CLI JSON document identities¶

This file is the record of which assay.<segments>.vN identities exist in production source, and which of them are top-level JSON documents a CLI command emits.

It is hand-edited. Nothing generates it — not CI, not a committed script, not a local "generate once and check in". A pin derived from its producer is not a record of that producer; it is the producer restating itself, and it cannot fail. crates/assay-cli/tests/cli_json_identities.rs reads this file and never writes it.

A scan can tell you that an identity string exists. It cannot tell you whether that string names a document, an evidence event, a nested object, an input, or a digest domain. That judgement is what this file stores, and it is why the two blocks below are a partition rather than a list.

Adding an identity to the source fails the test until it is recorded here, in the block that says what it is. Removing one from the source fails until the row goes. See #2167 for the convention and

2484 for this guard.¶

Convention (decided in #2167)¶

The discriminator is schema, carrying assay.<segments>.vN. .vN is the breaking generation.
A closed schema stays closed; a compatible addition keeps the identity and edits the schema in the same commit. Our CI stays green; a third party holding a vendored copy of that schema goes red later, unsignalled. That is the accepted cost.
An unknown identity is fail-closed everywhere. Unknown fields are rejected on emit validators, AEE readers, receipt validators and fail-closed inputs; third-party stdout readers field-pick.

Machine-checked: identities that ARE CLI JSON documents¶

A top-level JSON document a command writes to stdout or to a caller-named path, carrying this string as its schema.

Each row is identity | writing file | naming file, where - means the writing file also names the identity. The test asserts three things: the writing file calls a writer, the naming file carries the identity, and the writing file is about this identity — it names the identity itself, or, when the two files differ, mentions a symbol the naming file publishes. Without the third, a row could be bound to a file that writes some other document and happens to call a writer. The third column exists because some documents split the two: assay doctor sets its schema in diagnostics/probes.rs and writes in doctor/implementation.rs, and the three evidence schema reports are declared in schema/reports.rs and written in schema/write.rs. Recording the split is better than loosening the check until it fits. Membership alone was not enough: an earlier revision of this file classified six of these as non-documents on the strength of a word in their name — "carrier", "projection", "health artifact" — and the partition stayed green, because it only checked that every identity appeared somewhere. Two independent reviews found all six. Binding the row to a write is what turns that class of mistake into a failure.

