boshu2__agentops/docs/3.0-readiness.md

AgentOps 3.0-Readiness Level-Set (2026-05-23)

Bead: soc-hplds (W4, final wave of the 3.0 reconciliation epic soc-9y8hd). North star: docs/3.0.md. This is the honest "where we are" against the 3.0-ready acceptance criteria after the reconciliation push. It does not claim more than is true; remaining work is named with its tracking bead.

What the reconciliation did

The operator-directed 3.0 reconciliation (2026-05-23) ran as a cron-paced /evolve loop. No new features — tighten, match reality, finish the hexagon. Waves:

Wave	Bead	Outcome	PRs
W0	soc-25sgs	docs/3.0.md north star authored + woven into README/PRODUCT/GOALS/CLAUDE/docs-index/mkdocs	#468
W1a	soc-furq8	"2,514 LOC dead RPI code" claim was false (all load-bearing) — corrected CLAUDE.md to match reality; real removal → soc-1gbpz	#469
W1b	soc-icvff	Aspirational features (gt convoy, bd mol/cook, curation stages 3-6) → docs/ROADMAP.md, cut from live docs	#470
W1c	soc-sb5r1	Workbench Δ=0 claim kept (artifact exists); unsubstantiated 3-judge-parity claim downgraded to internal finding	#471
W1d	soc-onfm1	Stale dream_executor.go path fixed; 3 of 4 flagged items were already correct	#472
W2	soc-qk4b	Hexagon crank tail: 71/80 skills now carry canonical .feature acceptance (hexagonal_role already 100%)	#473-479
W3	soc-6zihw	Hook-noise audit then full hook removal: noise-injectors deleted, then all remaining gates/adapters removed outright (soc-57b7f) — AgentOps 3.0 ships hookless (skills + the ao CLI only)	#480, #483, #488, #494
W4	soc-hplds	This level-set	—

A recurring pattern: the discovery over-stated scope, and verify-before-acting caught it every time (W1a dead-code was live; W1c only one of two claims was unbacked; W1d 3 of 4 already correct). That is the reconciliation working as intended — match what reality is, not what the gap report assumed.

Acceptance status (against docs/3.0.md "What 3.0-ready means")

• [x] North star is the single referenced source of truth — docs/3.0.md authored (W0); README, PRODUCT, GOALS, docs-index, mkdocs nav, and CLAUDE.md all point to it.
• [x] Docs match reality — the dead-code claim, aspirational-feature claims, unsubstantiated market claim, and stale path that the audit found are all corrected (W1a-d). Aspirational items live in docs/ROADMAP.md; market claims are artifact-gated.
• [x] Hexagon complete (stable surface) — every skill carries hexagonal_role; 71/80 carry canonical .feature acceptance. The 9 without are intentionally exempt, not skipped: expert-council + inject (deprecated/removal-target), shared + standards + using-agentops + converter + provenance + flywheel (meta/internal/not user-invocable), llm-wiki (open proposal, no implementation — a .feature would assert unbuilt behavior).
• [x] Hooks audited for noise → none a noise-injector — audit complete (W3, docs/architecture/hook-noise-audit.md); the 14 flagged entries split into 7 true noise-injectors (deleted across soc-e2ju0 S1–S3) and 7 conditional injectors (reclassified as kept-gates). 46 hooks remain, all gates or bounded adapters — none a noise-injector (soc-cul67). The separate default-install-zero-hooks lift is soc-e2ju0 S4, still open and not a 3.0-close blocker.
• [x] The loop is executable — the operating loop (discovery → plan → implement → validate → ratchet) is documented in operating-loop.md and was exercised end-to-end across this reconciliation's 13 PRs via /discovery, /crank, worktree-per-bead, CI validation, and squash-merge ratchet.
• [ ] Every AgentOps bead worked or closed — the reconciliation waves (W0-W4) and the off-3.0 beads are closed; remaining agentops-scoped 3.0-aligned beads (coherence/ledger hardening: soc-ficdw, iha27, v3y80, et al.) are open and tracked, not abandoned. Fleet/other-repo soc-* beads (personal-site, mt-olympus, bushido-infra, showcase, platform-lab) are out of scope per the ratified content-scoping decision.

Net: 5 of 6 acceptance boxes met; the 1 open box (every AgentOps bead worked or closed) has named tracking beads, not unknowns.

Fitness snapshot (ao goals measure, 2026-05-23)

Score: 85.7% (26/31 gates passing). Two non-blocking fails:

• flywheel-lifecycle — 30s timeout; environmental CI flakiness (soc-iwrm2), not a code regression.
• executable-spec-link-integrity — warn-only (F1.6, continue-on-error in CI); flags GOALS.md directive↔scenario linkage gaps (all 14 directives show 0/0 linked scenarios), a pre-existing gap unrelated to the skill .feature crank.

Neither blocks PRs; both are tracked signals, not reconciliation regressions.

Remaining work (named, not blocking the 3.0-reconciliation close)

• soc-e2ju0 S4 — default install goes to zero hooks (the hooks-noise box is already met by S1–S3 deleting the 7 noise-injectors; S4 is the remaining default-install lift, not a 3.0-close blocker).
• soc-1gbpz — migrate-or-remove the legacy RPI lane (caller-migration refactor; the code is live, not dead).
• Coherence/ledger hardening beads (soc-ficdw et al.) — agentops 3.0-aligned, open.
• ADR-0002 S2-S5 — hookless default-install + hookless RPI; explicitly a follow-on, not a 3.0-close blocker (docs/3.0.md).
• GOALS directive↔scenario linkage (the warn-only F1.6 gap).

Verdict

The 3.0 reconciliation achieved its mandate: docs match reality, the hexagon is complete across the stable skill surface, and the hooks are audited. The repo is substantially 3.0-ready; the remaining items are bounded, tracked, and (for hookless-default) explicitly deferred. The loop that produced this is the loop docs/3.0.md describes.