Roller Compaction Batch Review

Skill workflow-skill many-tools-to-one-skill Proprietary (Paramus) — guidance only; execution performed by Paramus tools under their own license gating.
provenanceai generated statuspartial
Review a roller-compaction batch (or compare several) against its setpoints and its siblings - pull the batch record, summarize the 1 Hz process timeseries (roll force, gap, screw speed, ribbon density, torque), find excursions and setpoint-tracking deviations, and decide whether the batch passes review or needs investigation.When to use: a user names a batch id or asks how a roller-compaction run went.Multi-tool WORKFLOW with excursion and comparison judgment.
Procedure6 steps · 6 orchestrated tools
1
Discover tools via search/get_schema ("roller compaction", "batch").
2
roller_compaction_cmc_health - confirm the mounted app answers and learn the
Toolsroller_compaction_cmc_health
3
roller_compaction_cmc_list_batches (or, with a known id, straight to)
Toolsroller_compaction_cmc_list_batches
4
roller_compaction_cmc_batch_summary for the per-channel statistics and the
Toolsroller_compaction_cmc_batch_summary
5
roller_compaction_cmc_compare_batches against the product's recent batches -
Toolsroller_compaction_cmc_compare_batches
6
Verdict: pass when setpoint tracking is clean and no excursion touches a quality
You get
the batch record summary, per-channel statistics with excursion windows,

Overview

Roller compaction is continuous dry granulation: powder is pressed between two rolls into a ribbon, and the batch record is a set of setpoints plus a 1 Hz process timeseries. Reviewing a batch means answering three questions in order: what was ordered (setpoints, targets), what happened (the timeseries and its excursions), and how it compares (against sibling batches of the same product). This skill runs that review over the mounted roller-compaction CMC app and returns a pass / investigate verdict with the evidence. The value is the excursion and comparison judgment between the calls, not any single summary.

Guidance vs execution: batch records, timeseries and comparisons come from the mounted app's tools. Do not re-derive statistics by hand or import the data into ad-hoc code.

When to Use

  • A batch id is named ("review batch RC-2026-07") or a shift/day asks "how did the run go?"
  • Deviation triage: an alarm or a bad ribbon result needs the process-evidence pass
  • Campaign review: comparing several batches of one product for drift

Do not use for: formulation optimization of the compacted product (that is the formulation DoE skills), or reactor/process-kinetics questions (paramus-reactor-kinetics-screen is the reaction side). This skill's trigger is the batch record review of a roller-compaction run.

Domain Judgment

  • An excursion in a drive channel (roll force, torque, screw speed) is a process event; an excursion in a quality channel (ribbon density, gap) is a product risk - weigh them differently in the verdict.
  • Setpoint tracking matters more than absolute values: a batch that followed its setpoints cleanly but sits off the siblings' means points at the recipe, not the run.
  • One batch is a review; three drifting batches are a campaign question - escalate to the comparison view and say so.

Failure modes

  • App not mounted -> stop and name it: this skill reads a mounted app (roller-compaction-cmc); no mounted app, no review - do not invent batch data.
  • Unknown batch id -> list the batches and ask which one; never guess.
  • Timeseries channel missing for a batch -> report the gap; the verdict names which channel could not be reviewed.

Example

Asked: "Review batch RC-2026-07 - anything to investigate?" The skill confirms the app (roller_compaction_cmc_health), reads the batch record (formulation and roll-force/gap setpoints, target ribbon density), summarizes the timeseries (roller_compaction_cmc_batch_summary: per-channel mean/sd/min/max, the excursion list, setpoint tracking), and compares against the product's sibling batches (roller_compaction_cmc_compare_batches: per-channel deltas against the reference batch). The answer is the verdict plus its evidence lines - e.g. which excursion window touched which channel and how far the batch's ribbon-density mean sits from the siblings. The fixture dataset (roller-compaction-cmc: 15 batches, 1 Hz for 10 min each - roll force, roll speed, gap, screw speed, ribbon density, torque; synthetic, physically plausible, seeded) ships with the app, so the example runs as-is once the app is mounted; no numeric verdicts are quoted here - they come from the tools at run time.

Input

a batch id or product + date range; optionally the sibling set to compare

Output

the batch record summary, per-channel statistics with excursion windows,