assay.cli.describe.v0 | crates/assay-cli/src/cli/commands/describe.rs | -
assay.doctor_report.v0 | crates/assay-cli/src/cli/commands/doctor/implementation.rs | crates/assay-cli/src/diagnostics/probes.rs
assay.enforcement_health.v0 | crates/assay-cli/src/cli/commands/monitor_next/enforcement_health.rs | -
assay.enforcement_health.v1 | crates/assay-cli/src/enforcement_health_v1.rs | -
assay.enforcement_health_projection.v0 | crates/assay-cli/src/cli/commands/project_enforcement_health.rs | -
assay.evidence.attestation.verify.v1 | crates/assay-cli/src/cli/commands/evidence/verify_attestation.rs | -
assay.evidence.coverage_attestation.verify.v1 | crates/assay-cli/src/cli/commands/evidence/verify_coverage_attestation.rs | -
assay.evidence.schema.list.v1 | crates/assay-cli/src/cli/commands/evidence/schema/write.rs | crates/assay-cli/src/cli/commands/evidence/schema/reports.rs
assay.evidence.schema.show.v1 | crates/assay-cli/src/cli/commands/evidence/schema/write.rs | crates/assay-cli/src/cli/commands/evidence/schema/reports.rs
assay.evidence.schema.validation.v1 | crates/assay-cli/src/cli/commands/evidence/schema/write.rs | crates/assay-cli/src/cli/commands/evidence/schema/reports.rs
assay.experiment.runner_phase_timing.v0 | crates/assay-cli/src/cli/commands/runner_spike/phases.rs | -
assay.init_report.v0 | crates/assay-cli/src/cli/commands/init_report.rs | -
assay.mcp.execution-record-pairing.report.v0 | crates/assay-cli/src/cli/commands/evidence/mcp_execution_records.rs | -
assay.mcp.execution-record-supersession.report.v0 | crates/assay-cli/src/cli/commands/evidence/mcp_supersession.rs | -
assay.mcp.tunnel-observed.report.v0 | crates/assay-cli/src/cli/commands/evidence/mcp_tunnel_observed.rs | -
assay.declared_mcp_manifest.v0 | crates/assay-mcp-server/src/manifest_promotion.rs | crates/assay-mcp-server/src/declared_manifest.rs
assay.mcp_manifest_candidate.v0 | crates/assay-mcp-server/src/manifest_promotion.rs | crates/assay-mcp-server/src/manifest_promotion.rs
assay.mcp_preflight.v0 | crates/assay-cli/src/cli/commands/mcp/preflight.rs | -
assay.mcp_server_inventory.v0 | crates/assay-cli/src/cli/commands/inventory.rs | crates/assay-core/src/discovery/inventory_carrier.rs
assay.monitor.observed_peers.v0 | crates/assay-cli/src/cli/commands/monitor_next/observed_peers.rs | -
assay.otel_projection.v0 | crates/assay-cli/src/cli/commands/project_otel.rs | crates/assay-core/src/otel/projection.rs
assay.policy.resolved.v0 | crates/assay-cli/src/cli/commands/policy/resolve.rs | -
assay.runner.observation_health.v0 | crates/assay-cli/src/cli/commands/monitor_next/observation_health.rs | crates/assay-runner-schema/src/health.rs
assay.privileged_mcp_action.verify.report.v0 | crates/assay-cli/src/cli/commands/evidence/verify_privileged_mcp_action.rs | -
assay.run_report.v1 | crates/assay-core/src/report/json.rs | -
assay.run_summary.v1 | crates/assay-core/src/report/summary/writer.rs | -
assay.side_effect_verification.v0 | crates/assay-cli/src/cli/commands/evidence/verify_side_effects.rs | -
assay.skill_supply_chain.v0 | crates/assay-cli/src/cli/commands/evidence/skill_supply_chain_capture.rs | -
assay.skill_supply_chain.verify_report.v0 | crates/assay-cli/src/cli/commands/evidence/verify_skill_supply_chain.rs | -
assay.supply_chain_conformance.v0 | crates/assay-cli/src/cli/commands/supply_chain_conformance.rs | crates/assay-registry/src/supply_chain.rs
assay.tool_decision_truth.otel_projection.v0 | crates/assay-cli/src/cli/commands/project_otel.rs | crates/assay-core/src/otel/projection.rs
assay.tool_decision_truth.verify.report.v0 | crates/assay-cli/src/cli/commands/evidence/verify_tool_decision_truth.rs | -
assay.trust-basis.assert.v1 | crates/assay-cli/src/cli/commands/trust_basis.rs | -
assay.trust-basis.diff.v1 | crates/assay-cli/src/cli/commands/trust_basis.rs | crates/assay-evidence/src/trust_basis/types.rs
assay.validate_report.v1 | crates/assay-cli/src/cli/commands/validate.rs | -

assay.trust-basis.diff.v1 is declared in assay-evidence, outside the scanned crates. An earlier revision recorded that as a known hole. It is not a hole: the CLI writes the document (write_diff_json), and following the write rather than the crate puts it here with everything else.

Machine-checked: identities that are NOT CLI JSON documents¶

Evidence events inside a bundle, nested objects, read-only inputs, and one digest domain. Each row is identity | reason, and the reason is required.

