@@ -293,3 +293,45 @@ Ok, this is superb. One thing: would you review '/home/6ov/Pictures/Screenshots/
Parent gitlink bumped in 6b6373a. Each matplotlib-backed slide's SVG is large because matplotlib embeds the rendered paths for every data point — that's why the slide 4 and slide 7 SVGs are hundreds of KB; the PNGs are typical. Not a problem for
git, but worth flagging if the repo is space-constrained.
```
### Prompt 2.4
The scientists and I enjoyed the slides! Let us continue to refine them. I went ahead and merged 'quicknxsv2-modularization-with-mantid-elaboration' back into 'quicknxsv2-modularization' to continue with this prompt because the scientists are reviewing the slides from that branch now. I would love for you to work on the following tasks (in whatever order produces the best outcome - fewest transcription errors, highest accuracy to the behavior of the software and fidelity to reflectivity):
* I would like to have an .md file for every slide (if possible) or a .py file for every slide (if .md is not possible - similar to gen_slide4.py or gen_slide7.py) and a process for generating it. The purpose for this is to allow me to make modifications to the generated content. Let me know if there is any clarifying question on this.
* In the '01-repo-structure.md', please add live links to the repositories (to help people navigate to source if they wish)
* In the '01-repo-structure.md', you may revise the assessment to be authoritative now, the hack-a-thon begins tomorrow
* In all .md and slide files, please change 'mr_core' to 'ref_core' (the scientists have agreed to 'ref_core' as the name)
* In the slides, we need to express the message that the modularization must be done in such a way that it can support two instruments (even if we only work on one at a time).
* In the slides, we need to articulate clearly the "stretch" goals of the hack-a-thon.
* In the slides, we need to ensure that the "threshold" goals for the hack-a-thon are presented as the primary message.
* In 'slide-1-current-vs-ideal.svg', please modify:
- remove the 'lr_reduction (sibling)' box in the 'TODAY - tangled' section for now. It may clutter the message.
- lable every arrow line with a 'Calls' label (to make it clear that this is a call relation, not a class inheritance relation - this was confusing to the scientists)
- reduce the amount of test on the screen (I think it may be too complex to render well on a presentation screen)
* In 'slide-2-api-mapping.svg', please clarify that where lr_reduction is referenced, it is the 'standard' reduction in the 'next' branch and *NOT* the lr_reduction:new_workflow. This is the mantid-centric version. In fact, it is probably correct to create a data unit on cross-references between mr_reduction and lr_reduction (mantid-centric) as a separate understanding from lr_reduction:new_workflow (h5py+numpy) vs quicknxsv1. Do not include in the slides that discuss lr_reduction:new_workflow at this time (the scientists have agreed that it creates too much confusion of the message at this stage, and it is better left for a later part of the hack-a-thon if the direction that the team is taking a direction that is amenable, or for a future effort otherwise).
* Please create a series of detail slides derived from the contents of 'slide-3-silent-defaults.svg' such that:
- There is a slide for each of the items {'ErrorWeightedBackground', 'CropFirstAndLastPoints', 'RoundUpPixel', 'AcceptNullReflectivity', 'UseSANGLE'} that contains the detail text from the current 'slide-3-silent-defaults.svg'. This allows more room to express the detail in the slide.
- Simplify 'slide-3-silent-defaults.svg' such that the contents in the columns {'Parameter', 'quicknxs (GUI)', 'mr_reduction (auto)'} remain, while the {'Physics...', 'Observable...'} columns are replaced with a representative image/icon (which is now possible because the details are present in separate slides).
* On 'slide-3-silent-defaults.svg', in the Physics section, for 'ErrorWeightedBackground' remove the text that says 'Inverse-variance mean...' and keep only the sentence that starts with 'Uniform mean...'. The inverse-variance mean weighted by 1/sigma^2 is not correct for this dataset.
* On 'slide-3-silent-defaults.svg', in the Observable section, for 'ErrorWeightedBackground' rewrite the text 'Background estimates differ...' with 'Backgrounds used when high intensity off-specular reflections are present unfairly bias low count reflectivity'. I have sketched a graphic in 'slide3-ErrorWeightedBackground-Explanation.jpg' that I would like you to read/simplify/incorporate into the information slide detail for 'ErrorWeightedBackground'.
