analysis.

### A5 · HIGH · Low-gain average-T0 calculation uses the high-gain sum path
**Location:** `EventDisplay_v4/src/VCalibrator.cpp:899-903`  
`calculateAverageTZero(bool iLowGain)` calls `calcSums(..., iLowGainOnly=false)` unconditionally.  
**Impact:** Low-gain timing constants are derived from the wrong charge-extraction mode.

### A6 · HIGH · Shower-core validity check tests `ximp` twice, never `yimp`
**Location:** `EventDisplay_v4/src/VArrayAnalyzer.cpp:944-951`  
`fillShowerCore()` evaluates `!isnormal(ximp) || !isnormal(ximp)`; the duplicate check means an
invalid `yimp` is never screened out.  
**Impact:** NaN or invalid core-Y solutions propagate into core distances, energy reconstruction
inputs, and event-selection variables.

### A7 · HIGH · Method-4 stereo pair-angle rejection applies cut to the wrong telescope pair
**Location:** `EventDisplay_v4/src/VArrayAnalyzer.cpp:1241-1248`  
The per-pair rejection computes the current pair in `mm[0]`/`mm[1]` but evaluates the cut using
`m[0]`/`m[1]`.  
**Impact:** For events with more than two images, pair-angle rejection is applied to the wrong
telescopes, biasing the weighted line intersections used for direction reconstruction.

### A8 · HIGH · Log-likelihood image fit can be initialised with invalid centroid bounds
**Location:** `EventDisplay_v4/src/VImageParameterCalculation.cpp:1542-1555,1588-1601`  
For edge/degenerate images with `sigmaX <= 0` or `sigmaY <= 0`, only the upper Minuit bound is
set; the lower bound retains its sentinel value before being passed to the minimiser.  
**Impact:** Edge images can enter the recovery LL fit with invalid parameter bounds, degrading
parameterisation in the regime where the fit is intended to stabilise results.

### A9 · MEDIUM · Muon-ring image-size correction divides by zero at the first grid point
**Location:** `EventDisplay_v4/src/VImageParameterCalculation.cpp:737-748`  
`xi_tmp = i / numSteps` is zero when `i == 0`, but `kTest_tmp = ... / xi_tmp` is evaluated before
any guard.  
**Impact:** Muon calibration correction can produce `inf`/`NaN` candidates and an unstable
correction factor.

### A10 · MEDIUM · Emission-height pair weights undefined for small image sizes
**Location:** `EventDisplay_v4/src/VEmissionHeightCalculator.cpp:60-64`  
The weight formula `1./((1./log10(size[i])) + (1./log10(size[j])))` is not guarded against
`size <= 1`.  
**Impact:** Small-image events receive singular, negative, or ill-defined pair weights in the
emission-height estimate.

### A11 · MEDIUM · `VEmissionHeightCalculator` accumulates telescope positions across reinitalisations
**Location:** `EventDisplay_v4/src/VEmissionHeightCalculator.cpp:100-133`  
Both `setTelescopePositions()` overloads only `push_back()` and never clear previous contents.  
**Impact:** Reusing the same calculator instance duplicates telescope geometry, making pairwise
shower-height geometry wrong.

### A12 · HIGH · Core-resolution IRF formula uses `Xcore` twice instead of `Xcore` and `Ycore`
**Location:** `EventDisplay_v4/src/VInstrumentResponseFunctionData.cpp:472-485`  
The core-resolution error is computed as `sqrt(Xcore²+Xcore²) - sqrt(MCxcore²+MCycore²)`.  
**Impact:** Reconstructed core-error quantities are biased whenever `Ycore` differs from `Xcore`;
this is a direct physics error in the IRF-resolution path.

### A13 · HIGH · Gamma/hadron TMVA training has an off-by-one zenith-bin bounds check
**Location:** `EventDisplay_v4/src/trainTMVAforGammaHadronSeparation.cpp:255-266`  
The check uses `< iZenithBin` instead of `<= iZenithBin`, so `iZenithBin == size()` passes and is
then dereferenced.  
**Impact:** Certain bin configurations index beyond the configured zenith-bin vectors during
training.

### A14 · HIGH · Reconstruction-quality TMVA training configured as a classification factory
**Location:** `EventDisplay_v4/src/trainTMVAforGammaHadronSeparation.cpp:343-360`  
The non-gamma/hadron branch adds a regression target but creates the `TMVA::Factory` with the
default (classification) options string.  
**Impact:** The code path described as reconstruction-quality training is algorithmically
mismatched to the TMVA factory configuration.

### A15 · MEDIUM · Angular-reconstruction training silently rescales the requested train/test split
**Location:** `EventDisplay_v4/src/trainTMVAforAngularReconstruction.cpp:125-138`  
After deriving `ntrain` and `ntest` from the requested fraction, both are multiplied by `0.8`.  
**Impact:** Effective train/test statistics differ from the configured values, reducing
reproducibility.

### A16 · MEDIUM · Reduced-tree subsampling normalised by file count, not event population
**Location:** `EventDisplay_v4/src/trainTMVAforGammaHadronSeparation.cpp:147-157`  
The retained fraction is divided by `iTreeVector.size()` (number of input files), not by event
counts.  
**Impact:** Training preselection depends on file granularity and can distort class composition
across uneven file splits.

### A17 · HIGH · Sliding-window charge extraction reads one past the end of the FADC array
**Location:** `EventDisplay_v4/src/VTraceHandler.cpp:725-783`  
`iSearchEnd` is clamped to `n - iIntegrationWindow + 1`. At the final loop iteration
(`i = iSearchEnd - 1`), the expression `FADC[i + iIntegrationWindow]` evaluates to `FADC[n]`,
which is one past the end of the valid array range `[0, n-1]`.  
**Impact:** Undefined behaviour on every waveform where the sliding window reaches its rightmost
position; corrupts charge reconstruction silently on affected events.

### A18 · MEDIUM · `VTraceHandler::getQuickMax` uses the previous call's trace maximum for saturation counting
**Location:** `EventDisplay_v4/src/VTraceHandler.cpp:420-426`  
`nMax` is computed as `(int)(fDynamicRange * tmax)` before `tmax` is reset to −10000. If the
caller passes an uninitialised or stale value, the saturation counter `n255` is thresholded
against the wrong maximum.  
**Impact:** The number of saturated samples can be over- or under-estimated, misclassifying
pulses as saturated or unsaturated and biasing image-cleaning amplitude thresholds.

---

## B — C++ infrastructure defects

### B1 · MEDIUM · `CData::getXGBTree()` assumes `.root` is present in the filename
**Location:** `EventDisplay_v4/src/CData.cpp:1073-1086`  
The XGB friend-file name is rewritten via `string::replace(find(".root"), ...)` without checking
whether `.root` exists in the name.  
**Impact:** The class API is brittle and can generate an invalid replacement if reused with a
non-standard input name.

### B2 · MEDIUM · XGB gamma/hadron C++ interface hard-wired to `Is_Gamma_70`
**Location:** `EventDisplay_v4/src/CData.cpp:1118-1127`; `EventDisplay_v4/src/VGammaHadronCuts.cpp:727-731,1118-1128`  
The C++ consumer binds only `Gamma_Prediction` and `Is_Gamma_70`; the classification operating
point is baked into the upstream friend-tree producer, not enforced by `VGammaHadronCuts`.  
**Impact:** The 70% efficiency threshold cannot be adjusted from cut files or runtime parameters,
discarding all the optimisation flexibility of the XGB classifier.

### B3 · MEDIUM · `VTMVAEvaluator` recovers feature names by brittle XML string slicing
**Location:** `EventDisplay_v4/src/VTMVAEvaluator.cpp:62-116`  
Variable recovery depends on exact XML formatting and field ordering (`NVar`, `Expression`,
`Label`, `SpecIndex`) rather than a structured parser.  
**Impact:** A TMVA XML layout change can silently alter which variables the evaluator reads,
turning the deployed BDT interface into a schema-fragile contract.

### B4 · MEDIUM · TMVA weight grid initialisation tolerates missing interior bins with warnings
**Location:** `EventDisplay_v4/src/VTMVAEvaluator.cpp:190-279`  
Missing weight files at non-edge bins emit warnings and allow the evaluator to continue.  
**Impact:** Incomplete TMVA weight grids survive initialisation, making production behaviour
depend on partial bin coverage instead of a strict consistency check.

### B5 · LOW · `VTMVARunData` sanity message does not match the actual condition
**Location:** `EventDisplay_v4/src/VTMVARunData.cpp:795-800`  
Prints "need at least two energy bins" while the code only rejects `< 1` bin.  
**Impact:** Misleading diagnostics when debugging training-configuration problems.

### B6 · MEDIUM · `VInstrumentResponseFunction` mutates shared MC azimuth in place
**Location:** `EventDisplay_v4/src/VInstrumentResponseFunction.cpp:193-199`  
MC azimuth normalisation is applied as `fData->MCaz -= 360.` on the shared event object.  
**Impact:** Weak implementation pattern; mutates shared state instead of using a local normalised
copy.

### B7 · MEDIUM · `makeEffectiveArea` IRF duplication is order-dependent
**Location:** `EventDisplay_v4/src/makeEffectiveArea.cpp:162-208,194-208`  
IRF products are duplicated by index via `getDuplicationID()` and only work if the source has
already been initialised earlier in the vector.  
**Impact:** A future refactor changing the internal IRF ordering will silently break resolution
product generation.