The partition is total, so an identity in neither block fails the build. That stops a new emitter arriving unnoticed. It does not stop a misclassification, and no static check can: naming and writing live in different files for several of these, so nothing can decide from source alone that assay.otel_projection.v0 is the document assay project-otel writes. Requiring a reason is what this file can do about it — moving a real document into this block now means writing a sentence that is false, rather than deleting a line. Two independent reviews caught six such moves in the previous revision, when this block was a bare list.

assay.coding_agent.evidence_pack.v0 | declared in assay-evidence as a bundle pack schema; no CLI write opened
assay.content_hash_scope.v1 | nested object inside `assay evidence show --format json`; reader content_hash recomputation contract, not a top-level CLI document
assay.delegated_exec_observation.v0 | assay-evidence observation event inside a bundle; ignored by privileged-mcp-action/v0
assay.mandate.v1 | mandate event carried in evidence, not written by a command
assay.mcp_manifest_observed.v0 | assay-mcp-server observation event inside a bundle
assay.mcp_manifest_projection.v0 | assay-mcp-server projection event inside a bundle
assay.mcp_tool_field.v0 | a field-level record nested in the manifest observation
assay.provider_audit_record.v0 | assay-mcp-server audit record written to the audit log, not a command document
assay.redaction_receipt.v0 | runner receipt inside an archive
assay.runner.archive_manifest.v0 | runner archive manifest; #2552 owns archive integers and identities
assay.runner.capability_surface.v0 | runner capability surface read by project-otel as input
assay.runner.claim_support_parity.v0 | runner parity artifact inside an archive
assay.runner.correlation_report.v0 | runner correlation artifact inside an archive
assay.runner.coverage_descriptor.v0 | runner coverage descriptor inside an archive
assay.runner.event.v0 | runner event stream entry
assay.runner.fidelity_verdict.v0 | runner verdict inside an archive
assay.runner.kernel_event.v0 | kernel event stream entry
assay.runner.path_projection.v0 | runner path projection inside an archive
assay.runner.policy_event.v0 | policy event stream entry
assay.runner.sdk_event.v0 | SDK event stream entry
assay.semantic-digest.jcs-rfc8785.v1 | a canonicalization profile name, not a document
assay.token_passthrough_conformance.v0 | assay-mcp-server conformance carrier inside a bundle
assay.tool_annotation_conformance.v0 | assay-mcp-server conformance carrier inside a bundle
assay.aee_run_context.v0 | input the sandbox reads; exact key set enforced on read
assay.aee_seal_key.v0 | input the sandbox reads; exact key set enforced on read
assay.aee_trust_set.v0 | input the sandbox reads; exact key set enforced on read
assay.approval_artifact.structured_meta_jcs.v0 | a canonicalization domain for approval metadata, not a document
assay.denied_call_observation.v0 | nested inside the privileged-mcp-action verify report
assay.denied_call_observation.v1 | nested inside the privileged-mcp-action verify report
assay.enforcement_decision.v0 | evidence event read from a bundle; the command writes a verify report about it
assay.fallback_projection.v0 | an input mode name for mcp-execution-records, not a schema it emits
assay.mandate.revoked.v1 | MCP lifecycle event type inside a bundle
assay.mandate.used.v1 | MCP lifecycle event type inside a bundle
assay.manifest_establish.v0 | nested inside the privileged-mcp-action verify report
assay.mcp.policy.snapshot.v1 | policy snapshot event carried in evidence, not written by a command
assay.mcp.tool-definition.snapshot.v1 | tool-definition snapshot event carried in evidence
assay.mcp.tunnel_observed.v0 | the carrier event; the command writes assay.mcp.tunnel-observed.report.v0
assay.receipt.cyclonedx.mlbom-model-component.v1 | receipt written into an evidence bundle by BundleWriter, never to stdout
assay.receipt.cyclonedx.mlbom_model_component.v1 | event type beside that receipt schema, same bundle write
assay.receipt.livekit.tool-action.v1 | receipt written into an evidence bundle by BundleWriter, never to stdout
assay.receipt.livekit.tool_action.v1 | event type beside that receipt schema, same bundle write
assay.receipt.mastra.score_event.v1 | receipt written into an evidence bundle by BundleWriter, never to stdout
assay.receipt.openfeature.evaluation_details.v1 | receipt written into an evidence bundle by BundleWriter, never to stdout
assay.receipt.promptfoo.assertion-component.v1 | receipt written into an evidence bundle by BundleWriter, never to stdout
assay.receipt.promptfoo.assertion_component.v1 | event type beside that receipt schema, same bundle write
assay.receipt.pydantic.case_result.v1 | receipt written into an evidence bundle by BundleWriter, never to stdout
assay.render_safety_conformance.v0 | a golden corpus identity compared against, not a document a command mints
assay.supply_chain_conformance.input.v0 | the input descriptor the command reads; it writes assay.supply_chain_conformance.v0
assay.tool_args.v0 | a digest domain string; nothing serializes it as a document
assay.tool_decision_surface.v0 | event type read from a bundle by verify-side-effects
assay.tool_decision_truth.recipe_row.v0 | a row inside the projection, not the projection
assay.tool_decision_truth.v0 | `assay mcp --tool-decision-truth-out` DOES append these to a caller-named path, one JSON object per line; the file is NDJSON and no line is the top-level document. An earlier reason here said "read as input", which was the same keyword error as "projection"

