# Tau lphi REVIEW
###### tags: `marseille` `belleII`

## 2023/02/03

## Conference note editing ( first round)
### from Soeren
* answer to the 20delta coverage, not crucial since it reduced again in the final SR for the yield extraction.
- [x] describe the ISR and FSR tails in the plots
- [ ] remove fig.1 and rephrase the current most stringent experimetnal limits?
* comments on plots:
* on Fig.2: put decay mode? change legend?
* left/right align?
* resize binning (fig.4) and X axis ( fig.6)

* explain to the reviewer line 267-268: RSB does treat DeltaE differently, with the additional 2(3)delta band selection
* syst on LID, sum in quadrature stat and syst, not correct to take the maximum difference -> recompute!

## Answers to RC, second round
official answers sent to reviwers:

Draft document here: https://docs.google.com/document/d/1z9zkNIGbDFGktnH-ICRnc6qGBZPU05sfXdHiH_1zr_c/edit

## AFTER partial unblinding, post B2GM

* CLs method https://indico.belle2.org/event/6837/contributions/35354/attachments/16695/24887/tenchini_tla_17052022.pdf
* asymptotic used for observed UL computation in tla assumes to be in NOT free bkg scenario
* with single test poin and no template fit shoudl be used toys ( more rough estiame in larger range and finer toys study )
* to validate CLs show the profile likelihood ratio, pg.8
* it's about computing the right side area on this plots, hist factory implementation of CLs in C++, needs to show the likelihood of the test statistics, input used to produce Cls
* try to increase toys stats in fewer poins, but more dense around the intercept

## Pre-unboxing summary at tau meeting 28/02/2023
* presentation here: https://indico.belle2.org/event/8719/contributions/56615/attachments/21168/31337/zani_taulphi_unboxingRequest.pdf
* pg.5, uncertanties computation with Bayes theorem raised doubts and question by Diego. Daniel said he was jsut suggesting not to include in the poissonian fluctuation the MC scaling.
* paper refernce given https://arxiv.org/abs/physics/0701199v1
* let's expect also other collaborator may ahve questions here, not the standar way to provide uncertanties, but still well refernced
* we compute asymmetric uncertaties but show symmetric ones on data dots at pg.7. This should be correct, ucnertatinties have to be asymmetric in plots as well
* pg.6, start at 0 with both axis
* produce already the toys for CLs, data not needed for that computation
* * pg. 7, DeltaE dsitribution for electron channel show visible data excess around 0.2
* attributed to residual (~20%, not vetoes) phi-photoptoduction pollution
* suggested to add graphical artefact (stack uncertainty to qqbar SM bkg) to account for an additional 20% expected but not simulated in MC bkg to cover data excess at pg.7
### ACTION ITEMS:
- [ ] replot data/MC agreement pg.7 with
- [ ] asymmetric error bars
- [ ] added graphical artefact around $\Delta E$ = 0.2-0.4 of ~20% of MC bkg bin yield ( as dashed uncertainty?) to cover the data excess ( known to be possibly due to residual pollution from not simulated bkg)
- [ ] modify CLs axis starting from 0
- [ ] start computing toys for CLs unboxed plot with larger stats?

## Tau meeting 230/03/07: unboxing report
* remove delta and give the mass ranges, use different delta for electron and muon, $\delta_{e}$, $\delta_{\mu}$
- [x] modify fig.2, 4 caption removing any mention to $\delta$, only SR, SB, RSB , https://docs.belle2.org/record/3399/files/BELLE2-CONF-DRAFT-2023-002%20%28v4%29.pdf
- [x] unweigthed MC yields for SB corresponding to the 7.4 and 7.3 final weighted yield, l 267 "remaining background events from the simulation are 7.4 and 7.3 for the electron and muon modes; in data, we observe 12 and 9, respectively." here: https://docs.belle2.org/record/3399/files/BELLE2-CONF-DRAFT-2023-002%20%28v4%29.pdf
- [x] ask Paul on implementation of CLs validation plots ( profile likleihood ratios)
* UL variation as a function of the SR studied here: https://indico.belle2.org/event/7915/contributions/46720/attachments/20044/29692/TauGM-2022.12.20.pdf
* suggest not to optimize in future and fix the signal box

* analysis code must be publicly accssible by collaborators, migrate under Tau repository on gitLab
- [x] ask Michel to create tau l phi repo
- [ ] ask Leonard missing inputs: s-weigths, dataframe creation
- [ ] migrate Leonard code
- [ ]

* comments on v4 from Ami:

> Abstract:
> line 4.
> after you introduce ee -> tau tau, you need to explain that one tau decays to the signal mode, then what is happening to the other tau
> line 6 and throughout the draft
> - don't define C.L. in the abstract
> - don't give 23 as the 90% confidence level upper limit, always put x 10^-8
>
> l 109-110 Not clear which phenomenon you refer to. It reads like the accidental conversion of the lepton flavour is the phenomenon.
> l 111 - ref 1 is for muon decays. Give also the one for taus.
>
> l 117 , -> . delete and

> l 123 I would be hesitant to mention here the 50/ab given that we don’t know when and if we will get that 50/ab. Already with this 190/fb you reach 10^-8. So the motivation should be we are already there testing that particular model.
> l 128 report on a search 