* On 'slide-4-curves-disagree.svg', please do not synthesize a reflectivity mock data - the scientists object very strongly to unphysical reflectivity (and you need to make a note to remember this persistently in the memory system). This means that you must rewrite 'gen_slide_4.py' to do an *actual* reduction with both 'quicknxs' and 'mr_reduction' that highlights the callouts (1. autoreduce drops first and last 2 Q bins, 2. ErrorWeighted background oversubtracts). NOTE that callout 3 "post-hoc scaling" is not correct/does not exist and should not be referenced. If you need a third example, try finding a reduction that exposes the 'RoundUpPixel' silent default. This could be a substantial investigation to find a set of files that expose this, but you have access to the entire dataset in '/SNS/REF_M/**', and the night is young as they say...
* On 'slide-4-curves-disagree.svg', add a thumbnail image that refers to the line(s) in 'slide-3-silent-defaults.svg' that are being illustrated. I know you used callouts, but it helps to have a small thumbnail from a prior slide as a visual mnemonic.
* On 'slide-5-peak-drag-sequence.svg', please move item 9 to exist *below* item 8, and add an indicator between item 2 and item 9 that the UI is *locked out* during this time. Please make a separate slide to detail the section on the botton of 'Where the debt accumulates - five patterns scientists perceive as "Qt leaking into the reduction". The detail is needed, but it will be too hard to see on the screen - a 2nd slide will relieve the pressure.
* In general the slides are currently *very dense*. Consider ways to 'reduce density' of the textual information on the screen. You can make more slides to make the points (there is no limit to the number slides, compact dense slides are not a goal for this effort). You can convert certain points into graphics (like you did with Slide 4 - just please do not fabricate data; either make an obvious cartoon to illustrate the point (perfectly acceptable) or do an actual reduction from a real dataset (and reference the sources of data used).
* Cut 'slide-6-reduce-contract.svg' - this is a discussion for the hack-a-thon, it should not be a pre-conceived idea.
* Hide 'slide-7-debt-triage.svg' this is a great slide, but it should not be presented at this time. Instead, create an .md document in the knowledge base with the data in tabular/spreadsheet format for digestion and discussion.
* Hide 'slide-8-equivalence-harness.svg' this is another great slide, but it assumes that the "stretch" goal path is to be followed. It is not clear before the hack-a-thon if the developers will be able to achieve the "stretch" goals and putting it up at the start may have the opposite effect. It may be something that can be brought up during later days of the hack-a-thon if things are going welll.
* Please make a final slide (similar to the 'slide-1-current-vs-ideal.svg' that makes for a good closing slide to re-iterate the messages are trying to make clear, and which has a playful tone. Perhaps a tasteful graphic generated by your favorite image generator that captures the essence of this endeavor?
* Please make an opening slide that has a funny graphic. During the review, the scientists used the term "battle of the robots" to refer to the two sessions planned for Day 1: "Overview of Scientist Assessment" (this presentation), "Overview of Developer Assessment" (which the developer team is working on independently). I thought of a toy from my childhood called "Rockem-Sockem-Robots" and found this .gif on Giphy: . Can you incorporate it?
* Please collect all the individual slides (which I want to continue to have in .md, .svg and .png as you have made them) into a single .pptx file that I can upload and present for the "Overview of Scientist Assessment".
For additional background detail to the hack-a-thon, please review the 'hack-a-thon-2026' branch in the 'tasking' repository. It has the current agenda 'Hack-A-Thon-2026-Agenda.docx' updated within it.
May I request that you perform the above steps using Agent Teams such that you can maintain a proper context for the work? If your context gets close to the 'auto-compact' threshold, please ensure that you can write out a proper plan *prior* to the auto-compact occuring. In this way, I think you will produce better output result, is my thinking correct? If it is not correct, then I defer to your judgement as to the management of your agent teams and context buffer.
I really appreciate the enormous amount of work that you are doing for this presentation. Thank you so much!