### B8 · MEDIUM · `copyMCHistograms()` keeps the first ROOT file open
**Location:** `EventDisplay_v4/src/makeEffectiveArea.cpp:370-401`  
The first `MChistos` object is borrowed directly from the first input file, which is never closed
inside the helper.  
**Impact:** File-handle lifetime is coupled to the returned object, risking resource leaks or
dangling-file state.

### B9 · MEDIUM · `anasum` contains a stale unreachable `analysisType == 5` branch
**Location:** `EventDisplay_v4/src/anasum.cpp:47-54,106-108`  
`analysisType` is explicitly limited to 3 and 4, but the `VAnaSum::doStereoAnalysis()` handoff
still tests `analysisType == 5`.  
**Impact:** Dead code path; sign of an incomplete control-flow change.

### B10 · MEDIUM · Epoch reevaluation at runtime means analysis behaviour changes with epoch-file updates
**Location:** `EventDisplay_v4/src/VReadRunParameter.cpp:463-470`; `EventDisplay_v4/src/printRunParameter.cpp:100-117`  
`printRunParameter updated-runinfo` and `evndisp -epochfile` reevaluate epoch and atmosphere from
`VERITAS.Epochs.runparameter` each run rather than relying solely on the file-internal metadata.  
**Impact:** Reprocessing the same intermediate ROOT file with an updated epoch file can silently
change the science output.

### B11 · HIGH · `VOnOff` mutates the background histogram after using it, corrupting reflected-region normalisation
**Location:** `EventDisplay_v4/src/VOnOff.cpp:159-161`  
`doOnOffforParameterHistograms` computes the excess as
`hTemp->Add(hon, hoff, 1., -1.*i_norm_alpha)` (which consumes `hoff` at its original scale) and
then immediately calls `hoff->Scale(i_norm_alpha)`. Any caller that reuses `hoff` after this call
sees a rescaled background histogram.  
**Impact:** The reflected-background normalisation in the `anasum` analysis path (via `VAnaSum.cpp:556`)
can be doubly-scaled, producing wrong excess counts and significance values.

### B12 · HIGH · `makeEffectiveArea` comments that the MC run header is optional but hard-exits when it is missing
**Location:** `EventDisplay_v4/src/makeEffectiveArea.cpp:118`; `EventDisplay_v4/src/VInstrumentResponseFunctionRunParameter.cpp:430-437`  
The comment at line 118 states "might not be there, no problem", yet `readMCRunHeader()` calls
`exit(EXIT_FAILURE)` when no `MC_runheader` object is found in the input file.  
**Impact:** Misleading documentation causes operators to expect graceful handling; actual behaviour
is a hard crash with a non-zero exit code when the MC header is absent (e.g., for older files).

### B13 · HIGH · `VInstrumentResponseFunctionRunParameter` silently ignores unknown run-parameter keys
**Location:** `EventDisplay_v4/src/VInstrumentResponseFunctionRunParameter.cpp:109-358`  
The parser is a chain of `else if` branches ending at `RERUN_STEREO_RECONSTRUCTION_3TEL` with
no terminal `else` and no warning message. Unknown or misspelled keys are silently dropped.  
**Impact:** Typos in `makeEffectiveArea` parameter files produce no diagnostic and default values
are used; an operator has no way to detect that a setting was not applied.

### B14 · MEDIUM · `printRunParameter` exits with code 0 on file-open failure
**Location:** `EventDisplay_v4/src/printRunParameter.cpp:330-335`  
`new TFile(argv[1])` can return a zombie object; the code prints "exiting..." and calls `exit(0)`.  
**Impact:** Shell wrappers that check only the exit code (C6, several IRF helpers) will interpret
the empty output as a missing field rather than a fatal error, silently propagating bad metadata
through the pipeline.

### B15 · LOW · `VTableLookupRunParameter` error message says "2=read" but the code only accepts "0=read"
**Location:** `EventDisplay_v4/src/VTableLookupRunParameter.cpp:162-176`  
The `-fill` option accepts `0` (read) or `1` (fill); any other value triggers the message
"unknown parameter, choose 1=fill or **2=read** lookup tables", which is incorrect.  
**Impact:** Operators following the error message will pass `-fill=2`, which also triggers the
error in a confusing loop.

---

## C — Shell orchestration bugs

### C1 · HIGH · `ANALYSIS.anasum.sh` never submits any jobs
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/ANALYSIS.anasum.sh:90-92`  
Line 90 assigns the helper path as a string; the subsequent `eval` only echoes it back. Every
other equivalent script correctly uses backticks to execute the helper:
```bash
# Broken (line 90):
SUBC="$(dirname "$0")/helper_scripts/UTILITY.readSubmissionCommand.sh"
SUBC=`eval "echo \"$SUBC\""`    # still a string, never executed

# Correct (evndisp.sh:126):
SUBC=`$(dirname "$0")/helper_scripts/UTILITY.readSubmissionCommand.sh`
```
All `*qsub*`/`*condor*`/`*sbatch*`/`*simple*` tests fail, so generated job scripts are written
but never submitted.  
**Impact:** The full-runlist `anasum` path silently drops all jobs with no error message.

### C2 · HIGH · `get_disp_dir()` checks an undefined variable, always selects the wrong DispBDT directory
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/helper_scripts/ANALYSIS.mscw_energy_sub.sh:136-152`  
The function checks `${SIMTYPE} == "CARE_RedHV"` to decide between 55-deg and 60-deg DispBDT
fallback models. `SIMTYPE` is never assigned in this script; only `SIMTYPE_RUN` is set (lines
82–93). A secondary issue: the current value since v491 is `CARE_RedHV_Feb2024`, not
`CARE_RedHV`, so the comparison would fail even if the variable were correct. The proper guard
should use `HVSETTINGS == "obsLowHV"`.  
**Impact:** For all RedHV (and high-zenith UV) runs, `mscw_energy` loads 60-deg DispBDT models
instead of the 55-deg fallback, degrading direction reconstruction quality for runs at zenith ≥ 58°.

### C3 · MEDIUM · `FORCEDATMO` atmosphere override is silently ignored in cluster submissions
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/ANALYSIS.mscw_energy.sh:63`; `helper_scripts/ANALYSIS.mscw_energy_sub.sh:71`  
The parent script captures `FORCEDATMO=$6` locally. The sed-template substitution block does not
inject `FORCEDATMO` into the generated job script, and the job script references
`${FORCEDATMO:-...}` as an unexported shell variable, which is empty in batch environments.  
**Impact:** The documented `[ATM]` override parameter does nothing in production; users
attempting to force an atmosphere ID get silently wrong results.

### C4 · MEDIUM · `mscw_energy_sub.sh` missing atmosphere-ID normalisation for V4/V5 data
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/helper_scripts/ANALYSIS.mscw_energy_sub.sh:68-71`; `ANALYSIS.anasum_sub.sh:74-83`  
`mscw_energy_sub.sh` calls `printRunParameter … updated-runinfo`, which may return a modernised
6x atmosphere ID for old V4/V5 observations. `anasum_sub.sh` compensates with
`prepare_atmo_string()` (`${ATMO/6/2}` for V4/V5); `mscw_energy_sub.sh` has no equivalent
conversion.  
**Impact:** `mscw_energy` exits with "table file not found" for V4/V5 GRISU data when the epoch
files return a modernised atmosphere ID.

### C5 · MEDIUM · `DISPBDT` guard in `mscw_energy_sub.sh` is missing `$`
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/helper_scripts/ANALYSIS.mscw_energy_sub.sh:193`  
`if [[ DISPBDT != "NOTSET" ]]; then` references the literal string `DISPBDT` instead of
`$DISPBDT`, so the condition is never conditionally guarded.  
**Impact:** The log reports DispBDT XML provenance even when that branch is not actually active.

### C6 · MEDIUM · `printRunParameter -elevation` field is parsed positionally in the helper
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/helper_scripts/ANALYSIS.mscw_energy_sub.sh:127`; `EventDisplay_v4/src/printRunParameter.cpp:340-343`  
The helper extracts the zenith angle by `awk` field number from a human-readable CLI printout
rather than from a stable machine-readable interface.  
**Impact:** Fragile table and DispBDT directory selection; breaks silently if the printout format
changes.

### C7 · MEDIUM · `mscw_energy` hard-fails when a TMVA XML family is missing, no soft fallback
**Location:** `EventDisplay_v4/src/VTMVADispAnalyzer.cpp:97-105,164-170`  
When TMVA XML prefixes are provided and one family is missing, the code exits without graceful
degradation.  
**Impact:** Missing or incompletely copied DispBDT XMLs stop the run.

### C8 · MEDIUM · `anasum` default input directory hard-wired to `RecID0` with no consistency check
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/ANALYSIS.anasum.sh:47-49`; `helper_scripts/ANALYSIS.anasum_sub.sh:213-226`  
The default input directory is `.../RecID0`; no validation that the `mscw.root` files were
produced with the requested reconstruction ID.  
**Impact:** A workflow that reconstructs with a non-zero RecID can run `anasum` against stale
`RecID0` products, producing plausible but inconsistent science outputs.

