Project continuation — refresh current priorities after evidence convergence
Date: 2026-08-28
Repository: MSKazemi/idkmesh
Owner request
Continue improving the public repository.
Repository state inspected
The live repository had advanced far beyond the older docs/planning/CURRENT_PRIORITIES.md snapshot.
Observed facts at this continuation point:
main at inspection: af77440ba31aa9b53035818db02b09b9286401c7;- public GitHub metadata still reported
main as unprotected and required-status enforcement off;
- issue #35 remained the external/admin P0 protection gate;
- PR #91 was the only open pull request and remained intentionally draft pending genuinely separate human review of exact-head runtime evidence;
- issue #4 already recorded real success/success and real success/failure two-attempt evidence with exact replay;
- issue #5 already had the real verifier/report/replay substrate, and the first five Phase B2 benchmark definitions had subsequently been frozen on
main through merged #134;
- issue #20 IDKGraph P0 was closed completed;
- the read-only mathematical repository-evolution observatory had advanced through merged #143/#144/#148;
- ACE Bootstrap Cohort Observatory #109 reported capacity approximately
0.915 but zero claims, zero candidate PRs, zero verified descendants, and zero distinct external participants, with recommendation HOLD_COHORT_1.
Diagnosis
The repository is no longer mainly bottlenecked by missing mechanisms.
The limiting state is now:
unprotected integration boundary
+
canonical worker awaiting separate human review
+
insufficient held-out real-task corpus
+
zero external ACE descendants
This means the correct project posture is convergence + external evidence, not another controller, protocol, queue, or large theory layer.
Bounded action
Refresh docs/planning/CURRENT_PRIORITIES.md so it stops directing contributors toward already-completed or superseded work.
The refreshed ordering is:
- protect
main in GitHub settings (#35);
- obtain separate human review of exact #91 evidence;
- put the real node behind the existing adapter boundary;
- add one trivial heterogeneous real adapter;
- execute/replay the first five frozen benchmark tasks;
- grow the held-out real corpus for #70/#30;
- measure diversity vs replication under fixed budgets;
- only then feed real measurements into scheduler/evolution policy.
In parallel, ACE remains HOLD_COHORT_1: recovered review capacity is not reproduction evidence.
Preserved invariants
- proposal is not proof;
- CI success is not independent approval;
- worker success is not acceptance;
- observatory scores are decision support, not merge authority;
- no autonomous actor should propose, approve, and merge the same protected change;
- real negative/inconclusive evidence must be retained;
- new complexity should not outrun verification and reviewer capacity.
A current priority map reduces contributor misdirection and duplicate implementation. Newcomers and agents should now see that the highest-value work is no longer inventing mechanisms, but crossing the remaining protection, witness, benchmark, and external-community evidence gates.