tasks/task_043_antisolvent_crystallization/instruction.md

# Joint ATR-FTIR/FBRM Reconstruction of Supersaturation During Antisolvent Crystallization and Differentiation of Growth, Nucleation, and Agglomeration

The process team applied three antisolvent addition profiles—FAST, MEDIUM, and SLOW—to an API and evaluated both seeded and unseeded operation; each combination has two independent replicate batches. The attachments contain synchronized temperature, antisolvent volume, ATR-FTIR absorbance, and counts for six FBRM chord-length bins, together with ATR concentration calibration data, a solubility surface, FBRM solids-content calibration data, batch material records, and endpoint mother-liquor HPLC data.

Reconstruct the solution concentration, solubility, supersaturation, chord-length moments, and solids-content trajectories for each batch, and identify signatures of new nucleation, crystal growth, desupersaturation, and agglomeration. For seeded batches, first account for the initial seed-crystal counts and seed-crystal mass, and then evaluate newly formed particles. Summarize the mean/SD for the six process combinations, verify the endpoint HPLC results and material closure, and determine the recommended condition according to the process-selection criteria in the attachments. The report must explain the limitations of relating FBRM chord length to true particle size/mass.

Generate analysis.py, calibration_models.csv, process_trajectories.csv, run_metrics.csv, design_summary.csv, crystallization_qc.csv, supersaturation_and_solids.png, fbrm_population_diagnostics.png, process_window.png, and crystallization_PAT_report.md under /app/output. The tables must retain run_id, profile, seeded, replicate, time, and units; all numerical values in the figures and report must be verifiable from the CSV files. Do not read data from outside /app/input, and do not hard-code reference results.

## Deliverable Files

This section defines only the submission interface and does not prescribe the analytical methods or expected results. All 10 files listed below are required and must be written directly to `/app/output/`.

### General Table Conventions

- All CSV files must use UTF-8 encoding, with field names in the first row.
- Numeric fields must contain numeric values only; do not include unit strings in cells.
- The field names below are canonical field names; no alternative aliases are defined. Field order is not prescribed, and additional columns that do not duplicate canonical field names are permitted.
- Except for single-row summary tables, row order is not prescribed; records in each table must be uniquely identified by the listed stable key.
- `seeded` is a Boolean value and must be written canonically as `true` or `false`.

### `/app/output/analysis.py` — Required

A UTF-8 text source file used to reconstruct the remaining deliverables based solely on materials in `/app/input/`.

### `/app/output/calibration_models.csv` — Required

Each row represents one parameter of one calibration object, with (`model`, `parameter`) as the stable key.

| Field | Type | Unit | Meaning |
|---|---|---|---|
| `model` | string | — | Calibration object identifier |
| `parameter` | string | — | Parameter identifier |
| `estimate` | number | specified by `unit` | Parameter estimate |
| `unit` | string | — | Unit of `estimate` |
| `model_rmse` | number | response-variable unit of the corresponding calibration object | Summary of calibration error |

### `/app/output/process_trajectories.csv` — Required

Each row represents one sampling time point for one batch, with (`run_id`, `time_min`) as the stable key.

| Field | Type | Unit | Meaning |
|---|---|---|---|
| `run_id` | string | — | Batch identifier |
| `profile` | string | — | Antisolvent addition profile identifier |
| `seeded` | boolean | — | Whether seeding was performed |
| `replicate` | integer | — | Independent replicate identifier |
| `time_min` | number | min | Sampling time |
| `temperature_C` | number | °C | Temperature |
| `antisolvent_volume_L` | number | L | Antisolvent volume |
| `antisolvent_fraction` | number | dimensionless | Antisolvent fraction |
| `estimated_dissolved_concentration_g_L` | number | g/L | Estimated solution concentration |
| `estimated_solubility_g_L` | number | g/L | Estimated solubility |
| `supersaturation_ratio` | number | dimensionless | Supersaturation ratio |
| `FBRM_total_counts_per_s` | number | s⁻¹ | FBRM total chord count rate |
| `FBRM_M3_um3_per_s` | number | µm³/s | FBRM third-order chord-length moment |
| `number_weighted_mean_chord_um` | number | µm | Number-weighted mean chord length |
| `volume_weighted_mean_chord_um` | number | µm | Volume-weighted mean chord length |
| `estimated_solids_g_L` | number | g/L | Estimated total solids content |
| `seed_solids_g_L` | number | g/L | Seed-crystal solids content |
| `estimated_new_solids_g_L` | number | g/L | Estimated newly formed solids content |

### `/app/output/run_metrics.csv` — Required

Each row represents a result summary for one batch, with `run_id` as the stable key.