### C9 · HIGH · `ANALYSIS.evndisp_sub.sh` consumes DBTEXT-derived files without existence checks
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/helper_scripts/ANALYSIS.evndisp_sub.sh:97-107,137-166`  
After unpacking the DBTEXT tarball, the helper immediately `cat`s `${TMP_LASERRUN}` and parses
`${RUN}.runinfo` without checking that either file exists or is readable.  
**Impact:** Missing or corrupt DBTEXT tarballs silently propagate empty laser-run metadata or
wrong run dates into CVBF lookup and calibration selection.

### C10 · HIGH · Broken TMVA bin-parsing regex in training and optimisation scripts
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/IRF.trainTMVAforGammaHadronSeparation.sh:107-109`; `IRF.optimizeTMVAforGammaHadronSeparation.sh:91-93`  
`grep "^* ENERGYBINS"` and the matching `sed` expressions are invalid or broken (`*` is not
a valid regex anchor in this position).  
**Impact:** Training and optimisation bin parsing can fail or misbehave, directly affecting
gamma/hadron separation products.

### C11 · MEDIUM · Cut optimisation continues with missing effective-area input
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/IRF.optimizeTMVAforGammaHadronSeparation.sh:81-83`  
Logs an error on missing effective-area file but does not exit.  
**Impact:** Optimisation jobs may run with incomplete inputs, producing invalid cut tuning.

### C12 · MEDIUM · DL3 jobs install Python dependencies at runtime per job
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/helper_scripts/ANALYSIS.v2dl3_sub.sh:56-58`  
Each job activates the base environment and runs `pip install -e ...`.  
**Impact:** Slow, non-reproducible, and environment-dependent production.

### C13 · MEDIUM · `ANALYSIS.anasum_combine.sh` lacks a completion/integrity guard before merge
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/ANALYSIS.anasum_combine.sh:31-36,85-115`  
Assumes parallel jobs are finished; no strong completeness validation before combining.  
**Impact:** Partial combined outputs can be produced silently.

### C14 · MEDIUM · Data-side `mscw_energy` selects IRF products purely from filename templates
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/helper_scripts/ANALYSIS.mscw_energy_sub.sh:120-179,182-188`  
The helper derives the IRF table path and DispBDT directory from `printRunParameter` output and
string templates, then runs `mscw_energy` without verifying that the chosen XMLs and tables
match the input `evndisp.root` configuration.  
**Impact:** Stale or mismatched IRF products can be applied to otherwise valid `evndisp` files,
yielding apparently normal `mscw.root` outputs built with the wrong reconstruction tables.

### C15 · MEDIUM · DBTEXT run extraction is fragile against schema drift in `.runinfo` files
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/db_scripts/db_run.sh:69-115,165-185`  
`db_run.sh` re-parses `.runinfo` headers by name; if a header is absent, the helper falls back
to empty or stale column indexes.  
**Impact:** Downstream scripts can derive wrong time windows, source IDs, and DB queries from
superficially valid DBTEXT files.

### C16 · MEDIUM · Existing DBTEXT tarballs suppress regeneration of laser-calibration side files
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/db_scripts/db_run.sh:294-297`  
If `<run>.tar.gz` exists, the script skips `read_laser_calibration` entirely rather than
checking whether `.laserrun`, `.gain`, `.toffset`, and related files are current inside the
tarball.  
**Impact:** Stale or incomplete DBTEXT tarballs freeze outdated calibration metadata and prevent
later `evndisp` analyses from refreshing it.

### C17 · MEDIUM · FITS converter crashes on missing DQM-like DBTEXT files before the `None` check
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/db_scripts/db_write_fits.py:425-433`  
`convert_table_comment_to_ascii(table)` is called for `dqm` files before checking
`if table is not None`.  
**Impact:** One missing DBTEXT component aborts FITS generation and drops all DQM metadata for
the run.

### C18 · MEDIUM · DQM FITS combination requires identical schemas across all runs
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/db_scripts/db_combine_dqm_fits.py:40-54`  
Uses `vstack(..., join_type="exact")`, so any column drift makes the whole merge fail.  
**Impact:** Long-run DQM aggregation is brittle against minor schema evolution.

### C19 · MEDIUM · DQM rate/dead-time extraction does not guard malformed scaler intervals
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/db_scripts/db_write_fits.py:101-138`  
`np.diff()` and scaler divisions are performed with no protection against zero `time_diffs` or
zero `TenMHzScaler` increments.  
**Impact:** Invalid scaler records can inject `inf`/`NaN` values into exported DQM quantities.

### C20 · LOW · `ANALYSIS.v2dl3.sh` has typo `LOGIDR` and undefined variable `AFILE`
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/ANALYSIS.v2dl3.sh:53-58,93-106,117-122`  
**Impact:** Stale temp files and misleading job logs.

### C21 · LOW · `ANALYSIS.anasum_allcuts.sh` uses relative `./ANALYSIS...` calls
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/ANALYSIS.anasum_allcuts.sh:66-76`  
**Impact:** Fails unless launched from the scripts directory.

### C22 · LOW · IRF packaging script has misspelled function name and undefined `$F` in logging
**Location:** `Eventdisplay_AnalysisFiles_VTS/transfer/pack_irfs_for_ucla.sh:161-185`  
**Impact:** Fragile packaging and misleading archive diagnostics.

---

## D — IRF production chain

### D1 · HIGH · `IRF.production.sh` can resubmit the full TMVA grid multiple times
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/IRF.production.sh:337-387`  
The `TRAINTMVA`/`OPTIMIZETMVA` block contains inner loops that re-iterate over the full `EPOCH`
and `ATMOS` sets while already inside outer loops over the same sets.  
**Impact:** TMVA training and optimisation jobs are submitted `len(EPOCH) × len(ATMOS)` times,
wasting batch capacity and mixing results.

### D2 · HIGH · Inner TMVA block uses `${EPOCH:0:2}` instead of `${VX:0:2}` for epoch selection
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/IRF.production.sh:353-366`  
Inside the `for VX in $EPOCH` loop, size-cut file grep and TMVA runparameter selection still use
the outer `${EPOCH:0:2}` instead of the current iteration variable `${VX:0:2}`.  
**Impact:** Multi-epoch submissions select the wrong size-cut row and TMVA runparameter template
for all epochs beyond the first.

### D3 · HIGH · `IRF.mscw_energy_MC_sub.sh` quoted glob fails to copy uncompressed ROOT files
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/helper_scripts/IRF.mscw_energy_MC_sub.sh:133-136`  
`cp -v "${INDIR}/*[0-9].root" "$DDIR"` — the glob is inside double quotes and is not expanded
by the shell.  
**Impact:** If upstream `evndisp` products are uncompressed `.root` files, the MC `mscw_energy`
stage fails before building its file list.

### D4 · HIGH · Job-launch scripts generate one filename and submit another
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/IRF.generate_lookup_table_parts.sh:86-124`; `IRF.mscw_energy_MC.sh:97-139`  
Both scripts write the runnable script to `$FSCRIPT` but submit `$FSCRIPT.sh`, pointing to a
non-existent pathname.  
**Impact:** Core IRF production stages — lookup-table and MC `mscw_energy` generation — fail
before creating their outputs, blocking all downstream steps.

### D5 · HIGH · Gamma/hadron TMVA training ignores the documented UUID argument
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/IRF.trainTMVAforGammaHadronSeparation.sh:18-19,46,60`  
The help text places the optional UUID as argument 8, but the script reads it from `$12`.  
**Impact:** Repeated training campaigns reuse the same submit/log directory, mixing products and
overwriting job state.

### D6 · MEDIUM · Lookup-table combine checks the wrong path before overwriting
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/IRF.combine_lookup_table_parts.sh:77-80`  
Guard tests `[[ -f $OFILE ]]` instead of `[[ -f "$ODIR/$OFILE" ]]`.  
**Impact:** Existing combined lookup tables in the output directory are not reliably detected
before overwrite or merge attempts.

### D7 · MEDIUM · `zstd` availability checks are malformed in several IRF helpers
**Location:** `IRF.compress_evndisp_MC_sub.sh:46-54`; `IRF.evndisp_MC_sub.sh:323-330`; `IRF.lookup_table_parallel_sub.sh:67-79`; `IRF.mscw_energy_MC_sub.sh:137-149`  
`command -v zstd /dev/null` is not the standard existence test.  
**Impact:** Compression/decompression branches can report false missing-tool failures.

### D8 · MEDIUM · `IRF.evndisp_MC_sub.sh` passes literal glob to `zstd` when no files match
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/helper_scripts/IRF.evndisp_MC_sub.sh:88-103`  
The loop runs over `"$V4N"/*.root.zst` without checking whether any files match; the literal
pattern is then passed to `zstd`.  
**Impact:** Spurious decompression failures during duplicate-processing detection.

### D9 · MEDIUM · `IRF.optimizeTMVAforGammaHadronSeparation_sub.sh` uses `MVADIR` before definition
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/helper_scripts/IRF.optimizeTMVAforGammaHadronSeparation_sub.sh:54-69`  
`rm -f ${MVADIR}/rates.log` executes before `MVADIR` is assigned later in the file.  
**Impact:** The cleanup step targets the wrong directory, potentially using an inherited
environment value.

