CLI¶

The goals-scenario CLI provides two commands: draw and run.

Installation¶

Via pip (Python required)

pip install avenir_goals_scenario

After installation, goals-scenario is available on your PATH.

Windows standalone executable (no Python required)

Download goals-scenario-windows.zip from the releases page, unzip it, and run goals-scenario.exe from the extracted folder. Optionally add the folder to your PATH so the command is available globally.

Config file¶

Both commands are driven by a single JSON config file. Field names are case-insensitive (pjnz_dir, PJNZ_DIR, and Pjnz_Dir are all accepted).

{
  "pjnz_dir": "C:\\path\\to\\pjnz\\files",
  "definition_path": "C:\\path\\to\\scenario_definitions.json",
  "scenario_path": "C:\\path\\to\\draws.json",
  "output_dir": "C:\\path\\to\\output",
  "base_year": 2025,
  "output_indicators": [
    "p_hivpop",
    "p_infections",
    "p_hiv_deaths",
    "h_artpop"
  ],
  "n_simulations": 100,
  "n_workers": 4,
  "seed": null
}

Note you need to escape the \ in windows-style file paths, so use \\. Alternatively you can use unix-style / and the tool will translate them for you.

Field	Required	Description
`pjnz_dir`	Yes	Directory containing `.PJNZ` files
`output_dir`	Yes	Directory to write results to (created if absent; parent must exist)
`base_year`	Yes	First year of the output projection range
`output_indicators`	Yes	Goals output indicator names to extract
`definition_path`	No*	Path to the scenario definitions JSON file
`scenario_path`	No*	Path to a scenario draws JSON file
`n_simulations`	No	Number of draws per scenario (default: `100`)
`seed`	No	Integer RNG seed for reproducible draws (default: `null` - random)
`n_workers`	No	Parallel workers: `-1` for all CPUs, positive integer for explicit count (default: `4` or CPU count if fewer)
`scenarios_per_file`	No	Scenarios written per output Parquet file, which is also the unit of parallel work (default: `128`). Work units = `ceil(n_scenarios / scenarios_per_file) × n_pjnz`; keep that ≥ your core count or workers sit idle. A larger value gives bigger, better-compressed files but more memory (one batch in flight per worker).

* At least one of definition_path or scenario_path must be supplied for run. Both are required for draw.

Commands¶

`draw`¶

Generates scenario draws from a definition file and saves them to disk.

Both definition_path and scenario_path must be set in the config.

goals-scenario draw config.json

Argument	Description
`CONFIG_PATH`	Path to a JSON config file

`run`¶

Runs scenario analysis across a directory of PJNZ files. Behaviour depends on which of definition_path and scenario_path are set in the config:

`definition_path`	`scenario_path`	Behaviour
Set	Not set	Draws in memory, saves to `<output_dir>/draws.json`, runs
Not set	Set (file exists)	Loads draws from file, runs
Set	Set (file exists)	Uses existing draws (logs a message), runs
Set	Set (file missing)	Redraws, saves to `scenario_path`, runs

goals-scenario run config.json

Argument / Option	Description
`CONFIG_PATH`	Path to a JSON config file (positional)
`--retry PATH`	Re-run only the units listed in a `failures.json` from a previous run (see Fault tolerance & re-runs)

Fault tolerance and re-runs¶

A run is made up of one work unit per (PJNZ, scenario) combination. If an individual unit fails (for example a single scenario errors for one country), it is logged and skipped — the run continues and every other unit still completes and is written.

PJNZ import failures are the exception: a .PJNZ file that cannot be read is treated as a fatal data error and aborts the whole run (exit code 1).

When any unit fails, a summary is printed and a failures.json manifest is written to output_dir (see Failures manifest JSON). The command exits with code 2 to signal a partial run. Re-run just the failed units by pointing --retry at that manifest:

goals-scenario run config.json --retry path/to/output/failures.json

Retries reuse the same draws, so the failed units reproduce exactly. A retry writes the recovered scenarios to new supplemental part-retry-*.parquet files alongside the originals (closed Parquet files are immutable, so nothing is rewritten); because the dataset is read as the union of all part-*.parquet files and failed scenarios were never written to the originals, there are no duplicate rows. On a fully successful (re-)run the stale failures.json is removed.

