GCC2026 Talk — Context for Future Agents
Status: Active planning, MVP target ~1–2 weeks from 2026-05-23. North-star plan with explicit MVP/polish/stretch tiers so any of these slides can be presented even if downstream features slip.
Owner: jmchilton. Slide implementation lives in:
~/projects/repositories/jmchiltondotnet(Astro + reveal.js, MDX, route auto-publishes at/slides/gcc2026). New deck file:src/content/slides/gcc2026.mdx. Dev:make dev. Build:make build. Reveal.js version note: package.json declaresreveal.js ^5.2.1(CSS only, imported atSlideLayout.astroline 3); the actual JS runtime is loaded from CDN atreveal.js@4.5.0(lines 181/185/190). Target v4.5.0 JS features until the layout is upgraded. All planning docs (this file, slide outline, demo storyboards, status logs) live here invault/projects/history_markdown/gcc2026/.
Thesis (the two-minute version)
We’re about to automate the reproducibility crisis. Agents scale analysis faster than they scale trust. Galaxy’s response should not be to adapt away from its core principles; it should be to double down on them. Infrastructure designed for human reproducibility is exactly what agents need — and shipping it as agent-grade tooling accelerates and incentivizes reproducibility on both sides.
The case-study payoff: two new ways to construct Galaxy workflows that both lean on the same reproducibility infrastructure, demonstrated against the same scientific application.
Shared application — MRSA mobile-AMR comparative analysis
Both demos use the MRSA mobile-AMR demo vignette described in
jmchilton/galaxy-brain#12:
- 3–4 complete MRSA isolate assemblies (KUN1163, KUH140013, KUH140046, KUH180129 from BioProject PRJDB8599 / Hikichi 2019).
- Tools:
staramr,Bakta,ISEScan,IntegronFinder, table→GFF3 conversion, JBrowse. - Question: which resistance genes appear in mobile genomic contexts (plasmids, IS, integrons, SCCmec) across related isolates?
- Anchors in GTN: bacterial-genome-annotation, amr-gene-detection, mrsa-illumina tutorials.
Status (as of 2026-05-23): storyboard only. Real Galaxy history is expected to be ready in time but is not the gating risk — both demos can be presented from storyboard/seeded-history if the real run slips.
Presentation budget (10 min slot, 2 min Q&A reserved)
| Block | Time | Content |
|---|---|---|
| Title + thesis | ~2:00 | ”Automating the reproducibility crisis” + Galaxy’s role |
| Framing both case studies | ~0:15 | ”Two new ways to construct the same workflow” |
| Case study 1 — notebook-driven extraction | ~3:00 | embedded heavily-edited demo of #22709 against MRSA history |
| Case study 2 — building the workflow without Galaxy | ~3:00 | embedded demo of Foundry interview→Galaxy pipeline + gxwf validation, same MRSA app |
| Synthesis + paper pointers | ~0:30 | three papers as citations; site as long-video pointer |
| Total presenting | ~8:00 | leaves ~2:00 for questions |
Key referenced documents
Abstract & exec summaries (this directory)
ABSTRACT_CHATGPT.md— canonical talk abstract (Reproducibility in the Age of Agents).HISTORY_NOTEBOOKS_EXECUTIVE_SUMMARY.md— what was built for History Notebooks; framing for case study 1.WORKFLOW_STATE_EXECUTIVE_SUMMARY.md— Format2 + gxwf + Foundry stack; framing for case study 2.CLAUDE.md— narrower scoping note about this directory.
Paper manuscripts (the three citations the talk advertises)
vault/papers/galaxy-notebooks/manuscript.md— History Notebooks paper draft (substantial).vault/papers/gxwf/manuscript.md— Format2 + gxwf authoring/validation paper draft (substantial).vault/papers/foundry/manuscript.md— Foundry paper (empty; seevault/papers/foundry/outline.md,index.md,case-study.md).
Implementation references
- Galaxy issue galaxyproject/galaxy#22709 — notebook-driven workflow extraction (the case-study-1 mechanism).
vault/projects/history_markdown/GRAPH_WORKFLOW_EXTRACTION_PLAN.md— blocked-pending-gate plan for the extraction path.vault/projects/history_markdown/HISTORY_MARKDOWN_ARCHITECTURE.md— architecture for History Notebooks.vault/projects/history_markdown/SCREEN_CAST_PLAN.md— pre-existing screencast plan withseed_demo_histories.pypatterns. Reuse for MVP demo 1 if MRSA history slips.- Galaxy issue galaxyproject/galaxy#21475 — original History Notebooks proposal.
- Branch:
history_pagesin~/projects/worktrees/galaxy/branch/history_pages. - Workflow-state branch:
wf_tool_state. vault/projects/workflow_state/— CURRENT_STATE.md, PROBLEM_AND_GOAL.md, VSCODE_EXTENSION_PLAN.md.