### D10 · MEDIUM · `IRF.production.sh` mixes robust `$(dirname "$0")/...` calls with fragile `./IRF...` calls
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/IRF.production.sh:311,319,330,372,379`  
**Impact:** IRF production fails if launched from outside the scripts directory.

### D11 · MEDIUM · `makeEffectiveArea` IRF resolution duplication is order-dependent in C++
**Location:** `EventDisplay_v4/src/makeEffectiveArea.cpp:162-208`  
(See also B7.) IRF products are duplicated by index and only work if the source IRF was
initialised earlier in the vector.  
**Impact:** Any refactor changing internal IRF ordering silently breaks resolution product
generation.

### D12 · LOW · TMVA gamma/hadron preselection applies an unjustified 10× random retention factor
**Location:** `EventDisplay_v4/src/trainTMVAforGammaHadronSeparation.cpp:143-185`  
After computing the required retained fraction, the code multiplies it by `10.` before capping
at unity.  
**Impact:** Makes the effective preselection behaviour opaque and bin-statistics dependent.

---

## E — XGB / ML integration

### E1 · HIGH · XGB data-analysis step not triggered automatically before XGB `anasum` cuts
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/ANALYSIS.dispXGB.sh:11-95`; `ANALYSIS.anasum_allcuts.sh:6-10,57-71`; `ANALYSIS.anasum_parallel_from_runlist.sh:133-145`  
`ANALYSIS.dispXGB.sh` exists as a standalone script, and the `anasum` wrappers now expose XGB
cut aliases (`moderate2telXGB`, etc.), but the documented workflow goes straight from
`ANALYSIS.mscw_energy.sh` to `anasum`. The wrappers do not trigger `ANALYSIS.dispXGB.sh`
automatically before selecting `*-XGB-BDT` cuts.  
**Impact:** Users can request XGB cuts while analysing plain `mscw` files with no
`.xgb_stereo.root`/`.xgb_gh.root` sidecars, making XGB analyses a fragile manual side-band
instead of a reproducible pipeline stage.

### E2 · HIGH · `IRF.mscw_energy_MC_sub.sh` XGB branch uses stale CLI and wrong run-parameter keys
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/helper_scripts/IRF.mscw_energy_MC_sub.sh:237-271,322-347`; `Eventdisplay-ML/src/eventdisplay_ml/config.py:175-230`; `EventDisplay_v4/src/VInstrumentResponseFunctionRunParameter.cpp:249-263`  
The helper calls `eventdisplay-ml-apply-xgb-stereo --input-file --model-dir --output-file`, but
the current CLI expects `--input_file --model_prefix --output_file`. It also writes
`* XGBFILESUFFIX` into the effective-area parameter file, while `EventDisplay_v4` now parses
only `XGBSTEREOFILESUFFIX` and `XGBGAMMAHADRONFILESUFFIX`.  
**Impact:** The legacy in-subscript XGB branch cannot drive the current Python CLI or pass its
outputs to the current C++ consumer; any workflow relying on this path is broken or silently
stale.

### E3 · HIGH · XGB stereo training ignores the requested reconstruction ID
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/IRF.trainXGBforAngularReconstructionBinned.sh:48-56,76`  
The script accepts `<Rec ID>` as a required argument and carries it through naming and logging,
but hard-codes the training input directory to `.../MSCW_RECID0_DISP`.  
**Impact:** Training jobs submitted for reconstruction IDs other than 0 silently use wrong MC
products.

### E4 · HIGH · Non-XGB effective-area jobs still probe XGB sidecar files
**Location:** `EventDisplay_v4/src/VStereoAnalysis.cpp:1965-1972`; `EventDisplay_v4/src/CData.cpp:28-38,1073-1127`; `logFiles/irfs/20deg_0.5wob_NOISE200.effarea.log:21-23`  
`CData` opens XGB friend files from non-empty suffix strings before checking whether the active
reconstruction methods require them. Confirmed in a production IRF log with
`ENERGYRECONSTRUCTIONMETHOD 0` and a TMVA preselection cut file, yet the job still tries to open
`.xgb_stereo.root`/`.xgb_gh.root`.  
**Impact:** Non-XGB jobs emit avoidable file-open errors and retain hidden dependencies on
artifacts that are not required.

### E5 · HIGH · Eventdisplay-ML classification training and application can use different feature sets
**Location:** `Eventdisplay-ML/src/eventdisplay_ml/data_processing.py:807-817`; `Eventdisplay-ML/src/eventdisplay_ml/models.py:441-452`  
Classification training switches to `features_tmva_style(...)` when `tmva_style=True`, but the
application path always uses `features.features(analysis_type, training=False)` and does not
mirror that branch.  
**Impact:** A TMVA-style-trained classification model can be evaluated with the wrong input
feature set.

### E6 · HIGH · Eventdisplay-ML `config.py` calls `.get()` on a list
**Location:** `Eventdisplay-ML/src/eventdisplay_ml/config.py:162-165`  
`configure_training()` uses `model_parameters.get("energy_bins_log10_tev", []).get("E_min")`;
`energy_bins_log10_tev` is list-structured and a list does not have `.get()`.  
**Impact:** Classification training/apply configuration can fail with `AttributeError` on
valid-looking parameter files in certain call paths.

### E7 · MEDIUM · XGB training helper scratch directory uses undefined variable `MSCW_FILE`
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/helper_scripts/IRF.trainXGBforAngularReconstruction_sub.sh:15-18`  
`TEMPDIR` is derived from `$(basename $MSCW_FILE .root)`, but `MSCW_FILE` is never defined; the
actual input is the list file `LLIST`.  
**Impact:** Scratch-directory naming is accidental and environment-dependent, causing collisions
in batch production and making failed training jobs harder to debug.

### E8 · MEDIUM · `EVNDISP_MLSYS` documented but helpers require a hard-coded conda environment name
**Location:** `Eventdisplay_AnalysisScripts_VTS/scripts/helper_scripts/ANALYSIS.dispXGB_sub.sh:9-10,27-53`; `IRF.dispXGB_sub.sh:8-9`; `IRF.trainXGBforAngularReconstruction_sub.sh:9-10`; `IRF.trainXGBforGammaHadronSeparation_sub.sh:13-14`; `README.md:16-18,22-29`  
The public setup documents `EVNDISP_MLSYS`, but the helpers ignore it and require a pre-existing
conda environment named `eventdisplay_ml`.  
**Impact:** A user can satisfy all documented environment variables and still fail at runtime.

### E9 · MEDIUM · XGBoost hyperparameters are hard-coded in Python, not versioned in auxiliary parameter files
**Location:** `Eventdisplay-ML/src/eventdisplay_ml/hyper_parameters.py:9-45`  
`n_estimators`, `learning_rate`, `max_depth`, `subsample`, `colsample_bytree`, etc. are embedded
in code; the shipped JSON files define only binning and `tmva_style`.  
**Impact:** The effective ML model is not fully described by the analysis parameter files,
weakening reproducibility.

### E10 · MEDIUM · Production XGB training helpers omit `--random_state` and `--hyperparameter_config`
**Location:** `IRF.trainXGBforGammaHadronSeparation_sub.sh:59-67`; `IRF.trainXGBforAngularReconstruction_sub.sh:55-61`  
The helpers pass `train_test_fraction` and event limits but not a fixed random seed or explicit
hyperparameter file.  
**Impact:** Routine retraining is not strictly reproducible even when input event lists are
unchanged.

### E11 · MEDIUM · XGB command names are inconsistent between production scripts and documentation
**Location:** `IRF.generalproduction.sh:18-29`; `IRF.production.sh:24-31,392-403`; `docs/changes/124.feature.md:1-10`; `README.md:83-159`  
`IRF.generalproduction.sh` advertises `TRAINXGBANGRESBINNED`; `IRF.production.sh` implements
`TRAINXGBANGRES`; the change note references a non-existent
`IRF.trainXGBforAngularReconstruction.sh`; the main README omits the XGB production sequence
entirely.  
**Impact:** Operators are left with conflicting command names and incomplete instructions.

### E12 · MEDIUM · Legacy `TMVA.BDT.V4.runparameter` ships a malformed `PREPARE_TRAINING_OPTIONS` line
**Location:** `Eventdisplay_AnalysisFiles_VTS/ParameterFiles/TMVA.BDT.V4.runparameter:68-76`; `EventDisplay_v4/src/VTMVARunData.cpp:423-435`  
The line lacks the leading `*` required for parsing and contains a double `::` in the option
string.  
**Impact:** Legacy V4 BDT configuration does not reliably describe the training setup it
appears to encode.

### E13 · LOW · TMVA training options have drifted from the accepted backend behaviour
**Location:** `logFiles/irfs/mvaAngRes_20deg-BDTDisp-Tel2.log:81-82`  
`NegWeightTreatment=InverseBoostNegWeights` is not supported for `BoostType=Grad` and is silently
replaced by `Pray`.  
**Impact:** Effective training configuration no longer matches the nominal requested one.

---

## F — Auxiliary files & configuration

### F1 · HIGH · Epoch metadata and downloadable IRF manifests are out of sync
**Location:** `Eventdisplay_AnalysisFiles_VTS/ParameterFiles/VERITAS.Epochs.runparameter:47-53`; `Eventdisplay_AnalysisFiles_VTS/transfer/file_list_V6.dat:304-358`  
`VERITAS.Epochs.runparameter` defines `V6_2024_2024s`, `V6_2024_2025w`, `V6_2025_2025s`, and
`V6_2025_2026w`, but the shipped V6 download manifest stops at `V6_2023_2024w`.  
**Impact:** A fresh checkout following the README cannot fetch the IRF tarballs needed for the
newer epochs that the active scripts may select at runtime.

