agent-optimize's sibling fan-out, done behind the\nhonest val significance gate that evograph never applied, and everything else it did (failure\nclustering, rejected-edit memory, budget-aware fan-out, free-text stop condition) lives in\nagent-optimize + phases/diagnose. Use when reading or repairing an existing evograph run dir,\nor when writing the run-dir wiki/ format the dashboard's Weakness-graph tab reads — and to see\nwhat to select instead: agent-optimize for agent-mode search, hill-climb, gepa, or\nskillopt for a deterministic loop.\ncomponent: algorithm\nargument-hint: \"deprecated — use agent-optimize (agent mode) or hill-climb | gepa | skillopt\"\nallowed-tools: Read, Write, Edit, Bash, Task\nneeds: [scores, traces, candidate]\nprovides: [candidate]\nsources: [evo-graph]Do not select algorithm_skill: evograph for a new run. Use agent-optimize (agent mode) or\nhill-climb / gepa / skillopt (deterministic). This file stays so an existing evograph run dir\nis still readable and so the wiki file-format contract has an owner until it moves.
evograph advertised one distinctive capability — a collaborative weakness graph with one solver\nagent per weakness, merged into a shared candidate each round. Measured against its four siblings,\nthat capability is not distinctive and the part that was distinctive was a defect:
\nagent-optimize fans out N sibling candidates\nfrom the same parent, one diagnosed failure cluster each, every sibling in its own working copy\n(a git worktree when the capability is in git), gated one at a time with a re-gate after each\naccept so several fixes accumulate into one lineage honestly. That is evograph's round, minus\nthe flaws below. The clustering itself is phases/diagnose's job in both cases.Δ > k·SE. Between baseline and finalize an evograph run took no\nheld-out measurement at all, so the sealed test number was the first honest signal anyone saw.\nWhole-round revert existed only as a one-round-late substitute for the gate it lacked; wire the\nreal gate and there is nothing left for it to catch.wiki/ is\ngenuinely useful, but the dashboard renders the Weakness-graph tab from wiki/ presence alone,\nfor any algorithm that writes the format (core/cap_evolve/dashboard.py). An output contract\ndoes not earn a second agent-mode algorithm that users must choose between.There never was one, and scripts/run.py is a tombstone, not a stub: invoked deterministically it\nexits 2 with an agent-mode only payload rather than faking a loop. So evograph is also the one\nalgorithm that could not be routed through the shared per-iteration record — see\nthe hill-climb skill's references/run-step.md, which owns the shared iteration mechanics\n(parent selection, val gate, commit, iteration record) every other algorithm routes through. Read\nit if you are reconstructing what an evograph round should have done.
The run dir is authoritative. <run_dir>/wiki/ holds the weakness nodes, solution cards and\nper-round results; <run_dir>/runs/round-<N>/agents/<slug>.log holds solver progress. The formats\nare in references/dashboard.md — load it if you need to write or parse\nthat wiki. Treat any per-weakness \"kept / new record\" number in it as a train-subset self-report,\nnot a gated result; only finalize's sealed test number and any gate decision recorded in\nevents.jsonl are honest.
scripts/now.py is the one-clock timestamp stamper those formats require; it is correct and still\nused by anything writing the wiki.
Deprecation is reversible; removal is not. Still to decide (maintainer): move the wiki format\ncontract into agent-optimize as an optional output, stop dashboard.py inferring\nalgorithm = \"evograph\" from wiki/ presence, then delete this directory.
<run_dir>/wiki/.affected_tasks freeze rule, kept for reading historical run dirs.