Flatfile Agential Resource System Improving

What this is

FlatfileAgentialResourceSystemImproving is a specialization of FlatfileAgentialResourceSystemWorking (flatfile-agential-resource-system-working) that targets SkillfulWorkUnits (flatfile-agential-resource-system-skillful-work) specifically. FlatfileAgentialResourceSystemImproving MUST only target SkillfulWorkUnits — ad-hoc work goes through FlatfileAgentialResourceSystemWorking. It adds three things that plain FlatfileAgentialResourceSystemWorking lacks:

1. Selection bias — among available work units, FlatfileAgentialResourceSystemImproving prioritizes those that enable more work: unblocking dependencies, repairing broken skills, filling missing specs. A work unit that enables ten more units is preferred over one that enables none.

2. Feedback cycles — FlatfileAgentialResourceSystemImproving always ends with a learning capture step (skills/post-work-update.md). Discoveries made during work are routed back into PLANS and IDEAS. Nothing is discarded.

3. CoursOfWork improvement — for Derivation and Inquiry work units, FlatfileAgentialResourceSystemImproving triggers a tail-end improvement pass over the skills and units accessed during the work. See CoursOfWork below.

Selection bias rule

When choosing among open work items, FlatfileAgentialResourceSystemImproving MUST rank by:

1. Unblocking — items whose completion directly enables one or more currently blocked items to become available
2. Repair — items that fix broken or malformed SkillfulWorkUnits
3. Gap-filling — items that create new SkillfulWorkUnits where none exist
4. Ordinary work — items with no downstream enablement

Within each rank, tie-break by whichever item appears earliest in PLANS.

Plain FlatfileAgentialResourceSystemWorking takes item 1 from the queue regardless of rank. FlatfileAgentialResourceSystemImproving re-sorts the queue before taking item 1.

CoursOfWork

The CoursOfWork of a Derivation or Inquiry execution is the set of:

• Skills invoked (directly or as sub-skills)
• Spec files, math entitys, or other files read as context
• Skills or files that produced unexpected outputs or errors

At the end of any Derivation or Inquiry, FlatfileAgentialResourceSystemImproving MUST collect the CoursOfWork and pass it to skills/post-work-update.md as part of findings. Each item in the CoursOfWork that showed a problem triggers a feedback entry. Each item that worked perfectly is noted but does not generate work.

This is not optional. A Derivation or Inquiry that produces a judgment without examining what it used to get there has not closed its loop.

Error routing

When a Process or Procedure called during FlatfileAgentialResourceSystemImproving emits a SkillError (flatfile-agential-resource-system-skill-result), FlatfileAgentialResourceSystemImproving MUST:

1. Attempt local recovery if the error class is MissingInput or ValidationFailure and the recovery is straightforward.
2. If not locally recoverable: pass the error to skills/post-work-update.md as a finding — it becomes a PLANS item in the owning locale.
3. Mark the failed work unit as error in PLANS (not ✓) with the SkillError attached — so the next FlatfileAgentialResourceSystemImproving pass can target it.

An error that reaches FlatfileAgentialResourceSystemImproving and is not routed is a bug in FlatfileAgentialResourceSystemImproving.

Open questions

• Formal proof that the selection bias rule produces optimal ordering — blocked by: skill-algebra construction
• Formal definition of CoursOfWork as a typed object — blocked by: skill-algebra construction