### F2 · HIGH · Future epoch `V6_2025_2025s` is defined twice in the epoch file
**Location:** `Eventdisplay_AnalysisFiles_VTS/ParameterFiles/VERITAS.Epochs.runparameter:48-53`  
The second entry covers `2026-05-01 to 2026-11-24` while a TODO states it should become
`V6_2026_2026s`.  
**Impact:** Future epoch lookup becomes ambiguous and can map runs to the wrong gain/atmosphere
configuration.

### F3 · HIGH · Documented download path does not reproduce the full current auxiliary-file set
**Location:** `Eventdisplay_AnalysisFiles_VTS/README.md:10-19`; `transfer/file_list_V6.dat:1-45,170-186`; `auxiliary_files.txt:19632-19645,20236-20262`  
The download manifests expose only classic DispBDTs and effective areas; the fully unpacked
inventory in `auxiliary_files.txt` additionally contains a `DispXGB` tree and XGB
effective-area products.  
**Impact:** Two users following the official instructions end up with materially different
auxiliary installations.

### F4 · HIGH · Unpacked auxiliary inventory is effectively AP-only despite scripts advertising multi-type support
**Location:** `Eventdisplay_AnalysisScripts_VTS/README.md:58-81`; `IRF.production.sh:91-120`; `auxiliary_files.txt:19629-19631,21112-21169`  
Scripts and documentation present `AP`, `NN`, `TS`, and `CC` as supported analysis types; the
V6 file list even advertises `DispBDTs_NN_*` and `EffectiveAreas_nominalHV_NN_*` tarballs. The
unpacked inventory shows only an empty `DispBDTs/NN` directory; all table and effective-area
roots are `-AP.root` products.  
**Impact:** The supposedly complete auxiliary area cannot support NN (or TS/CC) analyses.

### F5 · HIGH · XGB instrument-file coverage is patchy while wrappers expose XGB as a generic AP mode
**Location:** `auxiliary_files.txt:19632-19645,19697,19734,19933,20010,20177,20236-20262`  
Most XGB model directories are `ATM61` only; XGB effective areas are confined to
`V6_2016_2017` for `ATM61/62`, far narrower than the generic XGB hooks now exposed by the AP
wrappers.  
**Impact:** The auxiliary-file area does not yet support XGB as a broadly deployable production
mode across the same epoch space as the classic AP products.

### F6 · MEDIUM · XGB effective-area naming is internally inconsistent
**Location:** `auxiliary_files.txt:20245-20248,20698`; `ANALYSIS.anasum_parallel_from_runlist.sh:133-145`  
The same tree contains both `...ModerateXGB...root` and `...Moderate-XGB-BDT...root` names;
the analysis wrapper resolves `moderate2telXGB` to `NTel2-PointSource-Moderate-XGB-BDT`.  
**Impact:** Shell wrappers, cut files, and effective-area selection logic can disagree on which
product family is canonical.

### F7 · MEDIUM · Maintainer README is stale and contradicts the transfer scripts
**Location:** `Eventdisplay_AnalysisFiles_VTS/README.md:29-30,77-89`; `transfer/pack_irfs_for_ucla.sh:11`; `transfer/put_irfs_to_ucla.sh:27`  
The README documents a non-existent `EVNDISP.reconstruction.runparameter`, refers generically
to `tar_packages`, and instructs maintainers to move packages to `archive` before calling an
upload script that only scans `tar_packages_${VERSION}/`.  
**Impact:** Release maintenance instructions are no longer self-consistent.

### F8 · MEDIUM · `download_irfs.sh` continues into extract/remove after a failed transfer
**Location:** `Eventdisplay_AnalysisFiles_VTS/download_irfs.sh:36-49`  
`tar --keep-newer-files` and `rm -v` execute immediately after the copy/`bbftp` call without
checking the transfer exit status.  
**Impact:** Failed or partial downloads trigger extraction and cleanup, making recovery harder.

### F9 · MEDIUM · `copy_dispBDTs.sh` and `copy_dispXGB.sh` silently overwrite `SIMTYPE` in a loop
**Location:** `Eventdisplay_AnalysisFiles_VTS/DispBDTs/copy_dispBDTs.sh:11-19`; `DispXGB/copy_dispXGB.sh:11-18`  
Both scripts assign `SIMTYPE` repeatedly, so only the last hard-coded value survives.
`copy_dispBDTs.sh` also prints `${IRVERSION}` instead of `${IRFVERSION}`.  
**Impact:** IRF-maintenance copies silently pull the wrong model family with misleading
provenance information.

### F10 · LOW · V4/V5 gamma/hadron copy script has dead atmosphere-selection code
**Location:** `Eventdisplay_AnalysisFiles_VTS/GammaHadronBDTs/copy_GammaHadron_V4_V5_BDTs.sh:21-31`  
Loop iterates over `ATM21 ATM22` but the branch condition checks for `ATM62`; the
summer/winter split is never exercised.  
**Impact:** Any future atmosphere-specific V4/V5 handling based on this template will be wrong.

### F11 · LOW · Global DB fallback is present in text only, not in parsed configuration
**Location:** `Eventdisplay_AnalysisFiles_VTS/ParameterFiles/EVNDISP.global.runparameter:9-16`  
Only the `romulus.ucsc.edu` line is active (`*`-prefixed); the `remus.ucsc.edu` fallback is
unstarred and ignored.  
**Impact:** No real failover at the parameter-file level if the primary DB endpoint is
unavailable.

### F12 · LOW · Backup content mixed into the main instrument-file tree
**Location:** `auxiliary_files.txt:19298`  
A `DispBDTs/AP.back` subtree sits alongside nominal production content.  
**Impact:** Ambiguity about which products are authoritative; risk of copying the wrong tree.

---

## G — Release tests

### G1 · HIGH · `EventDisplay_Release_v492` is still a skeleton, not an auditable release-test record

-  [x] Expected. v492 is in early development.

**Location:** `EventDisplay_Release_v492/README.md:1-3`  
The checkout contains only metadata plus `runparameter/V6.AP.runparameter.dat`; none of the
expected result trees (`Crab/`, `mc_data_comparison/`, `irf_plotting/`, `SourceTests/`,
`AP/radialAcceptances/`) are present.  
**Impact:** The `v492` artifact does not provide the evidence needed to review the release tests
the codebase expects to exist.

### G2 · HIGH · Release-test output naming is inconsistent

- [ ] Fixed in https://github.com/VERITAS-Observatory/EventDisplay_ReleaseTests_code/pull/34
 
**Location:** `EventDisplay_ReleaseTests_code/README.md:10-15`; `release_tests/montecarlo/energythresholds/plot.C:150-152`  
Most scripts use `EventDisplay_Release_<version>`, but the README points to
`EventDisplay_ReleaseTests_v492` and the energy-threshold macro writes to
`EventDisplay_ReleaseTests_<version>/energythresholds/`.  
**Impact:** Release products are scattered between differently named trees, making results look
missing and breaking documentation links.

### G3 · HIGH · MC/data comparison applies the summer-atmosphere filter with an undefined variable

- [ ] Fixed in https://github.com/VERITAS-Observatory/EventDisplay_ReleaseTests_code/pull/34

**Location:** `EventDisplay_ReleaseTests_code/release_tests/montecarlo/mc_data_comparison/compareDatawithMC.sh:138-146`  
The winter branch checks `${atm}`, but the summer branch checks `${A}` before `A` is assigned.  
**Impact:** Summer-epoch filtering silently uses the wrong atmosphere combinations, weakening one
of the main release-validation checks.

### G4 · HIGH · Crab run-list generation: `WOBBB` typo leaves zero-wobble classification undefined

- [ ] Fixed in https://github.com/VERITAS-Observatory/EventDisplay_ReleaseTests_code/pull/34

**Location:** `EventDisplay_ReleaseTests_code/release_tests/sources/Crab/runlist_generator_from_anasum_log.sh:127-132,156-165`  
When both wobble offsets are zero, the script assigns `WOBBB="0"` instead of `WOBB="0"`, but
later decisions key on `WOBB`.  
**Impact:** Zero-wobble runs inherit a stale wobble state and are grouped into the wrong run
lists, contaminating downstream source-test summaries.

### G5 · HIGH · Source-test plotting macros are pinned to the `v491` library and data layout

- [ ] Fixed in https://github.com/VERITAS-Observatory/EventDisplay_ReleaseTests_code/pull/34

**Location:** `release_tests/sources/Crab/plot_lightcurves.C:19`; `plot_energy_spectra.C:20`; `plot_skymaps.C:17`; `plot_sensitivity.C:11,27-40`; `plot_sensitivity_compare_cuts.C:16,44-62`  
These macros hard-code `/.../EventDisplay_v491-al9/lib/libVAnaSum.so` and for sensitivity plots
hard-code `v491`/`v490` result directories.  
**Impact:** Running on `v492` products loads the wrong code and compares against fixed historical
paths.