Machine-checked: documents with no identity at all¶

These are the rows a source scan structurally cannot produce, because none of them has a dotted identity constant. Twelve were measured from identity-less producers; nineteen more were added when

2555 followed writers to rows. A pin built only from identity strings would satisfy every other¶

check in the guard while omitting every document #2485 has to migrate. They are therefore required rows: a missing one fails the build even though there is nothing in the source to collect.

Each row is key | producer path | token, and the test asserts the producer file exists and still contains the token, so a row cannot outlive the thing it names.

baseline | crates/assay-core/src/baseline/mod.rs | schema_version
baseline_diff | crates/assay-cli/src/cli/commands/baseline.rs | baseline.diff(&candidate)
coverage_report | crates/assay-cli/src/cli/commands/coverage/report.rs | coverage_report_v1
discover_inventory | crates/assay-cli/src/cli/commands/discover.rs | DiscoverFormat::Json
hygiene_report | crates/assay-core/src/baseline/report.rs | schema_version
run_json_extended | crates/assay-cli/src/cli/commands/run_output.rs | write_extended_run_json
run_json_minimal | crates/assay-cli/src/cli/commands/run_output.rs | write_run_json_minimal
sarif | crates/assay-core/src/report/sarif.rs | 2.1.0
session_state_window | crates/assay-cli/src/cli/commands/session_state_window.rs | session_state_window_v1
sim_run_report | crates/assay-cli/src/cli/commands/sim.rs | to_string_pretty
soak_report | crates/assay-cli/src/cli/commands/sim/soak/report.rs | soak-report-v1
trust_basis_generate | crates/assay-cli/src/cli/commands/trust_basis.rs | generate_trust_basis
aee_landlock_seal | crates/assay-cli/src/cli/commands/sandbox/child.rs | maybe_emit_aee_seal
calibration_report | crates/assay-cli/src/cli/commands/calibrate.rs | to_string_pretty(&report)
coverage_legacy | crates/assay-cli/src/cli/commands/coverage/legacy.rs | LegacyOutputFormat::Json
evidence_attest_dsse | crates/assay-cli/src/cli/commands/evidence/attest.rs | sign_statement
evidence_diff | crates/assay-cli/src/cli/commands/evidence/diff.rs | DiffCliDocument
evidence_lint | crates/assay-cli/src/cli/commands/evidence/lint.rs | LintFormat::Json
evidence_lint_sarif | crates/assay-cli/src/cli/commands/evidence/lint.rs | LintFormat::Sarif
evidence_list | crates/assay-cli/src/cli/commands/evidence/list.rs | list_all
evidence_list_for_run | crates/assay-cli/src/cli/commands/evidence/list.rs | list_for_run
evidence_show | crates/assay-cli/src/cli/commands/evidence/mod.rs | "verify_mode"
evidence_store_status | crates/assay-cli/src/cli/commands/evidence/store_status.rs | StatusFormat::Json
explain_report | crates/assay-cli/src/cli/commands/explain.rs | ExplainFormat::Json
generated_policy | crates/assay-cli/src/cli/commands/generate/model.rs | PolicyOutputFormat::Json
mcp_config_path | crates/assay-cli/src/cli/commands/config_path.rs | "generated_server"
profile_file | crates/assay-cli/src/cli/commands/profile_types.rs | save_profile
profile_perf | crates/assay-cli/src/cli/commands/profile_next/mod.rs | ASSAY_PROFILE_PERF_JSON
profile_show | crates/assay-cli/src/cli/commands/profile_next/mod.rs | ProfileShowFormat::Json
signed_tool | crates/assay-cli/src/cli/commands/tool/sign.rs | sign_tool
skill_supply_chain_cdx | crates/assay-cli/src/cli/commands/evidence/project_skill_bom.rs | project_bom

