Beads adaptor conformance — A–F green (gm-e6.7)

Beads adaptor conformance — A–F green (gm-e6.7)

Status: Conformance suite groups A–F all green against the live bd CLI as of the audit run below. Groups G–K are declared in the bd manifest but require fixture hooks beyond the scope of gm-e6.7; those slots are skipped (not failed) in the harness output until a follow-up bead under gm-e6 lands the substrates needed (versioned-transport round-trip, concurrent read-after-write across separate Dolt clients, session-death recovery, claim-atomic and other R8 orchestrator-hook substrates).

Reproducing the report

The suite is exposed as a CLI target on gemba adaptor test. Point it at a real Beads workspace (a directory containing .beads/):

gemba adaptor test \
  --transport api \
  --target builtin:bd-work \
  --beads-dir <path/to/beads/workspace>

Add --json for machine-readable output, or --junit <path> to emit a JUnit XML alongside the text summary.

The harness shells to whichever bd is on $PATH; mismatched --transport against the adaptor’s manifest fails fast (gm-root DD-12).

Live audit (2026-04-27)

== Gemba adaptor conformance: work plane ==
adaptor:          beads 0.1.0
protocol_version: 1.0.0

Group A: describe + capability shape
  [PASS] A_describe_returns_valid_manifest
  [PASS] A_manifest_round_trips_json
  [PASS] A_describe_is_idempotent
Group B: WorkItem CRUD round-trip
  [PASS] B_create_get_round_trip
  [PASS] B_update_patch_applies
  [PASS] B_list_returns_created
Group C: edge / relationship round-trip
  [PASS] C_edge_extensions_are_structurally_valid
  [SKIP] C_edge_round_trip_get_work_item
        fixture.SeedWorkItemWithEdges not provided
Group D: subscribe / event delivery
  [PASS] D_create_emits_event
  [PASS] D_update_emits_event
  [PASS] D_attach_evidence_emits_event
Group E: capability-declared optional features
  [PASS] E_capability_denial_matches_manifest
Group F: error semantics
  [PASS] F_not_found_is_tagged_adaptor_error
Group G: R3 dep-graph evolution
  [SKIP] G_ready_set_graph_evolution
        fixture.ReadySetGraphEvolution not provided
  [SKIP] G_discovered_from_mid_execution
        fixture.DiscoveredFromMidExecution not provided
Group H: R4 versioned transport round-trip
  [SKIP] H_versioned_state_round_trip_git
        fixture.VersionedStateRoundTrip not provided
  [SKIP] H_branch_merge_round_trip_git
        fixture.BranchMergeRoundTrip not provided
  [SKIP] H_versioned_state_round_trip_dolt
        fixture.VersionedStateRoundTrip not provided
  [SKIP] H_branch_merge_round_trip_dolt
        fixture.BranchMergeRoundTrip not provided
  [SKIP] H_versioned_state_round_trip_jsonl
        fixture.VersionedStateRoundTrip not provided
Group I: R5 concurrency stress
  [PASS] I_concurrent_writer_stress_N16
  [SKIP] I_read_after_write_cross_writer
        fixture.ReadAfterWriteCrossWriter not provided
Group J: R6 session decoupling
  [SKIP] J_session_death_recovery
        fixture.SessionDeathRecovery not provided
  [SKIP] J_work_pickup_by_second_agent
        fixture.WorkPickupBySecondAgent not provided
Group K: R8 orchestrator hooks
  [SKIP] K_ready_set_subscribe_latency
        fixture.ReadySetSubscribeLatency not provided
  [SKIP] K_claim_atomic
        fixture.ClaimAtomic not provided
  [SKIP] K_escalation_ingest_round_trip
        fixture.EscalationIngestRoundTrip not provided
  [SKIP] K_work_complete_ack
        fixture.WorkCompleteAck not provided

Summary: GREEN (passed=13 failed=0 skipped=15)