Exit code	Meaning
`0`	Success — every unit completed
`1`	Fatal error — invalid config, PJNZ import failure, or no PJNZ files found
`2`	Partial — some scenario units failed; output produced and `failures.json` written

Typical workflows¶

One-shot - draw and run in a single command, no intermediate file:

{
  "pjnz_dir": "path/to/pjnz",
  "definition_path": "scenario_definitions.json",
  "output_dir": "path/to/output",
  "base_year": 2025,
  "output_indicators": ["p_hivpop", "p_infections"]
}

goals-scenario run config.json

Two-step - generate and inspect draws first, then run:

goals-scenario draw config.json   # writes draws to scenario_path
goals-scenario run config.json    # reuses the same draws

Performance tuning¶

The run parallelises over (PJNZ, batch) work units. Each batch is a group of scenarios_per_file scenarios, written as a single Parquet file per indicator in one write_table call (pyarrow sizes the row groups for good compression). Two settings control throughput.

n_workers - parallel worker processes. Use -1 to use every core.
scenarios_per_file - scenarios per output file, and the unit of parallel work. The total number of units is ceil(n_scenarios / scenarios_per_file) × n_pjnz, and that is what bounds core usage. Set it so there are at least as many units as cores, otherwise cores sit idle (the run warns when this happens). It is the single lever that trades off three things at once:
- Parallelism - smaller value → more units.
- File size / count - larger value → bigger, fewer, better-compressed files (output file count is ceil(n_scenarios / scenarios_per_file) × n_pjnz × n_indicators).
- Memory - one batch is held in flight per worker, so peak memory is roughly n_workers × scenarios_per_file × n_simulations × 3.5 MB (uncompressed Arrow). Unlike the rest of the run, this does scale with the setting, so keep it low enough to fit node RAM at high worker or simulation counts.

Example - 4 PJNZ files, 4096 scenarios each:

Environment	`n_workers`	`scenarios_per_file`	Units	Peak memory (5 sims)
6-core laptop	`-1`	`2731` (→ 6 units)	6	~0.3 GB
32-core node	`-1`	`512` (→ 32 units)	32	~29 GB

Output size. With the eight default indicators at 5 simulations per scenario, expect roughly 2 MB of compressed Parquet per scenario per PJNZ, so a full run is about n_scenarios × n_pjnz × 2 MB (e.g. 4096 × 4 ≈ ~32 GB). About 85% of that is the dense h_artpop array, so its file is the large one — roughly 1.7 MB × scenarios_per_file (~215 MB at the default 128); every other indicator's file is much smaller. Output scales linearly with the number of simulations and with the number and size of output_indicators (adding indicators adds their bytes, dominated by the largest arrays).

Why it matters: writing one Parquet file per (PJNZ, scenario) produces tens or hundreds of thousands of tiny objects, and each object create on object storage (S3/ADLS/DBFS) is a rate-limited, replicated, network-committed transaction. Batching scenarios into a handful of larger files replaces that with a handful of uploads, which is the single biggest lever on wall-clock time for large runs.

File formats¶

Scenario definition JSON¶

The scenario definition file specifies the set of scenarios to analyse. Each scenario is either a single set of interventions, or a combination of other single scenarios. The file is consumed by the draw command, which samples parameter distributions and writes the resulting draws to the scenario draws JSON.

{
  "scenarios": [
    {
      "id": "1",
      "pjnz_names": ["Zimbabwe"],
      "interventions": [ ... ]
    },
    {
      "id": "2",
      "interventions": [ ... ]
    },
    {
      "id": "3",
      "combines": ["1", "2"]
    }
  ]
}

Field	Description
`id`	Unique string identifier for this scenario
`pjnz_names`	Optional list of PJNZ file names (without `.PJNZ`) to restrict this scenario to
`interventions`	List of typed intervention objects (single scenarios only)
`combines`	List of two or more single scenario IDs to merge (combined scenarios only)

Intervention types¶

Each intervention is discriminated by its product field. The valid products and their required fields are listed below.

Every intervention has a parameters object. Products that target specific populations (PrEP/PEP, Vaccine, Cure, Cure (neonates), Vaginal microbiome modification, Adult ART) also have a targets list, and for those products target_coverage lives inside each target, not in parameters — each target gets its own coverage. The target-less products (AHD treatment, POC tests, Long-acting treatment) have no targets list, so target_coverage lives directly in parameters instead, since there is only one population to specify coverage for.

Per-year coverage arrays¶

Anywhere a target_coverage distribution is accepted — whether inside a target or in parameters — you may instead supply an explicit per-year array of values, e.g. "target_coverage": [0.80, 0.81, 0.82, ...]. Each element is a proportion in 0–1.