| Field | Type | Unit | Meaning |
|---|---|---|---|
| `run_id` | string | — | Batch identifier |
| `profile` | string | — | Antisolvent addition profile identifier |
| `seeded` | boolean | — | Whether seeding was performed |
| `replicate` | integer | — | Independent replicate identifier |
| `nucleation_onset_min` | number | min | Onset time of new nucleation |
| `maximum_supersaturation_ratio` | number | dimensionless | Maximum supersaturation ratio |
| `desupersaturation_time_min` | number | min | Desupersaturation time |
| `growth_rate_um_min` | number | µm/min | Chord-length growth rate |
| `nucleation_index_percent` | number | % | Nucleation index |
| `agglomeration_index_percent` | number | % | Agglomeration index |
| `endpoint_solids_g_L` | number | g/L | Endpoint solids content |
| `endpoint_mother_liquor_g_L` | number | g/L | Endpoint mother-liquor concentration |
| `ATR_minus_HPLC_endpoint_g_L` | number | g/L | Difference between endpoint ATR and HPLC concentrations |
| `crystallization_yield_percent` | number | % | Crystallization yield |
| `mass_closure_percent` | number | % | Material closure |
| `regime_classification` | string | — | Crystallization mechanism classification |

### `/app/output/design_summary.csv` — Required

Each row represents a replicate summary for one (`profile`, `seeded`) process combination, with (`profile`, `seeded`) as the stable key.

| Field | Type | Unit | Meaning |
|---|---|---|---|
| `profile` | string | — | Antisolvent addition profile identifier |
| `seeded` | boolean | — | Whether seeding was performed |
| `n_replicates` | integer | — | Number of independent replicates |
| `nucleation_onset_mean_min` | number | min | Mean onset time of new nucleation |
| `nucleation_onset_sd_min` | number | min | SD of the onset time of new nucleation |
| `maximum_supersaturation_mean` | number | dimensionless | Mean maximum supersaturation ratio |
| `maximum_supersaturation_sd` | number | dimensionless | SD of the maximum supersaturation ratio |
| `growth_rate_mean_um_min` | number | µm/min | Mean chord-length growth rate |
| `growth_rate_sd_um_min` | number | µm/min | SD of the chord-length growth rate |
| `nucleation_index_mean_percent` | number | % | Mean nucleation index |
| `nucleation_index_sd_percent` | number | % | SD of the nucleation index |
| `agglomeration_index_mean_percent` | number | % | Mean agglomeration index |
| `agglomeration_index_sd_percent` | number | % | SD of the agglomeration index |
| `yield_mean_percent` | number | % | Mean crystallization yield |
| `yield_sd_percent` | number | % | SD of the crystallization yield |
| `mass_closure_mean_percent` | number | % | Mean material closure |
| `mass_closure_sd_percent` | number | % | SD of the material closure |
| `dominant_regime` | string | — | Dominant crystallization mechanism classification |
| `recommendation` | string | — | Process-selection result |

### `/app/output/crystallization_qc.csv` — Required

A global results summary table containing only one summary record.

| Field | Type | Unit | Meaning |
|---|---|---|---|
| `n_runs` | integer | — | Number of batches |
| `n_time_points_per_run` | integer | — | Number of sampling points per batch |
| `ATR_absorbance_RMSE_AU` | number | AU | Summary of ATR calibration error |
| `solubility_RMSE_g_L` | number | g/L | Summary of solubility calibration error |
| `FBRM_solids_RMSE_g_L` | number | g/L | Summary of FBRM solids-content calibration error |
| `maximum_ATR_HPLC_endpoint_difference_g_L` | number | g/L | Summary of batch endpoint ATR/HPLC concentration differences |
| `minimum_mass_closure_percent` | number | % | Minimum batch material closure |
| `maximum_mass_closure_percent` | number | % | Maximum batch material closure |
| `recommended_profile` | string | — | Identifier of the recommended antisolvent addition profile |
| `recommended_seeded` | boolean | — | Whether the recommended condition is seeded |
| `recommended_regime` | string | — | Crystallization mechanism classification of the recommended condition |
| `ATR_status` | string | — | ATR calibration status |
| `solubility_status` | string | — | Solubility calibration status |
| `FBRM_moment_status` | string | — | FBRM chord-length moment calibration status |
| `seed_baseline_status` | string | — | Seed-crystal baseline treatment status |
| `mass_balance_status` | string | — | Material-balance status |
| `overall_status` | string | — | Overall status |
| `failure_reasons` | string | — | Explanation of failed criteria |

### Image Files — Required

- `/app/output/supersaturation_and_solids.png`: A PNG image showing the supersaturation and solids-content results.
- `/app/output/fbrm_population_diagnostics.png`: A PNG image showing the seed-crystal baseline and FBRM chord-length population results.
- `/app/output/process_window.png`: A PNG image showing the comparison of process combinations and the process-selection result.

### `/app/output/crystallization_PAT_report.md` — Required

A UTF-8 Markdown report. The report must cover calibration results and trajectory reconstruction, the seed-crystal baseline and newly formed particles, crystallization mechanisms and comparison of process combinations, endpoint HPLC verification and material closure, process selection, and the applicability limits of FBRM chord length relative to true particle size and mass; no conclusion is presupposed. Numerical values in the report must be verifiable from the CSV files above.