progress and prompt 1.7

# Day 4 Summary — Plan & Progress Checkpoint

**Status (as of pause):** ~70–75% context utilization. Sources read so far

listed below; remaining items listed at the end. Awaiting your go-ahead

to continue.

---

## Sources read in full

- **Day 4 AM transcript** (`2026-04-27-Part1-transcript.txt`) — opening

  remarks, transition to lunch, Bogdan's NDIP framing.

- **Day 4 PM transcript** (`2026-04-27-Part2-transcript.txt`) — covers

  John Duggan's NDIP presentation opening, then jumps to ~3 hr mark for

  the architecture discussion (settings.json critique, "ten levels of

  danger" electrical-wiring analogy, story-sketching for tomorrow).

- **`Notes 04-27-2026.docx`** (live collaborative Google Doc) —

  **authoritative for Day 4 outcomes.** Contains:

  - Working Plan (refactor sequencing, freeze/branch strategy, low-hanging

    fruit list)

  - 3 critical issues to resolve in current QuickNXS

    (run summing, off-spec binning, off-spec output)

  - Worth-exploring items (QML, MainView signal protocol)

  - **Proposed Structure/Design** with concrete `src/quicknxs/` tree

    (main / enums / assets / config / model / presenter / view / TBD)

  - Developer + scientist action items

- **`Background for architecture discussion.docx`** — Code excerpts of

  `MainWindow`, `MainHandler` showing the current handler-tangled state.

- **`Link to Bogdan AI repo.docx`** — pointer to the

  `quicknxsv2-modularization` branch.

- **AI Lunch and Learn transcript** — Bogdan presented his

  agentic-engineering work to NSD; covered Wireshark + Galil debugging

  precedents (S3 X-gap, BL4B TTHD), prompt-management discipline,

  CLAUDE.md memory architecture, ORNL legal authorization, ~$200/mo

  plan, context vs. token tradeoffs, ORNL-Assistant moderate-data work,

  and a long Q&A.

- **`research_mvp_architecture.md`** (962 lines) — Copilot's

  comprehensive MVP architecture assessment. Inventories all current

  components (`MainWindow`, `MainHandler` 2089 lines,

  `ConfigurationHandler`, `PlotHandler`, `PlotManager`, `DataManager`),

  documents the Qt signal/slot topology, the `auto_change_active`

  recursive-event guard problem, the direct-beam peak-change cascade,

  and proposes a 6-phase incremental migration with a two-phase

  Mutation+Render pattern as the recommended replacement.

- **Day 4 AM handwritten PDF (10 pages)** — covers:

  - "Ruff" testing discussion; Marie established in repos

  - Glass on pre-commit project template

  - Becky on documentation; agent automation possibilities

  - "Can I share?" — Marie's MVP framing for QuickNXS entanglement

  - Discussion of restructuring vs. add features

  - 3 critical Valeria items: cannot extract off-spec data; cannot

    add data; cannot rebin correctly

  - Refactor branch strategy: keep separate, "next" stays clean

  - Run summing fix — already in QA

  - Off-spec status — Valeria has lots of correct off-spec from old

    ders (back to 2019), V1 still works

  - Off-spec binning — keep in specular, disable in off-spec as

    proposed

  - Asma vs new pen to choose 'bin' (raw vs smoothed) — easy change

  - Spin assign (54) — discussion

  - Will provide V1 / V2 with double extraction — both nxs.histo and

    nxs.h5; Tim noted "lots of pieces in this window"

  - Becky: issue has been around for a while, scientists stopping

    their fixes — does Tim mean it will take time?

- **NDIP Overview PDF, pages 1–3** — covers high-level NDIP architecture

  (docker → ndip), components on github (basic script template),

  NOVA application options (TRAME script), Desktop apps (Qt, paravisor),

  Web-based applications (refl1d), example apps (TOPAZ Reduce, NOVA CT

  Scan Visualizer, JANA, 2030 application, refl1d as future).