The array holds one value per year from base_year (from the config) to the projection's final year inclusive, so its length must be projection_end_year - base_year + 1. The projection end year comes from each PJNZ file, so the length is checked when the run starts (right after the PJNZ files are imported). A wrong length aborts the whole run with a message naming the scenario, intervention, PJNZ, and the expected length.
Array values are passed straight through — the draw command does no sampling for them, so every simulation carries the same trajectory. The values are written directly into the model's yearly coverage array, bypassing the linear base-year→target_year ramp used for distributions.
target_year is only used to ramp distribution coverages. When every coverage in an intervention is an array, target_year is not needed and may be omitted; if supplied it is ignored. When an intervention mixes array and distribution coverages across its targets, target_year is still required and applies only to the distribution targets.

{
  "product": "Adult ART",
  "targets": [
    {"sex": "Female", "target_coverage": [0.80, 0.82, 0.84, 0.85, 0.85]},
    {"sex": "Male",   "target_coverage": [0.78, 0.80, 0.82, 0.83, 0.83]}
  ],
  "parameters": {}
}

PrEP interventions¶

Valid product values: "Oral PrEP (daily)", "Oral PrEP (monthly)", "Injectable PrEP (1 month)", "Injectable PrEP (2 month)", "Injectable PrEP (6 month)", "Oral PrEP plus contraceptive", "PrEP ring", "Implantable PrEP", "bNABs", "PEP"

targets: one or more risk group/sex combinations, each with its own target coverage.

Field	Values
`risk_group`	`"Low risk heterosexual"`, `"Medium risk heterosexual"`, `"High risk heterosexual"`, `"People who inject drugs"`, `"Men who have sex with men"`
`sex`	`"Male"`, `"Female"`, `"Both"`
`target_coverage`	Distribution parameters (`mean` & `sd`) or an array of per-year coverages for this target — see per-year coverage arrays

"Men who have sex with men" cannot have sex: "Female".

parameters:

Parameter	Description
`efficacy`	Distribution for intervention efficacy (proportion, 0–1)
`adherence`	Distribution for adherence (proportion, 0–1)
`target_year`	Distribution for target implementation year (integer ≥ 1970). Ignored if every target's `target_coverage` is passed as an array — see per-year coverage arrays
`substitution`	Distribution for substitution (proportion, 0–1). Only valid for `"Oral PrEP plus contraceptive"`.
`duration`	Distribution for implant duration in months (≥ 0). Only valid for `"Implantable PrEP"`.

Each distribution is {"mean": <float>, "sd": <float>} with optional min_value and max_value overrides.

substitution and duration are optional and product-specific: setting substitution on any product other than "Oral PrEP plus contraceptive", or duration on anything other than "Implantable PrEP", is a validation error. When omitted, the model's PJNZ default is used.

{
  "product": "Oral PrEP (daily)",
  "targets": [
    {"risk_group": "High risk heterosexual", "sex": "Female", "target_coverage": {"mean": 0.30, "sd": 0.05}},
    {"risk_group": "Men who have sex with men", "sex": "Male", "target_coverage": {"mean": 0.30, "sd": 0.05}}
  ],
  "parameters": {
    "efficacy":    {"mean": 0.95, "sd": 0.03},
    "adherence":   {"mean": 0.85, "sd": 0.05},
    "target_year": {"mean": 2028, "sd": 2}
  }
}

Implantable PrEP with a duration (months), and Oral PrEP plus contraceptive with a substitution:

{
  "product": "Implantable PrEP",
  "targets": [
    {"risk_group": "High risk heterosexual", "sex": "Female", "target_coverage": {"mean": 0.15, "sd": 0.05}}
  ],
  "parameters": {
    "efficacy":    {"mean": 0.90, "sd": 0.03},
    "adherence":   {"mean": 0.85, "sd": 0.05},
    "target_year": {"mean": 2028, "sd": 2},
    "duration":    {"mean": 12, "sd": 1}
  }
}

Vaccine¶

product: "Vaccine"

targets: one or more entries (one or more required), each with its own target coverage. Two targeting modes:

PLHIV — applies coverage across all PLHIV regardless of risk group. Use risk_group: "PLHIV" with sex: "Both" or omit sex.
Risk group — targets a specific risk group. "Both" applies coverage to both male and female indices for that group.

