Closing the Verification Gap

A concrete proposal to integrate Tandem's physical–digital assertion model into science-jubilee, so that Jubilee-derived laboratory machines (DuckBot, FungiBot, the Sonication Station) can detect their own failures instead of silently propagating them into data.

Built with research-infra — content lives in frontmatter-tagged Markdown under content/, and is assembled into both a LaTeX article and a Beamer deck.

The argument in one paragraph

Jubilee's machines are not meaningfully limited by their motion. They are limited by their inability to observe their own state: there is no motion-completion handshake, no tool-seating verification, no liquid-level detection, and no way to assert that the physical world matches the program's model of it. Every resulting failure is therefore silent — and a silent failure in a ten-day assay does not halt the experiment, it contaminates it. The Machine Agency already published the fix (Tandem, CHI 2024) and pointed it at a milling machine instead of at its own laboratory robots.

Build

Requires pandoc, xelatex, and the metropolis Beamer theme.

make            # both PDFs
make manuscript # manuscript/output/jubilee-tandem.pdf         (25 pp)
make slides     # manuscript/output/jubilee-tandem_slides.pdf  (20 frames)
make check      # verify: no unresolved citations, no missing glyphs, no frame overflow

Layout

content/            # the source of truth — frontmatter-tagged Markdown
  00-abstract.md
  10-introduction.md          The silent failure
  20-background-3dp-lineage.md        History of 3D printing → Jubilee
  21-background-liquid-handling.md    Hamilton, Opentrons, PyLabRobot, the physics
  22-background-science-jubilee.md    The platform and its machines
  23-background-tandem.md             The answer already written
  30-method-architecture.md           Three layers + the Sensor Rule
  31-method-sensors.md                The three sensors that make assertions bite
  32-method-workpackages.md           WP0–WP5
  40-results-measurement.md           The measurements the field is missing
  50-discussion.md                    Roads not taken, and why
  60-conclusion.md
  90-appendix-evidence.md             Evidence grade for every load-bearing number
  91-appendix-bom.md                  BOM + firmware checklist
manuscript/
  config.yaml       # research-infra ProjectConfig
  references.bib
  output/           # generated

Each file's YAML frontmatter carries section and weight (which set assembly order) and a slide_summary (which becomes the Beamer frame). The appendices deliberately carry no slide_summary, so they appear in the article but not the deck.

Conventions

• Every quantitative claim is graded in content/90-appendix-evidence.md: measured [M], claimed [C], no-data [N], or derived [D]. Where a number does not exist in the literature, the absence is the finding.
• Claims we could not substantiate are listed explicitly and not made.
• Two pandoc gotchas this document has already hit, recorded so you don't rediscover them: an inline $math$ whose closing $ is immediately followed by a digit is not parsed as math (write $\approx$ 390, not $\approx$390); and Unicode ≈ ≥ ≤ ₂ have no glyph in the default XeLaTeX text font — use LaTeX math instead.