Discovery yield trajectory: 53ยท18ยท24ยท13ยท3ยท2ยท0ยท7 โ iter-7 was the first dry iteration; iter-8 rebounded (+7) so the loop continued (loop-until-dry K=2 never tripped). Loop manually stopped at the operator's request.
Workflow runs (multi-agent orchestration)
Run ID
Phase
Sub-agents
Output tokens
Wall-clock
wf_164f517c
iter-1 โ broad SOTA sweep
18
โ
โ
wf_16995246
iter-2 โ missing-angle sweep
6
โ
โ
wf_a44fe2cb
iter-3 โ deep cluster sweep
5
โ
โ
wf_08c9c790
iter-4 โ remaining-cluster sweep
5
โ
โ
wf_5edba8cc
iter-5 โ niche taper-test
4
โ
โ
wf_a15e5b3a
iter-6 โ final-dredge
3
โ
โ
wf_0362d229
iter-7 โ dry-test sweep
2
โ
โ
wf_e916e0e7
iter-8 โ decider (rebound)
3
193,635
353s
wf_59e198ff
verify โ chronicle cross-check
5
2,649,861
298s
Token/wall-clock captured from run notifications for the two most recent workflows; earlier runs show sub-agent counts from their transcript directories.
Publish provenance
Built in local scratch โ rsync -az --delete โ bigblack:~/sites/2026-06-03-forex-parameterless-orthogonal-candidate-discovery/ โ build-nav.py --root ~/sites (pagefind + auto-nav) โ served by Caddy (:8087) โ https://odb.epatner.com/2026-06-03-forex-parameterless-orthogonal-candidate-discovery/ (behind CF Access). Read-only on bigblack ClickHouse throughout; audit markdown lands via PR-from-worktree (never direct to main).
Audit artifacts (in mql5/findings/audits/2026-06-03-โฆ/)