Field	Values
`risk_group`	`"Low risk heterosexual"`, `"Medium risk heterosexual"`, `"High risk heterosexual"`, `"People who inject drugs"`, `"Men who have sex with men"`, `"PLHIV"`
`sex`	`"Male"`, `"Female"`, `"Both"`. Optional — omitting it behaves the same as `"Both"`. For `risk_group: "PLHIV"`, only `"Both"` (or omitting `sex`) is allowed; `"Male"`/`"Female"` are rejected there since PLHIV coverage applies regardless of sex
`target_coverage`	Distribution parameters (`mean` & `sd`) or an array of per-year coverages for this target — see per-year coverage arrays

"Men who have sex with men" cannot have sex: "Female". "PLHIV" cannot have sex: "Male" or "Female".

parameters:

Parameter	Type	Description
`target_year`	Distribution	Target implementation year. Ignored if every target's `target_coverage` is passed as an array — see per-year coverage arrays
`reduction_in_susceptibility`	Distribution	Reduction in susceptibility to HIV due to vaccination (0–1)
`reduction_in_infectiousness`	Distribution	Reduction in infectiousness due to vaccination (0–1)
`increase_in_progression_time_to_aids`	Distribution	Increase in progression time to AIDS (0–1)
`vaccine_duration_years`	Distribution	Vaccine duration in years
`vaccine_action_type`	`"Take"` or `"Degree"`	Type of vaccine action on susceptibility
`targeting`	`"Vaccinate without HIV testing"` or `"Vaccinate only HIV-negative individuals"`	Vaccination targeting strategy

{
  "product": "Vaccine",
  "targets": [
    {"risk_group": "PLHIV", "target_coverage": {"mean": 0.50, "sd": 0.10}}
  ],
  "parameters": {
    "target_year":                           {"mean": 2035, "sd": 3},
    "reduction_in_susceptibility":           {"mean": 0.60, "sd": 0.05},
    "reduction_in_infectiousness":           {"mean": 0.40, "sd": 0.05},
    "increase_in_progression_time_to_aids":  {"mean": 0.20, "sd": 0.02},
    "vaccine_duration_years":                {"mean": 5,    "sd": 1},
    "vaccine_action_type":                   "Take",
    "targeting":                             "Vaccinate only HIV-negative individuals"
  }
}

Cure¶

product: "Cure (adults and children)"

targets: one or more entries (one or more required), each with its own target coverage. Same two targeting modes as Vaccine:

PLHIV — risk_group: "PLHIV" with sex: "Both" or omit sex.
Risk group — targets a specific risk group. "Both" applies coverage to both male and female indices.

Field	Values
`risk_group`	`"Low risk heterosexual"`, `"Medium risk heterosexual"`, `"High risk heterosexual"`, `"People who inject drugs"`, `"Men who have sex with men"`, `"PLHIV"`
`sex`	`"Male"`, `"Female"`, `"Both"`. Optional — omitting it behaves the same as `"Both"`. For `risk_group: "PLHIV"`, only `"Both"` (or omitting `sex`) is allowed; `"Male"`/`"Female"` are rejected there since PLHIV coverage applies regardless of sex
`target_coverage`	Distribution parameters (`mean` & `sd`) or an array of per-year coverages for this target — see per-year coverage arrays

"Men who have sex with men" cannot have sex: "Female". "PLHIV" cannot have sex: "Male" or "Female".

parameters:

Parameter	Description
`target_year`	Target implementation year. Ignored if every target's `target_coverage` is passed as an array — see per-year coverage arrays
`efficacy`	Efficacy of the cure (0–1)
`duration_of_cure`	Duration of cure effect

{
  "product": "Cure (adults and children)",
  "targets": [
    {"risk_group": "PLHIV", "target_coverage": {"mean": 0.20, "sd": 0.05}}
  ],
  "parameters": {
    "target_year":      {"mean": 2035, "sd": 3},
    "efficacy":         {"mean": 0.85, "sd": 0.05},
    "duration_of_cure": {"mean": 0.50, "sd": 0.10}
  }
}

Cure (neonates)¶

product: "Cure (neonates)"

targets: neonates are a single population with no risk-group or sex split, so this is normally a single entry.

Field	Values
`risk_group`	`"Neonates"`
`target_coverage`	Distribution parameters (`mean` & `sd`) or an array of per-year coverages — see per-year coverage arrays

parameters:

Parameter	Description
`target_year`	Target implementation year. Ignored if `target_coverage` is passed as an array — see per-year coverage arrays
`effectiveness`	Effectiveness of the cure (0–1)

{
  "product": "Cure (neonates)",
  "targets": [
    {"risk_group": "Neonates", "target_coverage": {"mean": 0.40, "sd": 0.05}}
  ],
  "parameters": {
    "target_year":   {"mean": 2032, "sd": 2},
    "effectiveness": {"mean": 0.65, "sd": 0.05}
  }
}

Vaginal microbiome modification¶

product: "Vaginal microbiome modification"

targets: one or more entries, each with its own target coverage. Women only (there is no sex field).

