Epic: provider SDK extraction (Phase 1)

[CameronBrooks11]

Successor to the Phase 0 conformance epic (#25, complete: 0A–0F). The 0G readiness review (#29) concluded GO for a shared provider SDK — spine-first, device-model framework deferred, with the conformance harness as the acceptance gate.

Outcome (0G)

• The protocol spine is ~96% identical across ezo/sim/bread. ~1500–2000 LOC of duplication lifts cleanly.
• One hard correctness prerequisite: T0.6 (sim/bread declare+populate CallResponse.results, like ezo) — done.
• One architectural prerequisite: converge the device-dispatch interface onto bread's adapter-module pattern in ezo+sim (Wave 4). The device framework (registry/threading/discovery generalization) stays deferred to Wave 5.
• Directional model holds: provider → SDK → contract.

Status (updated 2026-06-23)

Wave 1 — ADPP correctness — ✅ (1 harness item open)

• anolis-provider-sim#59 — CallResponse.results (T0.6) + min_timestamp (T0.7)
• anolis-provider-bread#51 — T0.6 + default signal set (T0.2) + provider_name (T0.5) + min_timestamp (T0.7)
• anolis-provider-ezo#53 — curated default signal set (T0.2)
• anolis-protocol#41 — harness assertions for §7.2 default-set + §8 results (the one remaining Wave-1 item)

Wave 2 — hygiene — ✅ COMPLETE

• anolis-provider-bread#52 — hardening parity (TSAN+valgrind) + proto pin → v1.4.0
• anolis-provider-ezo#54 — proto pin → v1.4.0
• anolis-provider-sim#60 — standardize handle_unimplemented signature
• Wave 2: shared clang-format/clang-tidy/editorconfig/justfile for providers .github#63 — shared clang-format/tidy/editorconfig/justfile
• Follow-ups (tracked): Enable CMAKE_EXPORT_COMPILE_COMMANDS in provider presets (out-of-box just lint) .github#67 (compile-commands for just lint), Adopt pre-commit (pinned hooks) as the org formatter/lint source of truth .github#69 (org-wide pre-commit — single pinned source of truth, supersedes the interim pipx-pinned format gate)

Wave 4 — device-dispatch convergence (SDK prerequisite) — ✅ DONE 2026-06-23

All three providers now share the DeviceAdapter descriptor + single typed dispatch shape (bread reference; ezo + sim converged). Verified green on merged main (ezo 28/28 + conf 38/0; sim provider 5/5 + all 16/16 + conf 41/0). Template-hardened (ezo#63: device layer independent of core). Plus: ezo#63 (template hardening), sim#70 (chaos_control cleanup — deferred hygiene). This closes the architectural prerequisite for Wave 5.
Decision: thin dispatch, NO registry. Converge ezo+sim onto bread's shape = per-type free-function modules + a thin dispatch ladder + a shared devices/common/. No registry/vtable — that's the deferred framework (Wave 5). Rationale, grounded in a full device-layer audit: ezo's value-copied RuntimeState makes a registry pure overhead; sim's real fix is a closed enum class + one dispatch site per op (-Werror=switch exhaustiveness, killing ~8 duplicated string ladders); 3 local registries would diverge the providers at the convergence wave and pre-bias the Wave-5 shared design. Behavior-preserving; conformance-gated.

• anolis-provider-ezo#55 — per-family read modules + shared devices/common/ + thin dispatch; control stays shared (reads-only adapters). Sequenced after ezo#52.
• anolis-provider-sim#61 — closed SimDeviceType enum + single dispatch site per op; engine/transport axis kept separate; promote sim's richer device_common/CallResult/StartupPolicy. sim#58 NOT folded (cross-repo pin).

Coordinated Wave-3 capability-convention fixes (same code, sequenced with Wave 4; enforced once the v1.5.0 'Bucket C' harness ships):

• anolis-provider-ezo#52 — snake_case signal_ids + per-type function_ids {1..N} (wire-safe; own PR before docs: bump README install pins to v1.6.0 + guard against drift #55)
• anolis-provider-sim#58 — motorctl function_id 10→1 (cross-repo pin in anolis tests/docs; paired anolis PR; not in chore(deps): update actions/setup-python action to v7 #61)

Deferred (re-gate after the above)

• Wave 3 remainder: Bucket C harness v1.5.0 release; Hello/WaitReady standard key sets; get_health decision.
• Wave 4 remainder: logging→sim's Logger, config toolkit unification, ezo runtime_state shrink, error-mapping→bread 1-stage, DeviceDiscoveryStrategy.
• Wave 5 — extract anolis-provider-sdk (spine + clean seams; migrate sim→ezo→bread, harness-gated). Needs a named owner before it starts.

Design refs (local working notes): provider-sdk-rfc.md, provider-consistency-audit.md, provider-sdk-0g-readiness-review.md, provider-sdk-wave4-device-dispatch-plan.md (the locked Wave-4 plan). Acceptance gate throughout: anolis-adpp-conformance + the 0F compatibility matrix.

[CameronBrooks11]

Pre-Wave-5 runway — landed (2026-06-23)

Completed the conservative A+B runway before SDK extraction:

Tier A (hard gates)

• ✅ Wave 1: harness assertions for default-signal-set (§7.2) and CallResponse.results (§8) #41 / feat(conformance): gate default-signal-set (§7.2) and call results (§8.1) #42 — harness now gates §7.2 default-set + §8.1 call-results (+ semantics clarifications). sim/ezo gated green; bread skips on its session-less mock (tracked anolis-provider-bread#62).
• ✅ Wave-4 completion — anolis-provider-sim#70 folded chaos_control into the device-adapter model (last dispatch path off the special-case branches).

Tier B (proto freeze + tooling)

• ✅ anolis-protocol v1.5.0 cut + released — Bucket C conventions + the A1 gates; no proto change from v1.4.0.
• ✅ anolis-provider-sim#58 — motorctl function_id 10→1 (+ paired anolis runtime change: integration suite now resolves the id by name).
• ✅ All three providers pinned to v1.5.0 (ezo merged; sim/bread bumps auto-merging) — conventions/§7.2/§8.1 gates verified green on each.
• ✅ Adopt pre-commit (pinned hooks) as the org formatter/lint source of truth .github#69 — org-wide pinned pre-commit template landed; ezo adopted as canary (sim/bread/anolis rollout following).

Design

• ✅ Wave-5 epic Epic: extract anolis-provider-sdk (Wave 5) #45 stood up (owner @CameronBrooks11) with the three resolved design decisions (neutral result type, taxonomy decoupling, union descriptor).
• Wave-4 trailing items (non-prerequisite) consolidated in Wave 4 trailing convergence items (non-prerequisite, pre/post-SDK) #44.

The architectural + correctness + proto-freeze prerequisites are complete. Wave 5 (extraction) is the next gate — see #45.