### G6 · MEDIUM · IRF plotting is hard-coded to a single slice instead of broad parameter coverage

**Location:** `EventDisplay_ReleaseTests_code/release_tests/montecarlo/irf_plotting/irf_plotting.sh:16-25,57,75-89`  
The driver hard-codes `COMPAREVERSION="v492"`, a fixed sim/cut family, and only runs the body
for `V6_2016_2017` / `ATM61`.  
**Impact:** The nominal IRF plotting step no longer spans the stated parameter space and is weak
as a release-regression check.

### G7 · MEDIUM · Shipped `v492` run-parameter format encodes a fragile cut-name duality

- [ ] Fixed in https://github.com/VERITAS-Observatory/EventDisplay_ReleaseTests_code/pull/34

**Location:** `EventDisplay_Release_v492/runparameter/V6.AP.runparameter.dat:35-39`; `release_tests/sources/Crab/anasum_from_runlists.sh:34-40`  
Each `* CUT` line stores both a full IRF cut identifier and a short alias; consumers rely on the
fourth field specifically.  
**Impact:** Easy for future scripts to read the wrong cut token and silently select the wrong
inputs.

---

## H — Performance & efficiency

### H1 · HIGH · `evndisp` keeps time-sliced pedestal analysis enabled during normal data reconstruction
**Location:** `EventDisplay_v4/src/VEventLoop.cpp:490-495,1442-1461`; `EventDisplay_v4/src/VPedestalCalculator.cpp:30-199,280-381`; `logFiles/analysis/64080.log:71,598-599`  
On the exercised run, `evndisp` processes ~205 events/s (`543 481` events in 44:04). The active
path still performs pedestal-variance bookkeeping for every telescope during standard
reconstruction: `VPedestalCalculator::doAnalysis()` loops over hit channels, recomputes trace
sums for every summation window, and fills per-pixel histograms.  
**Impact:** A calibration-style side analysis remains in the hottest reconstruction loop and is a
plausible first-order contributor to the 16× throughput gap between `evndisp` (~205 ev/s) and
`mscw_energy` (~3 229 ev/s).

### H2 · MEDIUM · `evndisp` repeats invariant low-gain/sum-window validation in the per-event loop
**Location:** `EventDisplay_v4/src/VEventLoop.cpp:1204-1292`  
The analysis path rechecks availability of low-gain multipliers for `sumwindow_1`,
`sumwindow_2`, and `sumwindow_pass1` on every event for every telescope, even though these are
run-level constants after initialisation.  
**Impact:** Avoidable branch-heavy work repeated millions of times in long runs.

### H3 · HIGH · `anasum` double-scans `mscw.root` and reinitialises TMVA twice per run
**Location:** `EventDisplay_v4/src/VAnaSum.cpp:464-465`; `EventDisplay_v4/src/VStereoAnalysis.cpp:264-349,434-448`; `EventDisplay_v4/src/VTMVAEvaluator.cpp:366-511`; `logFiles/analysis/64080.anasum.log:97-103,237-243`  
In the reflected-background path, ON and OFF both point to the same `mscw.root`, yet `VAnaSum`
runs separate passes. Each pass reads `174 937` entries and rebuilds the gamma/hadron TMVA
readers from disk.  
**Impact:** For every run, `anasum` roughly doubles tree I/O and repeats expensive TMVA and
effective-area initialisation that could be shared.

### H4 · MEDIUM · `anasum` uses a linear TMVA-bin search with recomputed mean energies per event
**Location:** `EventDisplay_v4/src/VTMVAEvaluator.cpp:714-745,802-839`  
Each gamma/hadron evaluation calls `getDataBin()`, which scans all loaded bins and recomputes
the mean energy of each candidate before evaluating the selected reader.  
**Impact:** Avoidable per-event classifier overhead in the inner `anasum` loop, especially with
many energy/zenith bins.

### H5 · MEDIUM · Multi-run `anasum` jobs scale as O(runs × total entries in chain)
**Location:** `EventDisplay_v4/src/VAnaSum.cpp:464-465`; `EventDisplay_v4/src/VStereoAnalysis.cpp:434-450`  
`fillHistograms()` loops over all chain entries and then filters by `runNumber == irun`; each
run-specific pass re-reads entries from all other runs.  
**Impact:** Runtime can grow toward O(runs × total_entries) instead of near-linear.

### H6 · MEDIUM · `evndisp` reads pedestal calibration inputs redundantly during initialisation
**Location:** `logFiles/analysis/64080.log:247-252,263-268,275-280,291-296`  
The same telescope-level pedestal products are read twice in succession during calibration setup.  
**Impact:** Duplicated calibration I/O in the slowest stage of the data pipeline.

### H7 · MEDIUM · `mscw_energy` repeatedly rebuilds TMVA datasets instead of amortising setup
**Location:** `logFiles/analysis/64080.mscw.log:156-169`; `logFiles/irfs/20deg_0.5wob_NOISE200.mscw.log:261-277`  
Both data-side and MC-side logs contain repeated `Rebuilding Dataset Default` messages.  
**Impact:** Avoidable setup churn, expensive when multiplied across large IRF productions.

---

## Cross-cutting issues (confirmed from logs)

### X1 · HIGH · `v2dl3` silently sets quality flag to 0 when Eventdisplay logfile is missing
**Location:** `logFiles/analysis/64080.v2dl3.log:14`  
`v2dl3` reports "Eventdisplay logfile not found in anasum root file. Quality flag set to 0."  
**Impact:** Downstream DL3 products lose quality metadata without a loud failure.

### X2 · MEDIUM · `anasum` continues under missing TMVA-weight warnings
**Location:** `logFiles/analysis/64080.anasum.log:243-249,541-547`; `EventDisplay_v4/src/VTMVAEvaluator.cpp:190-279`  
`VTMVAEvaluator` reports "TMVA root file not found or incomplete file (0)" and continues;
see also B4.  
**Impact:** Incomplete classifier inputs survive initialisation silently.

### X3 · MEDIUM · `anasum` reinitialises the same TMVA state twice, doubling warning noise
**Location:** `logFiles/analysis/64080.anasum.log:237-247,535-545`  
See also H3. The same missing/incomplete low-energy TMVA file warning is emitted once for ON
and once again for OFF.  
**Impact:** Duplicated warnings make it harder to distinguish a new problem from a repeated one.

### X4 · LOW · Data run contains large numbers of GPS-faulted events; pipeline continues without escalation
**Location:** `logFiles/analysis/64080.log:565-580,587-596`  
Missing telescope events and GPS faults on telescopes 1 and 4 are logged and skipped.  
**Impact:** Run-quality issues are not escalated; pipeline favours "continue with warnings" over
explicit run-quality gating.

---

## Hyperparameter locations (reference)

| Model | Location |
|---|---|
| TMVA gamma/hadron BDT | `Eventdisplay_AnalysisFiles_VTS/ParameterFiles/TMVA.BDT.runparameter:13-15` |
| TMVA DispBDT | `Eventdisplay_AnalysisFiles_VTS/ParameterFiles/TMVA.BDTDisp.runparameter:1-2` |
| XGB regression/classification | `Eventdisplay-ML/src/eventdisplay_ml/hyper_parameters.py:9-45` (hard-coded in Python) |
| XGB binning / style | `Eventdisplay_AnalysisFiles_VTS/ParameterFiles/XGB-stereo-parameter.json`; `XGB-classify-parameter.json` |

---

## Priority action list

### Immediate (HIGH severity, silent wrong-physics or silent job failure)

1. **Fix `ANALYSIS.anasum.sh` SUBC initialisation** (`anasum.sh:90-92`) — jobs are silently never submitted (C1)
2. **Fix `get_disp_dir()` to check `HVSETTINGS == "obsLowHV"` instead of undefined `SIMTYPE`** (`mscw_energy_sub.sh:136-152`) — wrong DispBDT directory for all RedHV/UV runs at high zenith (C2)
3. **Fix shower-core `yimp` validity check** (`VArrayAnalyzer.cpp:944-951`) — NaN core-Y propagates into reconstruction (A6)
4. **Fix method-4 pair-angle rejection variable** (`VArrayAnalyzer.cpp:1241-1248`) — wrong telescope pairs used in cut (A7)
5. **Fix low-gain T0 assignment and calculation** (`VCalibrationData.cpp:365-373`; `VCalibrator.cpp:899-903`) — high-gain state corrupted (A4, A5)
6. **Fix core-resolution formula** (`VInstrumentResponseFunctionData.cpp:472-485`) — Xcore used twice, Ycore ignored (A12)
7. **Fix TMVA zenith-bin off-by-one** (`trainTMVAforGammaHadronSeparation.cpp:255-266`) — out-of-bounds vector access (A13)
8. **Fix TMVA training factory type for reconstruction-quality path** (`trainTMVAforGammaHadronSeparation.cpp:343-360`) — classification factory used instead of regression (A14)
9. **Fix IRF job-launch filenames** (`IRF.generate_lookup_table_parts.sh:86-124`; `IRF.mscw_energy_MC.sh:97-139`) — submitted path does not exist (D4)
10. **Fix `IRF.production.sh` double-loop and epoch variable** (`IRF.production.sh:337-387,353-366`) — TMVA grid submitted multiple times with wrong epoch variable (D1, D2)
11. **Fix `IRF.mscw_energy_MC_sub.sh` quoted glob and stale XGB CLI** (`IRF.mscw_energy_MC_sub.sh:133-136,237-271`) — files not copied; Python CLI mismatch (D3, E2)
12. **Propagate `FORCEDATMO` via sed template** (`ANALYSIS.mscw_energy.sh:63`) — ATM override is a no-op in batch (C3)
13. **Add atmosphere-ID normalisation for V4/V5 in `mscw_energy_sub.sh`** consistent with `prepare_atmo_string()` — table lookup fails for V4/V5 GRISU data (C4)
14. **Fix `ANALYSIS.evndisp_sub.sh` DBTEXT file existence checks** — missing files silently inject wrong metadata (C9)
15. **Fix broken TMVA bin-parsing regex** (`IRF.trainTMVAforGammaHadronSeparation.sh:107-109`) — bin parsing fails (C10)
16. **Fix sliding-window FADC out-of-bounds read** (`VTraceHandler.cpp:783`) — `FADC[n]` accessed on every event where the window reaches the rightmost search position; UB corrupts charge reconstruction (A17)
17. **Fix `VOnOff::doOnOffforParameterHistograms` background mutation** (`VOnOff.cpp:159-161`) — `hoff->Scale` after `Add` double-scales background, corrupting reflected-region excess counts and significance in `anasum` (B11)
18. **Fix `VInstrumentResponseFunctionRunParameter` silent key drop** (`VInstrumentResponseFunctionRunParameter.cpp:109-358`) — add a terminal `else { cerr << "unknown key: " << temp << endl; }` to catch parameter typos (B13)
19. **Fix `printRunParameter` exit code on file-open failure** (`printRunParameter.cpp:330-335`) — replace `exit(0)` with `exit(EXIT_FAILURE)` so calling shell scripts detect errors (B14)

