Date: 2026-08-28
Repository: MSKazemi/idkmesh
The project owner asked to continue development, inspect the repository for bugs/inconsistencies, choose the most useful next step, execute it, and keep the work public in the repository.
Issue #82 still described a local_verifier.py self-test default-path bug. Current main already used isolated candidate-root directories, so #82 was documented and closed as completed rather than left as a misleading open defect.
Issues #4 and #16 contained historical worker SHA/checklist state. Corrective comments recorded the active exact worker candidate and were then corrected again as concurrent merges #117 and #116 moved the repository baseline forward.
This matters in a self-evolving repository: tracker state must follow actual current evidence rather than preserve an obsolete roadmap as if it were pending work.
The repository had a real evidence-retention inconsistency:
The older diverged PRs #115 and #119 were retired as superseded. PR #122 unified both paths using pinned actions/upload-artifact v4.6.2, 30-day retention, if-no-files-found: error, read-only repository permissions, and no selection/approval/merge authority.
The unified branch was explicitly synchronized with the merged #116 EvaluatorPlan-v0.2 orchestrator state and re-tested. Both real E2E paths succeeded and produced retained artifacts before #122 was squash-merged.
PR #122 merged as:
2c312e124743dc677333fbf3b08477a619d615ad
The exact compatibility runs before merge were:
33187645308, success, artifact 9692375594, digest sha256:42af1a108a2bb73d1c36fa7ad4319e01442e968e45338818dacbec3c1575c967;33187645373, success, artifact 9692376706, digest sha256:e68d6d4f9a8cdc2b2200d43f44ba5f9aa2c263c5dddff31838a8e69597994233.PR #125 introduced the useful Free Resource Mesh v0:
schemas/resource-offer-registry-v0.1.schema.json;scripts/free_resource_planner.py;The repository already had a separate zero-cost runtime layer:
schemas/compute-offer-pool-v0.1.schema.json;config/compute-policy.json;experiments/free_compute_router.py;experiments/local_compute_offer.py.These contracts are complementary rather than duplicates:
The original planner JSON did not make that distinction machine-readable. A future integration could therefore treat a planner selected entry as an executable compute offer and accidentally evolve a second execution router.
While this fix was being tested, main independently gained:
docs/architecture/FREE_RESOURCE_MESH_COMPUTE_BRIDGE.md
That document explicitly establishes the canonical separation and states that a registry entry is not automatically a compute offer.
This was semantic overlap with an architecture note initially drafted on this branch. Rather than merge two normative documents that say essentially the same thing, the duplicate branch document was deleted.
This is a useful self-evolution rule:
independent work discovers same boundary
-> prefer the already-canonical artifact
-> remove duplicate architecture
-> retain only missing executable enforcement
The remaining useful delta is machine-readable and regression-tested.
scripts/free_resource_planner.py emits:
runtime_materialization.required_before_execution = true
runtime_materialization.planner_output_is_executable_compute_offer = false
runtime_materialization.boundary_doc = docs/architecture/FREE_RESOURCE_MESH_COMPUTE_BRIDGE.md
runtime_materialization.discovery_contract = schemas/resource-offer-registry-v0.1.schema.json
runtime_materialization.runtime_contract = schemas/compute-offer-pool-v0.1.schema.json
runtime_materialization.runtime_router = experiments/free_compute_router.py
runtime_materialization.repository_compute_policy = config/compute-policy.json
tests/test_free_resource_planner.py locks all of those values.
The intended pipeline remains:
resource catalog / freshness / policy evidence
-> discovery planner
-> provider-specific live materializer
-> canonical Compute Offer Pool
-> repository compute policy
-> canonical free compute router
-> bounded worker
-> independent verification
-> explicit integration decision
The key safety statement is:
discovery eligibility != execution eligibility
A provider must supply live capacity/availability evidence and any required operator consent/caps before it can materialize into the existing runtime offer contract.
Before the concurrent bridge landed, the initial machine-readable boundary passed the Free Resource Mesh Plan workflow at exact head 011160fc8327b06ae8519f817afa7ce996242fdc, including:
After converging on the canonical bridge document and synchronizing current main, the same policy workflow should be treated as the required final merge evidence for the reduced patch.
Public GitHub branch metadata continued during this audit to report main as unprotected with required status enforcement off.
Repository workflows can detect and document that state, but they cannot substitute for an actual GitHub ruleset/branch-protection configuration. The canonical worker PR #91 also retains its deliberate genuinely separate human/reviewer inspection gate; this audit did not manufacture independence by self-approving that worker.
The real same-worker two-attempt path and EvaluatorPlan-v0.2 orchestration are already integrated. The most informative next product experiment is not another orchestration layer. It is one deliberately simple heterogeneous second real adapter/worker plus a small frozen real-task set, measuring:
That will test whether diversity actually improves IDKMesh rather than merely increasing agent count.