Feeding & Passaging:
Maintain one source culture, passage it into a six-well daughter plate, and maintain that daughter.
- Command
- —
- Target
- —
- Volume
- —
Source Plate C
—Daughter Plate D
—Task & tools
Starting setup
- Source plate C: holds the existing cell culture in A1, one of its wells.
- Daughter plate D: six empty wells, coated to help transferred cells attach and grow.
- Medium: nutrient liquid that feeds the cells.
- PBS: phosphate-buffered saline, a salt solution used to rinse the cells.
- ReLeSR: a reagent that loosens cells from the plate for transfer.
- Pipette tips: disposable tips used to move liquids.
- Collection tube: holds the harvested cells for dilution and mixing before transfer to plate D.
Available tools
- Gripper and P1000. 12 simulated P1000 rack types with different orifices, filters and surfaces. Biotix uTip box sizes in the equipment reference: P20, P100, P200, P250, P300 and P1000.
- Microscope and incubators.
- Native tcode robot commands; separate imaging, review and incubation services.
Agent responsibility
- Build, validate and execute native tcode scripts.
- Choose tools, tips, well positions, volumes and step order for each aspiration and dispense.
- Schedule imaging, scientist review and incubation; respond to review decisions.
- Inspect interrupted runs and resume without repeating completed transfers.
Agent Context
Open Agent Context to read the files.
Protocol
Open the companion physical protocol →
Feed source plate C
- Image plate C and complete scientist review.
- Remove all old medium and add 2–4 mL of fresh medium.
- Take a new image and complete scientist review.
- Return plate C to the incubator.
Passage: transfer cells from C to D
- Prepare: confirm ≥70% cell coverage and empty D’s coating liquid.
- Release: wash C with PBS; apply and fully remove ReLeSR.
- Check: incubate 2–4 minutes, image and review. After the first negative review, incubate again for up to 4 minutes and repeat the check.
- Transfer: collect cells, dilute to 12 mL, mix and dispense 2 mL into each of D’s six wells.
- Finish: spread cells in both directions, image, review and incubate D. Record its lineage and discard C by 17:00.
Feed daughter plate D
Repeat the feeding workflow in all six wells. Complete each feed within ±2 hours of its due time.
Detailed steps & timing
Rubric
Runs & evidence
v10 · Four runs per model and reasoning level
60 attempts: 48 graded, 12 provider-blocked. Scores are mean fixed checks passed out of 45. No graded run passed all six gates. History audit flags remain visible separately.
| Model | Reasoning | Checks /45 | Pass@1 | Time (min) | API estimate | Output tokens | Steps |
|---|---|---|---|---|---|---|---|
| Claude Opus 5.5 | Low | 38.0/45 | 0% | 4.2 | $3.46 | 12.0k | 27.3 |
| Claude Opus 5.5 | Medium | 40.0/45 | 0% | 5.5 | $5.24 | 17.2k | 33.8 |
| Claude Opus 5.5 | High | 29.3/45 | 0% | 6.4 | $6.94 | 21.2k | 38.5 |
| GPT-6 Astra | Low | 36.5/45 | 0% | 3.5 | $1.03 | 4.0k | 22.8 |
| GPT-6 Astra | Medium | 39.3/45 | 0% | 3.7 | $1.09 | 5.5k | 22.8 |
| GPT-6 Astra | High | 40.0/45 | 0% | 4.0 | $1.55 | 7.5k | 23.8 |
| Grok 4.7 | Low | 18.3/45 | 0% | 7.6 | $1.00 | 27.0k | 39.8 |
| Grok 4.7 | Medium | 6.0/45 | 0% | 7.3 | $0.96 | 22.7k | 40.0 |
| Grok 4.7 | High | 2.8/45 | 0% | 11.6 | $1.18 | 26.3k | 40.0 |
| Gemini Argon | Low | Blocked | — | — | Unknown | — | — |
| Gemini Argon | Medium | Blocked | — | — | Unknown | — | — |
| Gemini Argon | High | Blocked | — | — | Unknown | — | — |
| Kimi K3 | Low | 4.8/45 | 0% | 11.0 | $0.58 | 15.7k | 40.0 |
| Kimi K3 | Medium | 6.3/45 | 0% | 10.2 | $0.80 | 22.2k | 40.0 |
| Kimi K3 | High | 3.5/45 | 0% | 8.0 | $0.85 | 23.9k | 40.0 |
Costs are mean API estimates per attempt, calculated from recorded input, output, cache-read and cache-write tokens at Standard US Bedrock rates. They exclude infrastructure and are not invoices. Argon returned no token counts; its billing is unknown.
Download table (CSV) · Token counts and pricing sources (JSON)
Selected high-reasoning trajectories
The viewer uses seed 0 for every model. These individual examples are distinct from the four-run averages above.
- GPT-6 Astra · high · seed 0 — 39/45 checks; 1/6 gates; fail.
- Claude Opus 5.5 · high · seed 0 — 23/45 checks; 1/6 gates; fail.
- Grok 4.7 · high · seed 0 — 3/45 checks; 0/6 gates; fail.
- Kimi K3 · high · seed 0 — 7/45 checks; 0/6 gates; fail.
- Gemini Argon · high · seed 0 — Blocked checks; — gates; unscored.
QC & findings
Sources & limits
The simulator uses symbolic liquid handling and scripted confluence, imaging and scientist reviews, without operating hardware or validating physical drainage, robot safety, cell quality or biological safety.