Patterns Workflow Benchmark
Seven experimental modules and two composite workflows connect laboratory procedures with inspectable programs and execution records. The collection is maintained as an ongoing benchmark. Versioned results retain the original paper evaluation alongside later implementation results.
Results by version
| Snapshot | Source steps | Completed execution steps | Status |
|---|---|---|---|
| 1.0.9 | 173/173 | 2700/2700 | Release evaluation, 19 October 2026 |
| 1.0.7 | 172/173 | 2675/2675 | Paper evaluation |
Culsma 1.0.9 represents the ELISA plate-tapping step with TAP and an independent GENTLE requirement. Module 07 reaches 42/42, and the nine-entry release inventory reaches 173/173 source steps with all nine entry programs completing. The input source inventory and counting rule remain the same. Each snapshot retains its own inputs, runtime identity, manifest, results and reproduction instructions.
Additional workflows — Qubit and double-sided SPRI
Evaluated with Culsma 1.0.9 on 19 October 2026, using the same numbered-source correspondence rule and reproduction script. The two submitted module procedures and their composite are recorded separately from the fixed nine-entry paper inventory.
| Entry | Source steps | Coverage | Completed execution steps |
|---|---|---|---|
| Module 08: Qubit 1X dsDNA HS quantification | 13/13 | 100% | 35/35 |
| Module 09: Double-sided SPRI selection | 21/21 | 100% | 56/56 |
| Composite 03: Qubit quantification and double-sided SPRI selection | 3/3 | 100% | 86/86 |
The two modules cover 34/34 source steps (100.00%). The separate three-step composite source describes input setup and module handoff; including it gives 37/37 (100.00%) across the three new entries. All three programs complete with zero failures, skips or runtime diagnostics.
Qubit S8–S10 are represented together by the calibration procedure's declared readout structure and ordered standard measurements. Each invocation expresses the calibration procedure for that run at the language level. Structural matches retain finer boundaries described with each source: timed holds do not verify clarity or dryness, and a repeat wash does not model a separate residual-ethanol removal. Stub readouts do not provide a numeric DNA concentration; the fixed-volume SPRI procedure passes the field without using it to calculate bead amounts or DNA recovery. Six nonblocking header-return deprecation warnings remain in the composite validation record.
- Source inventory, boundaries and reproduction command
- Pinned inputs and expected results
- Generated summary and coverage table
- Original submission and author records
Paper evaluation — Culsma 1.0.7
Released on 21 September 2026.
Paper source-step correspondence
The evaluation matches 172/173 source steps (99.42%). All 9/9 programs complete in software, with 2,675 execution steps and zero runtime diagnostics.
| Module | Source steps | Coverage |
|---|---|---|
| Module 01: Cell transfection | 13/13 | 100% |
| Module 02: Selective plating and colony isolation | 12/12 | 100% |
| Module 03: Cell lysate preparation | 10/10 | 100% |
| Module 04: Suspension-cell staining | 29/29 | 100% |
| Module 05: Magnetic-bead immunoprecipitation | 23/23 | 100% |
| Module 06: Western blot | 32/32 | 100% |
| Module 07: Sandwich ELISA | 41/42 | 97.6% |
| Composite 01: Flow-cytometry immunophenotyping | 5/5 | 100% |
| Composite 02: Magnetic-bead IP with Western blot | 7/7 | 100% |
Source steps are reported as matched / total numbered source steps. Coverage checks structural correspondence between source instructions and program statements; it does not establish complete semantic fidelity, biological outcomes, or physical device execution.
Programs and reproduction
The versioned GitHub packages are the authoritative source for programs, evidence and reproduction instructions. The links below reproduce the paper evaluation; newer snapshots are linked in the version table above.
Evolution after the paper evaluation
The 1.0.7 paper snapshot retains plate tap-mixing as an explicit uncovered step; the separately recorded 1.0.9 evaluation resolves it. Module 02 represents surface spreading, while mapping selected observation rows to colony material references still requires an explicit declaration. The repository records these boundaries alongside the source procedures.
A maintained collection
Future snapshots can add workflows and incorporate verified improvements. Comparisons use identified source inventories and checking rules, with previous snapshots retained in Git history. The worked manuscript examples have their own checks and are excluded from the coverage totals above.
- Package: a76a98b98dab5e8edd76eefbb7dc5cc7ac447a64
- Evaluated implementation: 5364676bc2437b4981a00ce0133c730af243e469