### Short-term (MEDIUM / HIGH, reproducibility and efficiency)

20. **Gate time-sliced pedestal analysis** in the standard `evndisp` analysis loop — primary throughput bottleneck (H1)
21. **Avoid double `mscw.root` scan and double TMVA initialisation** in `anasum` — doubles tree I/O (H3)
22. **Fix Eventdisplay-ML train/apply feature mismatch** for `tmva_style=True` classification (E5)
23. **Fix `config.py` `.get()` on list** — `AttributeError` on valid parameter files (E6)
24. **Wire `ANALYSIS.dispXGB.sh` into the default `anasum` workflow**, or add a guard that prevents `*-XGB-BDT` cut selection without XGB sidecars (E1)
25. **Fix XGB stereo training RecID hard-coding** (`IRF.trainXGBforAngularReconstructionBinned.sh:48-56`) (E3)
26. **Gate XGB friend-file probing** in `CData` to the active reconstruction methods (E4)
27. **Synchronise epoch file, download manifests, and downloadable IRF tarballs** (F1, F2, F3)
28. **Add end-to-end IRF → analysis consistency test** — shell and C++ interface drift surface only in production
29. **Fix `makeEffectiveArea` "no problem" comment vs. `exit(EXIT_FAILURE)` mismatch** (`makeEffectiveArea.cpp:118`; `VInstrumentResponseFunctionRunParameter.cpp:437`) — either remove the misleading comment or implement graceful fallback (B12)
30. **Fix `VTableLookupRunParameter` error message** (`VTableLookupRunParameter.cpp:175`) — change "2=read" to "0=read" (B15)
31. **Fix `VTraceHandler::getQuickMax` tmax initialisation** (`VTraceHandler.cpp:420-426`) — initialise `tmax` to 0 or the expected maximum before computing `nMax` (A18)


Stage	IRF chain entry	Data-analysis entry
Image reconstruction	`IRF.evndisp_MC.sh` → `evndisp`	`ANALYSIS.evndisp.sh` → `evndisp`
Stereo / table lookup	`IRF.mscw_energy_MC.sh` → `mscw_energy`	`ANALYSIS.mscw_energy.sh` → `mscw_energy`
Angular reco (ML)	DispBDT / XGB training (`IRF.train*`)	DispBDT XMLs consumed by `mscw_energy`
Gamma/hadron (ML)	BDT / XGB training, optimisation, effective areas	TMVA/XGB friend trees consumed by `anasum`
Science analysis	—	`ANALYSIS.anasum*.sh` → `anasum`
DL3 export	—	`ANALYSIS.v2dl3.sh` → `v2dl3`

Model	Location
TMVA gamma/hadron BDT	`Eventdisplay_AnalysisFiles_VTS/ParameterFiles/TMVA.BDT.runparameter:13-15`
TMVA DispBDT	`Eventdisplay_AnalysisFiles_VTS/ParameterFiles/TMVA.BDTDisp.runparameter:1-2`
XGB regression/classification	`Eventdisplay-ML/src/eventdisplay_ml/hyper_parameters.py:9-45` (hard-coded in Python)
XGB binning / style	`Eventdisplay_AnalysisFiles_VTS/ParameterFiles/XGB-stereo-parameter.json`; `XGB-classify-parameter.json`

Issue Report #352

Description

Issue Report

Pipeline overview (reference)

A — C++ algorithm defects

A1 · HIGH · Out-of-bounds read in pulse-timing extraction

A2 · HIGH · Sliding-window charge branch returns FADC[1] instead of integrated charge

A3 · MEDIUM · Fitted-trace T0 mixes pedestal-subtracted and absolute amplitude conventions

A4 · HIGH · Low-gain average-T0 assignment always writes the high-gain field

A5 · HIGH · Low-gain average-T0 calculation uses the high-gain sum path

A6 · HIGH · Shower-core validity check tests ximp twice, never yimp

A7 · HIGH · Method-4 stereo pair-angle rejection applies cut to the wrong telescope pair

A8 · HIGH · Log-likelihood image fit can be initialised with invalid centroid bounds

A9 · MEDIUM · Muon-ring image-size correction divides by zero at the first grid point

A10 · MEDIUM · Emission-height pair weights undefined for small image sizes

A11 · MEDIUM · VEmissionHeightCalculator accumulates telescope positions across reinitalisations

A12 · HIGH · Core-resolution IRF formula uses Xcore twice instead of Xcore and Ycore

A13 · HIGH · Gamma/hadron TMVA training has an off-by-one zenith-bin bounds check

A14 · HIGH · Reconstruction-quality TMVA training configured as a classification factory

A15 · MEDIUM · Angular-reconstruction training silently rescales the requested train/test split

A16 · MEDIUM · Reduced-tree subsampling normalised by file count, not event population

A17 · HIGH · Sliding-window charge extraction reads one past the end of the FADC array

A18 · MEDIUM · VTraceHandler::getQuickMax uses the previous call's trace maximum for saturation counting

B — C++ infrastructure defects

B1 · MEDIUM · CData::getXGBTree() assumes .root is present in the filename

B2 · MEDIUM · XGB gamma/hadron C++ interface hard-wired to Is_Gamma_70

B3 · MEDIUM · VTMVAEvaluator recovers feature names by brittle XML string slicing

B4 · MEDIUM · TMVA weight grid initialisation tolerates missing interior bins with warnings

B5 · LOW · VTMVARunData sanity message does not match the actual condition

B6 · MEDIUM · VInstrumentResponseFunction mutates shared MC azimuth in place

B7 · MEDIUM · makeEffectiveArea IRF duplication is order-dependent

B8 · MEDIUM · copyMCHistograms() keeps the first ROOT file open

B9 · MEDIUM · anasum contains a stale unreachable analysisType == 5 branch

B10 · MEDIUM · Epoch reevaluation at runtime means analysis behaviour changes with epoch-file updates

B11 · HIGH · VOnOff mutates the background histogram after using it, corrupting reflected-region normalisation

B12 · HIGH · makeEffectiveArea comments that the MC run header is optional but hard-exits when it is missing

B13 · HIGH · VInstrumentResponseFunctionRunParameter silently ignores unknown run-parameter keys

B14 · MEDIUM · printRunParameter exits with code 0 on file-open failure

B15 · LOW · VTableLookupRunParameter error message says "2=read" but the code only accepts "0=read"

C — Shell orchestration bugs

C1 · HIGH · ANALYSIS.anasum.sh never submits any jobs

C2 · HIGH · get_disp_dir() checks an undefined variable, always selects the wrong DispBDT directory

C3 · MEDIUM · FORCEDATMO atmosphere override is silently ignored in cluster submissions

C4 · MEDIUM · mscw_energy_sub.sh missing atmosphere-ID normalisation for V4/V5 data

C5 · MEDIUM · DISPBDT guard in mscw_energy_sub.sh is missing $

C6 · MEDIUM · printRunParameter -elevation field is parsed positionally in the helper

C7 · MEDIUM · mscw_energy hard-fails when a TMVA XML family is missing, no soft fallback

C8 · MEDIUM · anasum default input directory hard-wired to RecID0 with no consistency check

C9 · HIGH · ANALYSIS.evndisp_sub.sh consumes DBTEXT-derived files without existence checks

C10 · HIGH · Broken TMVA bin-parsing regex in training and optimisation scripts

C11 · MEDIUM · Cut optimisation continues with missing effective-area input

C12 · MEDIUM · DL3 jobs install Python dependencies at runtime per job

C13 · MEDIUM · ANALYSIS.anasum_combine.sh lacks a completion/integrity guard before merge