Slide infrastructure
- Existing slide system:
~/projects/repositories/jmchiltondotnet/src/content/slides/*.mdx+src/layouts/SlideLayout.astro. - Reference deck for style/format:
claude-lab-talk-2025-12.mdx. - New deck (to create):
src/content/slides/gcc2026.mdx. - Build:
make dev/make buildfromjmchiltondotnet. - Decision: extend existing infra. Reveal.js is rich enough (fragments, video embeds, transitions, code highlighting). Don’t migrate.
Slide outline
The slide-by-slide outline (with MVP/polish/stretch markers and per-slide notes) lives in SLIDE_OUTLINE.md in this directory. Read that file alongside this one.
Demo videos
Two embedded YouTube videos, each ~3 minutes after heavy editing. Storyboards are in:
DEMO1_STORYBOARD.md— history → notebook → graph extraction → workflow report (#22709 over MRSA).DEMO2_STORYBOARD.md— Foundry interview-to-Galaxy → gxwf-validated Format2 workflow (same MRSA app, no Galaxy server in the loop).
Note: demo 2 deliberately does not mention Nextflow conversion. The Foundry mode being demonstrated is the interview→galaxy pipeline (currently in development). Keep this consistent across all materials.
Tiered scope — MVP / Polish / Stretch
Every component is tracked at one of three readiness tiers so that if a feature slips, the talk can present a degraded-but-coherent version of the same material.
Tier 0 — MVP (present even if everything below slips)
- Title + thesis slides — all content already written in
ABSTRACT_CHATGPT.md. - Both case studies present as conceptual slides + storyboards: still-frame diagrams, parameter-tree screenshots, narrated walkthrough.
- A single static screenshot or 30-second clip per case study, even if the full 3-minute video isn’t recorded yet.
- Paper pointers at end.
- “Long video coming” pointer (site link), even if the long video isn’t recorded yet.
Tier 1 — Polish (target for talk day)
- Full 3-minute edited demo 1 video against a real (or convincingly seeded) MRSA history.
- Full 3-minute edited demo 2 video showing Foundry interview→galaxy on MRSA application with gxwf round-trip.
- Site auto-publishes the deck at
jmchilton.net/slides/gcc2026. - All three paper drafts have public URLs to cite.
Tier 2 — Stretch (nice-to-haves)
- Foundry visualization showing the workflow growing step-by-step under Mold execution with gxwf gating each transition. (See
FOUNDRY_VIZ_IDEAS.md— placeholder, to be filled in by jmchilton.) - Real (not seeded) MRSA history with biological story (mobile context differences across the four isolates) usable as a paper figure.
- Long-form standalone video (see parallel track below).
Parallel track — longer standalone thesis video
This is planned in parallel with the conference talk and shares assets but has its own timeline. Conference talk is essentially a “trailer” for this video and the three papers.
Goals:
- Longer thesis development (10–15 minutes on the reproducibility-crisis-meets-agents argument, vs. 2 minutes in the talk).
- Deeper Foundry segment (covering the compiler framing from
vault/papers/foundry/index.md). - Reuse the 3-minute case-study clips from the talk verbatim where useful; record extended cuts where not.
- Live independently on YouTube; link from the talk’s final slide and from each paper’s index.
Tracked in LONG_VIDEO_PLAN.md in this directory.
Risk register
| Risk | Mitigation |
|---|---|
| MRSA history not run by talk day | Use seeded history from seed_demo_histories.py style; demos still functional. |
| #22709 / graph extraction merge slips | Demo 1 reverts to existing extraction surface; narrative still about “notebook-driven extraction” with graph view in storyboard. |
| Foundry interview→galaxy pipeline not demoable | Demo 2 falls back to gxwf-only authoring in VS Code; Foundry framed as forthcoming layer. Still tells the “Format2 as agent-writable artifact” story. |
| Long video slips past talk day | Talk’s “long video coming” pointer becomes a “papers in preparation” pointer to vault drafts. |
| Slide infra changes needed | Existing reveal.js setup is rich; no migration anticipated. |
Open questions
- Is the conference recording posted publicly? (Affects whether long video should redundantly cover the same case studies.)
- Does jmchilton want a hosted demo URL on slides for live audience interaction, or video-only?
- Should the long-video and conference-talk script share a single narration source-of-truth (markdown), or diverge?
- Does the foundry interview→galaxy pipeline have a name yet for the slide titles?
- When the gxwf-ui browser editor demo is mature enough, does it replace one of the embedded videos or live alongside?
Working agreements for future agents on this project
- Don’t introduce new abstractions in the slide source unless asked. MDX files are short — keep them short.
- Storyboard before recording. Don’t record demo footage without an updated storyboard in this directory.
- Track readiness in
STATUS.md(create if absent) — one line per Tier 0/1/2 item with date stamp. - Match the abstract’s voice.
ABSTRACT_CHATGPT.mdis the canonical voice; older drafts inold/are not. - Don’t mention Nextflow conversion in demo 2 materials. Foundry mode for this talk is interview→galaxy.
- The talk is an ad for the three papers. Every slide should be either a thesis beat or one of: notebooks-paper preview, gxwf-paper preview, foundry-paper preview.
- Plot for failure. A talk that needs every feature to land is fragile; we explicitly track Tier 0 so that something credible is presentable from day one.