>
> l 130 why 12 and 13 are cited here? you need the detector only citation
> l 150 the lumi should be 189.9/fb -> also throughout the draft
> l 153 remove centrally produced
> l 153 don't use MC samples, instead simulated data or samples. -> also throughout the draft
> l 155 remove "is forced to", decay -> decays
> l 155 SM -> not defined. I would suggest to spell it out
> l 157 "whose decay" -> rephrase, reads like the decay of KKMC
> l 157 EVTGEN doesn’t start from e+e-. it starts from γ*, remove e+e-
> l 158 it is KKMC for ee → qq̄ and then the hadronisation with PYTHIA
> l 162 delete the last sentence
>
> l 174 usually we avoid “tracks” . You can say charged particle trajectory -> also throughout the draft
> l 189 the π0 reconstruction description should come first
> l 194 event shape variable is not a common knowledge. Define the variables. But you I suggest instead you define your photons. That would mean remove "... are also taken.... "
> l 195 Move kaon sepection before the neutral selection. First concentrate on the signal reconstruction: leptons, then K and φ, then neutrals
> l 195 hadron -> kaon. you use kaonID, kaon mass hypothesis
> l 198 PID not defined, spell it out
> Fig 1 define the BG and signal in the caption
> Fig 1 Caption what are electron and muon modes? define in the text
> Fig 1 have you checked how this plots will look if printed black and white?
> lines before 201 - You should start from the ee → ττ, then talk about the τ decays to signal.
>
> l206-210 this is very long and hard to read sentence. Introduce the ROE, then in a separate sentence the requirements.
> Table II caption - don't put moun in (), make it part of the sentence. -> -> also throughout the draft
>
> l 227 you need a sentence saying there is a BG contribution in the signal window.
> l 229-231 To remove the low-mult BG, we expolit …. for the electron channel. Then a similar sentence for the muon channel.
> l 234 another very complicated and long sentence. We require the 0th CLEO cone to be less than 8.5 GeV/c, etc….
> l 239 surviving -> remaining
> l 240 event-shape variables - define
>
> l 252 remove "at the time that analysis is performed"
> l 252 . Therefore...
> l 255 ... is required to be larger…. -> same for the second half.
> l 258 sys unc - where is the sys coming from?
> Fig 3 you have points that are not visible. change the y axis scale
>
> l 275 , → . Since
> l 277 1-> one, 3 -> three
> l 279 remove "by applying Eq 2 and we obtain" -> to be
> l 281 the last sentence without line number belongs to the caption of Fig 4
> Fig 4 Caption, the last sentence belong to the main body
>
> l 298 15 -> 15 x 10^-8
> l 301 - what are detector effects?
> l 312 within their stat and sys uncertainties
> l 315 avoid putting in (). better say 0.8% and 0.3% for the electron and muon modes, respectively.
> l 317 ref[30, 31] Belle II notes cannot be citations. If no public document is available, remove the citation.
> l 322 new sentence after ","
> l 323 why unity ? Do you mean you calculated the trigger eff. in data and MC, then the ratio, that has to be a unity? otherwise not clear why the trigger efficiency deviated from unity is a sys source.
> l 326 this is not clear. You are testing whether the BG suppression affects the signal distributions?
> l 328 data-MC -> this is jargon
> l 330-331 we extract correction factors, measured on variables used in 
the BDT selection which show the largest disagreements. 

> l 331-332 We then retrain the BDT on the bin-by-bin corrected….
> l337 imperfections of the B-field description used in the event reconstruction, misalignment of the detector, and material budget mismodeling.
> l 337 momentum
> l 341 when using the nominal scale factors
> l 349 there is a gap here. You estimated the various sources of sys… and then? how do you get the impact on the UL? In section 6 it is only about the UL without taking into account the sys. effects.
> l 356 don't put the reaction in the middle of the numbers and in (). It should be part of the sentence.
> l 356 remove "according... "
>
> l 359 190 -> 189.9
> l 360 experiment -> detector
> l 362 to the end. This has to be rewritten
> For the muon channel, the OBSERVED 90% C.L. upper limit improves by a factor of X% and Y %, compared to the previous measurements conducted at belle and babar experiments. For the electron ….
>
> I don’t think the lumi is important here. If you want to stress the lumis, then it should be in a separate sentence.
>
> l 363 why do you talk about the expected UL and not observed?
> l 364 don’t give a range for the UL improvement. There are two numbers to be compared for each experiment, so two numbers, not a range.

### PRofile likelihood
* checking on muon this might have two components
* why do we redo toys everytime?
* generates ones, save them, use always the sme toys
* for electron seems the observed input is recomputed for 0 obs event
* bugged found: data_obs was set equal to bkg

### AMi meeting:
* lepton flavour with hyphen
* 5/ab, or pre LS2 data set?

# On upper limit setting (17.03.23)
* hypo inverter in roostat: https://root.cern.ch/root/html532/tutorials/roostats/StandardHypoTestInvDemo.C.html
* oneside likelihood based approach with bands:
https://root.cern/doc/master/OneSidedFrequentistUpperLimitWithBands_8C.html
* for frequentist calculator: https://root.cern/doc/master/FrequentistCalculator_8h_source.html
* configuring the workspace, model building:
* https://root.cern/doc/master/ModelConfig_8h_source.html