key	version field	class	note
`baseline`	integer, `load` fails on `!= 1`	©	live gate; minting an identity is a #2485 review, not a rename
`coverage_report`	string `coverage_report_v1` + `report_version`	(b)	closed schema, validated on emit
`hygiene_report`	integer	©	second document, not `baseline`
`run_json_extended`	none	(d)	not `assay.run_report.v1`; may be consumed by `bundle create --from`
`run_json_minimal`	none	(d)	early-exit shape; same-or-split identity is a #2485 review
`sarif`	SARIF `2.1.0`	(g)	foreign identity (OASIS); do not mint an `assay.*` name
`session_state_window`	string `session_state_window_v1` + `report_version`	(b)	closed schema, validated on emit
`sim_run_report`	none	(d)	`assay sim run --report`; not the soak report
`baseline_diff`	none	(d)	`assay baseline check --format json` prints `BaselineDiff`
`discover_inventory`	none	(d)	`assay mcp discover --format json`; not `assay.mcp_server_inventory.v0`
`trust_basis_generate`	none	(d)	`assay trust-basis generate` writes canonical JSON; the `assert` report is a separate, named document
`soak_report`	string `soak-report-v1` + `report_version`	(b)	closed schema, validated on emit
`aee_landlock_seal`	DSSE payload type `application/vnd.assay.aee-landlock-seal.v1+json`	(g)	`assay sandbox --aee-seal`; foreign envelope, not an `assay.*` document identity
`calibration_report`	integer `schema_version` on the core model	©	`assay calibrate --out`; minting an identity is a #2485 review
`coverage_legacy`	none	(d)	`assay coverage` legacy `--format json`; not `coverage_report_v1`
`evidence_attest_dsse`	in-toto/DSSE envelope	(g)	`assay evidence attest`; foreign envelope, integers stay under #2552
`evidence_diff`	none	(d)	`assay evidence diff --format json` prints `DiffCliDocument`
`evidence_lint`	none	(d)	`assay evidence lint --format json`; not the SARIF emit
`evidence_lint_sarif`	SARIF `2.1.0`	(g)	`assay evidence lint --format sarif` via `to_sarif_with_options`; not the `sarif` row, whose producer is `assay-core` run-report SARIF
`evidence_list`	none	(d)	`assay evidence list --format json` without `--run-id` (`list_all`)
`evidence_list_for_run`	none	(d)	`assay evidence list --run-id --format json` (`list_for_run`); different object keys than `evidence_list`
`evidence_show`	none	(d)	`assay evidence show --format json` prints manifest + events + `content_hash_scope`
`evidence_store_status`	none	(d)	`assay evidence store-status --format json`
`explain_report`	none	(d)	`assay explain --format json`
`generated_policy`	none	(d)	`assay generate --format json` writes a policy, not an `assay.*` document
`mcp_config_path`	none	(d)	`assay mcp config-path` JSON object
`profile_file`	none	(d)	`save_profile` writes a Profile JSON file
`profile_perf`	none	(d)	`ASSAY_PROFILE_PERF_JSON` sidecar from `assay profile update`; not `profile_show`
`profile_show`	none	(d)	`assay profile show --format json`; not the perf sidecar
`signed_tool`	none	(d)	`assay mcp tool sign` writes a signed tool definition
`skill_supply_chain_cdx`	CycloneDX 1.6	(g)	`assay evidence project-skill-bom`; foreign identity, do not mint `assay.*`