## Sources still to read

- **NDIP Overview PDF, pages 4–10** (7 pages) — examples of deployed

  NDIP apps and discussion details

- **Day 4 PM handwritten PDF, 14 pages** — main session 7+8+9 content,

  story-sketching for tomorrow, the "ten levels of danger" architectural

  critique from the transcript

- **Notes 04-23-2026.docx** vs `2026-04-23-Notes-GoogleDoc.md` — diff

  to find any Day-4 edits to Day-2 notes

- **Notes 04-24-2026.docx** vs `2026-04-24-Notes-GoogleDoc.md` — diff

  to find any Day-4 edits to Day-3 notes (Notes 04-24 is 224KB vs the

  .md that was 7.5KB, so substantial Day-4 additions are likely)

## Summary plan (skeleton I will produce)

```

# Hack-A-Thon 2026 — Day 4 Summary (2026-04-27)

[Header: location B202; threshold = generate EWM tickets]

[Outcome: working plan agreed; 6 design pillars articulated;

 3 critical scientific issues catalogued with assignees;

 NDIP overview confirms web/Qt deployment paths;

 AI Lunch-and-Learn delivered (separate side track)]

[Sources block: notes 04-27 docx, both transcripts, 3 PDFs,

 architecture background, MVP research doc, AI lunch transcript]

[Attendees: room + remote, with NDIP guests for afternoon]

## 09:30 - 09:45 · Day 3 recap and Day 4 charge (Bogdan)

[draws on Bogdan-Day-4-Overview slide we built Saturday]

## 10:00 - 12:00 · Session 6 (refactor planning)

[AM PDF + AM transcript]

- Marie's "Can I share?" — MVP framing

- Three critical issues catalogued (run summing / off-spec binning / off-spec output)

- Refactor-branch strategy: freeze main dev, branch from refactor branch

- Low-hanging-fruit list (Configuration class split, settings.json removal,

  GenX templates, blockSignals, variable naming)

- Valeria's three off-spec gaps; V1 has correct outputs

- Asmaa+Valeria assigned to off-spec output reconciliation

## 12:00 - 13:00 · Lunch / AI Lunch and Learn (separate track in C-152)

[from AI lunch docx]

- Bogdan presented agentic engineering to NSD audience

- Concrete precedents: Wireshark+Galil S3 X-gap, BL4B TTHD

- Tooling: Claude Code, $200/mo plan, CLAUDE.md memory, SSHFS+rclone

- Concurrent discussion: code.ornl.gov outage during the session,

  prompt-organization discipline, "10x productivity" claim, ORNL

  legal-clearance status, Codex/Copilot alternatives, mono-repo Q&A

## 13:00 - 14:00 · NDIP Overview side meeting (John Duggan)

[NDIP PDF + PM transcript intro]

- High-level: NDIP = docker → analysis filesystem → HPC orchestration

- Three deployment paths:

  - Basic script (template on GitHub)

  - NOVA application (TRAME script)

  - Desktop Qt apps (TOPAZ paradigm)

  - Web-based apps (refl1d precedent)

- Examples: TOPAZ Reduce; NOVA CT Scan Visualizer (~200GB);

  Jana 2030 application; refl1d (general web app, NDIP provides path)

- Resource model: dedicated VM per session vs. analysis-node 64-core

  shared (today's "Quick Access" experience)

- Long-term: re-evaluate shared analysis system; per-user dedicated

  resources mean responsiveness, not contention

- Q&A on whether NDIP virtual interface will feel as responsive as

  shared analysis (John says: shouldn't notice)

## 14:00 - 16:30 · Session 7+8+9 (architecture deep-dive)

[Day 4 PM PDF + PM transcript late]

- "Ten levels of danger" — electrical-wiring analogy as the heuristic

  for how deep to refactor; Bogdan's friend-the-electrician story

- settings.json critique: dangerous if user-editable, structurally

  flawed — eliminate, replace with Python dict in code

- Tomorrow's plan: sketch stories for tech debt (continue design as one

  story); group by Day 3 AI-generated tech-debt list; combine with

  notes from prior days

- Splitting Configuration class — "going to be a bear, not low-hanging

  fruit but worth it"

- Spec output: update descriptions

- Folder reorganization

- "Ten levels of danger" → must fix dangerous things even if outside

  the original scope

## Working Plan (the centerpiece — from Notes 04-27-2026.docx)

[verbatim or near-verbatim from the doc]

## Proposed Structure/Design

[the src/quicknxs/ tree from the doc; reference research_mvp_architecture.md

 sections for justification]

## Consensus & Decisions (Day 4)

1. MVP pattern adopted (Phase-by-phase per research doc)

2. Two-phase Mutation+Render pattern (replaces auto_change_active)

3. Eliminate settings.json + Settings singleton

4. Run-summing critical fix already in QA — scientist verification

5. Off-spec binning hidden in UI for now

6. Off-spec output to be reconciled against V1 (Valeria's correct outputs)

7. GenX templates removed (Day 2 decision confirmed)

8. Refactor-branch strategy committed (freeze main; branch from refactor)

9. NDIP confirmed as deployment platform for both Qt and web frontends

10. AI tooling scope: agent-augmented refactor is in play (per Bogdan

    AI Lunch and Learn precedents)

## Open questions / non-decisions

## Day 5 charge

## Notes on fidelity