C14 · MEDIUM · Data-side mscw_energy selects IRF products purely from filename templates

C15 · MEDIUM · DBTEXT run extraction is fragile against schema drift in .runinfo files

C16 · MEDIUM · Existing DBTEXT tarballs suppress regeneration of laser-calibration side files

C17 · MEDIUM · FITS converter crashes on missing DQM-like DBTEXT files before the None check

C18 · MEDIUM · DQM FITS combination requires identical schemas across all runs

C19 · MEDIUM · DQM rate/dead-time extraction does not guard malformed scaler intervals

C20 · LOW · ANALYSIS.v2dl3.sh has typo LOGIDR and undefined variable AFILE

C21 · LOW · ANALYSIS.anasum_allcuts.sh uses relative ./ANALYSIS... calls

C22 · LOW · IRF packaging script has misspelled function name and undefined $F in logging

D — IRF production chain

D1 · HIGH · IRF.production.sh can resubmit the full TMVA grid multiple times

D2 · HIGH · Inner TMVA block uses ${EPOCH:0:2} instead of ${VX:0:2} for epoch selection

D3 · HIGH · IRF.mscw_energy_MC_sub.sh quoted glob fails to copy uncompressed ROOT files

D4 · HIGH · Job-launch scripts generate one filename and submit another

D5 · HIGH · Gamma/hadron TMVA training ignores the documented UUID argument

D6 · MEDIUM · Lookup-table combine checks the wrong path before overwriting

D7 · MEDIUM · zstd availability checks are malformed in several IRF helpers

D8 · MEDIUM · IRF.evndisp_MC_sub.sh passes literal glob to zstd when no files match

D9 · MEDIUM · IRF.optimizeTMVAforGammaHadronSeparation_sub.sh uses MVADIR before definition

D10 · MEDIUM · IRF.production.sh mixes robust $(dirname "$0")/... calls with fragile ./IRF... calls

D11 · MEDIUM · makeEffectiveArea IRF resolution duplication is order-dependent in C++

D12 · LOW · TMVA gamma/hadron preselection applies an unjustified 10× random retention factor

E — XGB / ML integration

E1 · HIGH · XGB data-analysis step not triggered automatically before XGB anasum cuts

E2 · HIGH · IRF.mscw_energy_MC_sub.sh XGB branch uses stale CLI and wrong run-parameter keys

E3 · HIGH · XGB stereo training ignores the requested reconstruction ID

E4 · HIGH · Non-XGB effective-area jobs still probe XGB sidecar files

A2 · HIGH · Sliding-window charge branch returns `FADC[1]` instead of integrated charge

A6 · HIGH · Shower-core validity check tests `ximp` twice, never `yimp`

A11 · MEDIUM · `VEmissionHeightCalculator` accumulates telescope positions across reinitalisations

A12 · HIGH · Core-resolution IRF formula uses `Xcore` twice instead of `Xcore` and `Ycore`

A18 · MEDIUM · `VTraceHandler::getQuickMax` uses the previous call's trace maximum for saturation counting

B1 · MEDIUM · `CData::getXGBTree()` assumes `.root` is present in the filename

B2 · MEDIUM · XGB gamma/hadron C++ interface hard-wired to `Is_Gamma_70`

B3 · MEDIUM · `VTMVAEvaluator` recovers feature names by brittle XML string slicing

B5 · LOW · `VTMVARunData` sanity message does not match the actual condition

B6 · MEDIUM · `VInstrumentResponseFunction` mutates shared MC azimuth in place

B7 · MEDIUM · `makeEffectiveArea` IRF duplication is order-dependent

B8 · MEDIUM · `copyMCHistograms()` keeps the first ROOT file open

B9 · MEDIUM · `anasum` contains a stale unreachable `analysisType == 5` branch

B11 · HIGH · `VOnOff` mutates the background histogram after using it, corrupting reflected-region normalisation

B12 · HIGH · `makeEffectiveArea` comments that the MC run header is optional but hard-exits when it is missing

B13 · HIGH · `VInstrumentResponseFunctionRunParameter` silently ignores unknown run-parameter keys

B14 · MEDIUM · `printRunParameter` exits with code 0 on file-open failure

B15 · LOW · `VTableLookupRunParameter` error message says "2=read" but the code only accepts "0=read"

C1 · HIGH · `ANALYSIS.anasum.sh` never submits any jobs

C2 · HIGH · `get_disp_dir()` checks an undefined variable, always selects the wrong DispBDT directory

C3 · MEDIUM · `FORCEDATMO` atmosphere override is silently ignored in cluster submissions

C4 · MEDIUM · `mscw_energy_sub.sh` missing atmosphere-ID normalisation for V4/V5 data

C5 · MEDIUM · `DISPBDT` guard in `mscw_energy_sub.sh` is missing `$`

C6 · MEDIUM · `printRunParameter -elevation` field is parsed positionally in the helper

C7 · MEDIUM · `mscw_energy` hard-fails when a TMVA XML family is missing, no soft fallback

C8 · MEDIUM · `anasum` default input directory hard-wired to `RecID0` with no consistency check

C9 · HIGH · `ANALYSIS.evndisp_sub.sh` consumes DBTEXT-derived files without existence checks

C13 · MEDIUM · `ANALYSIS.anasum_combine.sh` lacks a completion/integrity guard before merge

C14 · MEDIUM · Data-side `mscw_energy` selects IRF products purely from filename templates

C15 · MEDIUM · DBTEXT run extraction is fragile against schema drift in `.runinfo` files

C17 · MEDIUM · FITS converter crashes on missing DQM-like DBTEXT files before the `None` check

C20 · LOW · `ANALYSIS.v2dl3.sh` has typo `LOGIDR` and undefined variable `AFILE`

C21 · LOW · `ANALYSIS.anasum_allcuts.sh` uses relative `./ANALYSIS...` calls

C22 · LOW · IRF packaging script has misspelled function name and undefined `$F` in logging

D1 · HIGH · `IRF.production.sh` can resubmit the full TMVA grid multiple times

D2 · HIGH · Inner TMVA block uses `${EPOCH:0:2}` instead of `${VX:0:2}` for epoch selection

D3 · HIGH · `IRF.mscw_energy_MC_sub.sh` quoted glob fails to copy uncompressed ROOT files

D7 · MEDIUM · `zstd` availability checks are malformed in several IRF helpers

D8 · MEDIUM · `IRF.evndisp_MC_sub.sh` passes literal glob to `zstd` when no files match

D9 · MEDIUM · `IRF.optimizeTMVAforGammaHadronSeparation_sub.sh` uses `MVADIR` before definition

D10 · MEDIUM · `IRF.production.sh` mixes robust `$(dirname "$0")/...` calls with fragile `./IRF...` calls

D11 · MEDIUM · `makeEffectiveArea` IRF resolution duplication is order-dependent in C++

E1 · HIGH · XGB data-analysis step not triggered automatically before XGB `anasum` cuts

E2 · HIGH · `IRF.mscw_energy_MC_sub.sh` XGB branch uses stale CLI and wrong run-parameter keys

E6 · HIGH · Eventdisplay-ML `config.py` calls `.get()` on a list

E7 · MEDIUM · XGB training helper scratch directory uses undefined variable `MSCW_FILE`

E8 · MEDIUM · `EVNDISP_MLSYS` documented but helpers require a hard-coded conda environment name

E10 · MEDIUM · Production XGB training helpers omit `--random_state` and `--hyperparameter_config`

E12 · MEDIUM · Legacy `TMVA.BDT.V4.runparameter` ships a malformed `PREPARE_TRAINING_OPTIONS` line

F2 · HIGH · Future epoch `V6_2025_2025s` is defined twice in the epoch file

F8 · MEDIUM · `download_irfs.sh` continues into extract/remove after a failed transfer

F9 · MEDIUM · `copy_dispBDTs.sh` and `copy_dispXGB.sh` silently overwrite `SIMTYPE` in a loop

G1 · HIGH · `EventDisplay_Release_v492` is still a skeleton, not an auditable release-test record

G4 · HIGH · Crab run-list generation: `WOBBB` typo leaves zero-wobble classification undefined

G5 · HIGH · Source-test plotting macros are pinned to the `v491` library and data layout

G7 · MEDIUM · Shipped `v492` run-parameter format encodes a fragile cut-name duality

H1 · HIGH · `evndisp` keeps time-sliced pedestal analysis enabled during normal data reconstruction

H2 · MEDIUM · `evndisp` repeats invariant low-gain/sum-window validation in the per-event loop

H3 · HIGH · `anasum` double-scans `mscw.root` and reinitialises TMVA twice per run

H4 · MEDIUM · `anasum` uses a linear TMVA-bin search with recomputed mean energies per event

H5 · MEDIUM · Multi-run `anasum` jobs scale as O(runs × total entries in chain)

H6 · MEDIUM · `evndisp` reads pedestal calibration inputs redundantly during initialisation

H7 · MEDIUM · `mscw_energy` repeatedly rebuilds TMVA datasets instead of amortising setup

X1 · HIGH · `v2dl3` silently sets quality flag to 0 when Eventdisplay logfile is missing

X2 · MEDIUM · `anasum` continues under missing TMVA-weight warnings

X3 · MEDIUM · `anasum` reinitialises the same TMVA state twice, doubling warning noise