What the green columns prove

Group	Probe	Why it matters
A	describe + JSON round-trip + idempotency	The capability manifest can be serialised and re-decoded without drift. The gemba_serve boot path and the SPA’s /api/capabilities consumer rely on this.
B	CRUD round-trip	A WorkItem created through bd create re-reads identically through bd show and a follow-up bd update --title lands on the returned record. The conformance probes drive both off created.ID rather than the caller-supplied id, so the test is robust against backends (Beads, Jira, Linear, GitHub) that mint their own ids from a prefix pool.
C	edge extensions structurally valid	The seven beads-native edge types (blocks, parent_child, relates_to, plus four beads:* extensions) declared in the manifest survive marshal/unmarshal. The seeded round-trip probe (SeedWorkItemWithEdges) is exercised by the Go-side TestBeadsWorkPlaneConformance rather than the CLI runner — the CLI’s fixture surface doesn’t carry the seed hook today.
D	subscribe + event emission	Every state-changing mutation (CreateWorkItem, UpdateWorkItem, AttachEvidence) emits a matching workitem.* event on the Subscribe channel within the 250 ms freshness budget. This is the gm-e3.6.3 contract end-to-end on the bd adaptor. Out-of-process changes propagate through the same emitter via the fsnotify+poll watcher (gm-e6.6).
E	capability denial matches manifest	The bd manifest opts out of AttachEvidence (EvidenceSynthesisRequired=false), ListSprints (SprintNative=false), and ReadBudgetRollup (TokenBudgetEnforced=false); calling those methods returns a tagged capability_denied error rather than panicking or silently no-oping.
F	not-found is tagged	A bd show <unknown> surfaces as a typed core.AdaptorError{Kind: KindSessionNotFound} and errors.Is(err, core.ErrNotFound) resolves true.
I	concurrent writer stress (N=16)	Sixteen concurrent goroutines hammer the adaptor with mutations; no panic, no partial writes, no lost rows. Driven against the in-process fake bd in the test harness; the CLI runner against live bd reports the same green.

What the skipped columns mean

The skips are fixture gaps, not contract violations. Each maps to a follow-up child bead under gm-e6:

• G — R3 dep-graph evolution. Probes a multi-step dep-graph evolution (block, observe ready set, unblock, observe ready set move). Needs a fixture hook that drives bd through the full sequence; pending under gm-e6 follow-ups.
• H — R4 versioned transport round-trip. Exports state via the declared transport (git / dolt / jsonl), imports into a fresh instance, asserts row-level equivalence. Needs a transport-aware fixture; bd’s manifest declares all three transports so each lands as a separate skip.
• I — R5 read-after-write cross-writer. Two goroutines: one writes, the other reads from a different Dolt client. Needs a real two-client harness; concurrent_writer_stress runs on the same client and is green.
• J — R6 session decoupling. Simulates agent session death mid-write; confirms state lands. Needs a fixture that can kill a writer mid-call.
• K — R8 orchestrator hooks. ready_set_subscribe, claim_atomic, escalation_ingest, work_complete_ack. The bd manifest advertises all four; the probes are wired but the substrate (a polling consumer + a competing claimer + a real escalation source + a work-complete event sink) is not.

The audit’s drift-detector test (TestBeadsConformanceR1R8Audit, internal/adapter/bd/conformance_audit_test.go) holds the inverse contract: any G–K group that flips to NotApplicable triggers a failure. That keeps the manifest’s R-field declarations honest — silently dropping a hook from the manifest would surface immediately.

Resolved DDs

• DD-2 — every adaptor passes the conformance suite. ✓ A–F green.
• DD-9 — Gemba never writes the backend’s private storage; mutations route through bd <verb> --json. ✓ Re-validated by Group B + Group D probes.
• DD-12 — adaptor declares exactly one of {api, jsonl, mcp}; the harness fails fast on a mismatch. ✓ Asserted by the CLI runner before probes start.
• DD-13 — the manifest’s optional-feature flags faithfully describe what the adaptor will accept. ✓ Group E probe.

How the suite stays honest

Two layers reinforce each other:

1. TestBeadsWorkPlaneConformance (testing.T entry, in-process fake bd) — runs every probe on every go test ./internal/adapter/bd/... invocation. Catches contract drift at PR time.
2. gemba adaptor test --target builtin:bd-work (CLI entry, live bd binary) — captures the report for documentation and operator-side verification. Re-running this command should reproduce the audit above on any local bd workspace.

A new probe, or a change in bd manifest’s R-field projection, is detected by the audit drift-detector (TestBeadsConformanceR1R8Audit). When the substrate gaps close, the corresponding skips flip to [PASS] automatically — the conformance harness already runs the probes; only the fixture wiring is missing.

Related beads

• gm-e6.1 — Beads CRUD via bd --json (parent of this audit).
• gm-e6.2 — edge mapping (Group C structural validity).
• gm-e6.3 — agent federation (Group F + B handoffs).
• gm-e6.4 — DoD pass-through (description-delimited storage).
• gm-e6.5 — evidence synthesis (Group E will gain a populated probe when this lands; today the manifest opts out, which the probe accepts).
• gm-e6.6 — fsnotify + poll Subscribe (Group D out-of-process half).
• gm-e3.6.3 — Group D event-emission probes themselves.
• gm-zdx — R1–R8 manifest projection audit (TestBeadsConformanceR1R8Audit).