```

## Time / token budget remaining

If you say "go," I will:

1. Read remaining 7 NDIP pages + 14 Day-4-PM pages (≈21 page-images;

   each consumes ~1.5–2K context tokens depending on dense handwriting)

2. Diff Notes 04-23 / 04-24 docx against originals

3. Write the summary (target ~500–700 lines of markdown)

Total additional context cost estimate: ~30–40K tokens for inputs +

~10–15K tokens for the writing. This puts me right at the edge.

## What I'd ask you to consider

- **Skip the Notes 04-23 / 04-24 diff?** — These would only matter if

  the team retroactively edited Day 2 / Day 3 conclusions on Day 4. I

  consider that low likelihood; the Day 4 outcomes are best captured

  in Notes 04-27 directly.

- **Skip part of the NDIP PDF?** — The PM transcript already has John's

  high-level NDIP framing; pages 4–10 of the NDIP PDF are likely

  example-deep-dives that could be summarised more briefly. Reading

  half (pages 4–7) and skipping the rest would save ~10K tokens.

- **Read all 14 Day-4-PM pages?** — These are the *core* afternoon

  architecture discussion and should be read in full. Non-negotiable.

If you agree with the plan above, reply "go" (or "go and skip the docx

diffs") and I will proceed straight through to the summary. If you want

me to add or remove sections, say so before I continue.

Please read ~/Desktop/Hack-A-Thon-2026/2026-04-27-Files.md. Please use that information to summarise the Day 4 events and write them to tasking/Hack-A-Thon-2026-Day4-Summary.md for me to review. If there is any other information that you would like to have, please ask me. This one has two breakout sessions for which I have handwritten notes for only on (NDIP Overview). The other is a Teams transcript in which I (Bogdan) presented an overview slide that you helped me create in the 'ai-lunch-slides' branch of tasking. I have also downloaded the current contents of the shared Google drive that the team has been working on during the meeting (they are referenced in 2026-04-27-Files.md). This is going to be a big summary, so take your time, be thorough. Watch your context - if you get to 75% utilization, make sure you write a plan and let me review it. This summary may be the most important one of the hack-a-thon, so please work extra carefully!

```

  Where I am:

  - 8 of 10 source-reading tasks complete (transcripts, all docx, MVP research markdown, AI Lunch transcript, Day 4 AM handwritten PDF

  in full, NDIP pages 1–3)

  - Remaining: NDIP pages 4–10 (7 pages), Day 4 PM PDF (14 pages), optional docx diffs of Day 2/3 notes, then writing the summary

  Token-budget tradeoffs I'd like your input on:

  1. Notes 04-23 / 04-24 docx diffs (task #29) — likely low value (the team's Day-4 conclusions are captured directly in Notes 04-27,

  which I've read in full). Skipping saves ~5K tokens of input.

  2. NDIP PDF pages 4–10 — likely example-deep-dives; the transcript already gave the high-level framing. Reading first 3 pages +

  summarizing rest more briefly saves ~10K tokens.

  3. Day 4 PM PDF (14 pages) — the core architectural deep-dive; must read in full.

  The plan document at the path above describes the proposed final-summary skeleton, the consensus list I'm building toward, and the

  verbatim Working Plan + Proposed Structure I will pull from Notes 04-27-2026.docx (which is the authoritative Day 4 deliverable from

  the team itself).

  Reply when you're ready with one of:

  - "go" — read everything, full summary

  - "go skip docx-diffs" — read all PDFs, skip the docx diff (recommended)

  - "go skip docx-diffs and skim NDIP" — minimum-viable to land Day 4 within context budget

  - Or any specific edits to the plan you want before I continue

✻ Crunched for 2h 22m 19s

```

I review the [Hack-A-Thon-2026-Day4-Summary-PLAN.md](../Hack-A-Thon-2026-Day4-Summary-PLAN.md).

### Prompt 1.7

go skip docx-diffs. I have inspected the files and the only difference I saw, I collected into ~/Desktop/2026-04-24-Notes-GoogleDoc-diffs.md which you may read. When you finish, please produce a prompt that I can use in a fresh session to incorporate the upcoming Day 5 activities that will capture the most important aspects of this session. Thank you so much for this amazing work!