Three of these rows — baseline_diff, discover_inventory and trust_basis_generate — came from an independent read, not from the author of this file. Two separately written inventories both missed them. That is the argument for the required-rows check stated as evidence rather than as principle: the documents a scan cannot see are also the ones a single careful reader does not think of, and there is no instrument that fixes that.

Nineteen of the rows above were not on that first measured twelve. They were found by reading the emit path of each unaccounted serializer on 1c560a912, which is #2555: the file writes JSON, so the inventory has to say what that JSON is. No production command changed. Adding the row does not prove the classification; it records that a document was shipped unnamed. The converse detector is still file-level, so several of these keys share a producer; the extra rows exist so a second emit form in that file is not hidden behind the first.

Machine-checked: JSON-serializing command files that are not document producers¶

A production file under crates/assay-cli/src/cli/commands that serializes JSON through write_stdout_json, to_string_pretty, to_vec_pretty, to_writer, serde_json::to_string(, or serde_json::to_vec( — after inline #[cfg(test)] modules and external */tests.rs files are removed — must be named by a cli-documents writer/namer, an unnamed-documents producer, or listed here with a non-empty motive that names the emit or helper role. Paths are the exact workspace-relative POSIX path; a basename does not match. A stale opt-out (missing path, file that no longer serializes, or path already named by a document row) is a failure. The motive is a reviewable declaration, not mechanical proof. The machine block is a text fence of path | motive lines — pipe-delimited text, not JSON.

crates/assay-cli/src/cli/commands/ci.rs | stdout helper: write_stdout_json of render_summary_json for already-recorded assay.run_summary.v1 (writer is assay-core report/summary/writer.rs)
crates/assay-cli/src/cli/commands/coverage/io.rs | emit helper: to_vec_pretty of coverage_report_v1 whose unnamed producer is coverage/report.rs
crates/assay-cli/src/cli/commands/evidence/adapt_skill_scan.rs | second emit: write_stdout_json of an augmented assay.skill_supply_chain.v0 carrier; the cli-documents writer is skill_supply_chain_capture.rs
crates/assay-cli/src/cli/commands/evidence/explore.rs | TUI helper: to_string_pretty of event.payload for the detail pane, not a CLI document
crates/assay-cli/src/cli/commands/evidence/livekit_tool_action/canonical.rs | hash helper: serde_json::to_string of keys and strings while building canonical JSON for a digest, not a CLI document
crates/assay-cli/src/cli/commands/import.rs | NDJSON helper: serde_json::to_string of each TraceEvent into a .trace.jsonl file; no line is a top-level document
crates/assay-cli/src/cli/commands/mcp/coverage_input.rs | NDJSON helper: serde_json::to_string of one {tool, tool_classes} object per line as coverage_report_v1 input
crates/assay-cli/src/cli/commands/replay/fs_ops.rs | config helper: to_string_pretty rewriting settings.seed in an existing JSON config
crates/assay-cli/src/cli/commands/replay/provenance.rs | rewrite helper: to_string_pretty annotating provenance onto an existing run.json
crates/assay-cli/src/cli/commands/run.rs | stdout helper: write_stdout_json of render_json for already-recorded assay.run_report.v1 (writer is assay-core report/json.rs)
crates/assay-cli/src/cli/commands/sandbox/profile.rs | digest helper: serde_json::to_string of a degradation record as hasher input, not a CLI document
crates/assay-cli/src/cli/commands/sim/soak/mod.rs | emit helper: to_string_pretty of soak-report-v1 whose unnamed producer is sim/soak/report.rs
crates/assay-cli/src/cli/commands/trace.rs | NDJSON helper: serde_json::to_string of each ingested V2 event to --out-trace; no line is a top-level document
crates/assay-cli/src/cli/commands/trace/import_mcp.rs | NDJSON helper: serde_json::to_string of each TraceEvent to --out-trace; no line is a top-level document