Field	Values
`risk_group`	`"Percent of women treated"`, `"Not sexually active"`, `"Low risk heterosexual"`, `"Medium risk heterosexual"`, `"High risk heterosexual"`
`target_coverage`	Distribution parameters (`mean` & `sd`) or an array of per-year coverages for this target — see per-year coverage arrays

"Percent of women treated" applies coverage to all women and, when used, must be the only target on the intervention.

parameters:

Parameter	Description
`target_year`	Target implementation year. Ignored if every target's `target_coverage` is passed as an array — see per-year coverage arrays
`effectiveness`	Effectiveness of the intervention (0–1)

{
  "product": "Vaginal microbiome modification",
  "targets": [
    {"risk_group": "Percent of women treated", "target_coverage": {"mean": 0.30, "sd": 0.05}}
  ],
  "parameters": {
    "target_year":   {"mean": 2032, "sd": 2},
    "effectiveness": {"mean": 0.40, "sd": 0.05}
  }
}

AHD treatment¶

product: "AHD treatment"

No targets field — coverage applies globally.

parameters:

Parameter	Description
`target_year`	Target implementation year. Ignored if `target_coverage` is passed as an array — see per-year coverage arrays
`target_coverage`	Distribution parameters (`mean` & `sd`) or an array of per-year coverages — see per-year coverage arrays
`reduction_in_mortality`	Reduction in AHD mortality (0–1)

{
  "product": "AHD treatment",
  "parameters": {
    "target_year":          {"mean": 2026, "sd": 1},
    "target_coverage":      {"mean": 0.70, "sd": 0.08},
    "reduction_in_mortality":{"mean": 0.40, "sd": 0.05}
  }
}

POC VL test¶

product: "POC VL test"

No targets field.

parameters:

Parameter	Description
`target_year`	Target implementation year. Ignored if `target_coverage` is passed as an array — see per-year coverage arrays
`target_coverage`	Distribution parameters (`mean` & `sd`) or an array of per-year coverages — see per-year coverage arrays
`effect`	Effect size (0–1)

{
  "product": "POC VL test",
  "parameters": {
    "target_year":     {"mean": 2027, "sd": 1},
    "target_coverage": {"mean": 0.70, "sd": 0.08},
    "effect":          {"mean": 0.50, "sd": 0.05}
  }
}

POC CD4 test¶

product: "POC CD4 test"

Same structure as POC VL test.

{
  "product": "POC CD4 test",
  "parameters": {
    "target_year":     {"mean": 2027, "sd": 1},
    "target_coverage": {"mean": 0.70, "sd": 0.08},
    "effect":          {"mean": 0.50, "sd": 0.05}
  }
}

Adult ART¶

product: "Adult ART"

targets: one or more entries, each specifying a sex and its target coverage.

Field	Values
`sex`	`"Male"`, `"Female"`, `"Both"`
`target_coverage`	Distribution parameters (`mean` & `sd`) or an array of per-year coverages for this target — see per-year coverage arrays

parameters:

Parameter	Description
`target_year`	Target implementation year (integer ≥ 1970). Ignored if every target's `target_coverage` is passed as an array — see per-year coverage arrays

Coverage ramps linearly from its base-year value to target_coverage at target_year and is held thereafter. Coverage values are interpreted as proportions (ratio between 0 and 1); art15plus_isperc is automatically set to 1 for the targeted sex from the base year onward.

{
  "product": "Adult ART",
  "targets": [
    {"sex": "Female", "target_coverage": {"mean": 0.85, "sd": 0.05}},
    {"sex": "Male",   "target_coverage": {"mean": 0.85, "sd": 0.05}}
  ],
  "parameters": {
    "target_year": {"mean": 2028, "sd": 2}
  }
}

Long-acting treatment¶

Valid product values: "Long-acting treatment", "Long-acting treatment (Oral weekly)", "Long-acting treatment (Injectable 6 month)", "Long-acting treatment (Implant)".