What this guard cannot do, stated plainly¶

It cannot verify a classification, and the reason column is a speed bump, not a gate. The test checks that a row has a | and a non-empty right-hand side. It does not read the sentence: a banana passes. A misclassification costs writing a false sentence instead of deleting a line, which is why two independent reviews caught six of them — but the build stays green either way. The theory that produced those six ("a projection, not a document") writes a perfectly non-empty reason. #2485 must not read these sentences as evidence that a row is right. Three of them have been wrong today, and the third was written after conceding the first two.
The writer-to-identity tie is a token check, not dataflow. A writer that mentions the right constant while serializing something else would pass. Closing that means following the value into the schema field, which nothing here does. Both sides now ignore comments and generic published names (new, from, parse, …); one row was briefly tied by a single //! line, and another would have stayed tied through Sha256::new() alone.
The split-row tie has no demonstrated source-mutation red, and cannot have one. Severing it in source — removing the writer's use of the naming file's symbol — does not compile (unresolved import). That is a stronger guarantee than a red test and it is why the mutations for this property are edits to this file rather than to code. Stated because a guard whose reds are all document edits has not been shown to notice a source change, and here the reason is structural rather than an omission.
A shared naming file is one bit for several rows. evidence/schema/write.rs and schema/reports.rs serve three identities, and the tie for all three is the same use. Requiring a symbol unique to each identity would need a fourth column naming it; that is not done here.
The writer list is a fixed set of idioms, and it has been wrong twice. serde_json::to_writer was missing until assay describe exposed it; to_vec_pretty, tokio::fs::write, serde_json::to_string and a bare fs::write were missing until a review opened the emit paths. A file emitting through some route still not on the list is invisible to documents_are_bound_to_a_writer.
The converse detector is file-level, not dataflow, and its idiom list is not WRITERS. A command file that serializes through println! or a filesystem helper not on the six issue idioms stays invisible. A second writer of an already-recorded identity is an opt-out, not a second cli-documents row, because that block is keyed by identity. import and trace write NDJSON; they are opted out for the same reason assay.tool_decision_truth.v0 is not a document. Classifications cite the emit path and remain reviewable declarations.
Twenty-one of the non-document rows are declared in dependency crates and their emit paths were not opened. They are classified from the declaring module. That is a weaker basis than the rows whose producer I read, and three such judgements have already been wrong today.
Five crates declare identities the CLI writes — assay-cli, assay-core, assay-evidence (trust-basis.diff), assay-registry (supply_chain_conformance) and assay-runner-schema (runner.observation_health). The collector scans the first two; the rest are reached only through a document row's naming column, which is why runner.observation_health was missing from the first revision of this file and the guard was green without it. A sixth crate is now caught: every_production_identity_is_classified also collects pub const identities from every crate assay-cli depends on, and DEPENDENCY_CRATES is pinned against assay-cli/Cargo.toml so the list cannot drift away from the manifest. A new JSON-serializing command file in a crate already scanned is #2555's converse rule.
describe ⊆ pin is one-directional. BINDING_ROWS is seven clap paths and omits the run report and run summary. Forcing describe to grow is a product decision, not bookkeeping.

Provenance of this file¶

The identity strings were located by searching source; the classification, the writer bindings and every reason are hand-made judgements, and they are the content. Three rows — baseline_diff, discover_inventory, trust_basis_generate — came from an independent read after two separately written inventories missed them. Six identities were misclassified as non-documents in the first revision on the strength of a word in their name ("carrier", "projection", "health artifact"); two independent reviews found all six, which is why the reason column exists.

What is not here¶

Bundle manifests, trust cards, attestation and lint packs keep their own integers under #2552. The CLI field-name rule does not cross that crate boundary by implication.