No targets field — coverage applies globally.

parameters:

Parameter	Description
`target_year`	Target implementation year. Ignored if `target_coverage` is passed as an array — see per-year coverage arrays
`target_coverage`	Distribution parameters (`mean` & `sd`) or an array of per-year coverages — see per-year coverage arrays
`interruption_rate_reduction`	Reduction in treatment interruption rate (0–1)
`viral_load_suppression_ratio`	Viral load suppression ratio (0–1)

A scenario may include more than one of the 4 products above (e.g. both the oral weekly and the injectable variant) — they are combined before being passed to Goals, since the model only has a single coverage slot for long-acting treatment: each product's target_coverage is ramped independently, then summed per year (clamped to 1.0). interruption_rate_reduction and viral_load_suppression_ratio are each blended into a single value, weighted by every included product's own target_year coverage — Goals has no way to vary either rate by year, so this blend is necessarily a single number for the whole run, not a per-year series.

{
  "product": "Long-acting treatment (Oral weekly)",
  "parameters": {
    "target_year":                  {"mean": 2030, "sd": 2},
    "target_coverage":               {"mean": 0.30, "sd": 0.05},
    "interruption_rate_reduction":   {"mean": 0.20, "sd": 0.05},
    "viral_load_suppression_ratio":  {"mean": 0.75, "sd": 0.05}
  }
}

Scenario draws JSON¶

The draws file produced by draw (or saved automatically by run) has this structure:

{
  "scenarios": [
    {
      "id": "1",
      "interventions": [
        {
          "id": "oral_prep_daily",
          "product": "Oral PrEP (daily)",
          "targets": [
            { "risk_group": "High risk heterosexual", "sex": "Female" },
            { "risk_group": "Men who have sex with men", "sex": "Male" }
          ]
        }
      ],
      "simulations": [
        {
          "oral_prep_daily": {
            "efficacy": 0.976158,
            "adherence": 0.942526,
            "target_coverage": 0.202123,
            "target_year": 2028
          }
        }
      ]
    }
  ]
}

Each entry in simulations maps intervention slug → sampled parameter values for one draw. Categorical parameters (e.g. vaccine_action_type) are passed through unchanged. A coverage supplied as a per-year array appears verbatim in place of the sampled scalar (as the coverage value, or as target_coverage for target-less products), identical across every simulation.

Failures manifest JSON¶

Written to <output_dir>/failures.json whenever one or more (PJNZ, scenario) units fail. Pass it back to run --retry to re-run only those units.

{
  "failures": [
    { "pjnz": "Zimbabwe", "scenario_id": "3", "error": "..." }
  ]
}

Field	Description
`pjnz`	Stem of the PJNZ file whose scenario failed (no `.PJNZ`)
`scenario_id`	Identifier of the scenario that failed for that PJNZ
`error`	Short error message describing the failure

Output data (Parquet)¶

Results are written as a Hive-partitioned Parquet dataset, one directory per indicator:

{output_dir}/{indicator}/pjnz_name={pjnz}/part-{batch}.parquet

pjnz_name is a partition directory. scenario_id is a data column (not a partition), so a partition may contain several part-*.parquet files (one per batch, plus part-retry-*.parquet from any --retry run). Readers should treat a partition as the union of all its part-*.parquet files - arrow::open_dataset(<indicator dir>) and read_parquet(..., hive_partitioning = true) do this automatically.
Columns are scenario_id, the indicator's dimension columns, simulation (int32), and value (float64). Row groups are sized by pyarrow within each batch file, and their scenario_id statistics give predicate pushdown via the _metadata file.
scenario_id is stored as a string ("1", "all_products", ...), because scenario identifiers are not always numeric. Filter with scenario_id == "1", not scenario_id == 1.

Global options¶

Option	Description
`--version`	Show version and exit
`--help`, `-h`	Show help and exit
`-v`, `--verbose`	Enable debug logging

Tab completion¶

goals-scenario --install-completion

CLI¶

Installation¶

Config file¶

Commands¶

draw¶

run¶

Fault tolerance and re-runs¶

Typical workflows¶

Performance tuning¶

File formats¶

Scenario definition JSON¶

Intervention types¶

Per-year coverage arrays¶

PrEP interventions¶

Vaccine¶

Cure¶

Cure (neonates)¶

Vaginal microbiome modification¶

AHD treatment¶

POC VL test¶

POC CD4 test¶

Adult ART¶

Long-acting treatment¶

Scenario draws JSON¶

Failures manifest JSON¶

Output data (Parquet)¶

Global options¶

Tab completion¶

`draw`¶

`run`¶