Everyday Agent Validation Product v1

Status: local evidence only, unsupported and unpublished. This document freezes the bounded validation product built for issue ABI-09A.17 (#166); it is not a support or publication decision, and it does not narrow or pre-empt PG-9 (#165/#175).

Audience: reviewers of #166, maintainers deciding whether to raise MAX_BUILDER_BYTES (SPX-G171) or MAX_BYTES_COPY_SITES (SPX-T267) for bounded workloads, and any later agent extending examples/everyday-agent-project.

Why this document exists

#166 asks for a "versioned end-to-end validation product/profile" pinning request/fixture schema, Agent definition, checkpoint/recovery rules, output schema, evidence schema, bounds, supported target(s), and nonclaims. Those facts already existed, spread across examples/everyday-agent-project/README.md and tests/agent_runtime_v1/everyday_agent_product.rs. This document is the one place that pins them together with exact identifiers, matching this repository's convention for a versioned product profile (compare Project Profile Admission v1, Public Generic Boundary Profile v1). It restates no capability the linked sources do not already have; it adds no code, .spx source, diagnostic, or CLI surface.

Profile identity

Layer	Identifier
Product profile	semaprax.everyday-agent-public-generic-validation.v1
Project	examples/everyday-agent-project (semaprax.toml, profile = "filesystem-io.v3")
Data/JSON/filesystem module	everyday.manifest (src/manifest_json.spx)
Source-declared Agent	everyday.agent (src/agent.spx:84)
Agent module	everyday.agent.lifecycle
Validation harness	tests/agent_runtime_v1/everyday_agent_product.rs
Depends on (blocked)	ABI-09A.13 (#162), ABI-09A.14 (#163)
Governed by	PG-9 decision — not affected by this document

This product is two real, already-shipped foundations composed in one project, deliberately not wired together into one combined Agent-effect-calls-fs pipeline. See Nonclaims for exactly why, with reproductions.

What it proves

Verified by semaprax check/semaprax test on the project, semaprax check on the Agent module standalone, and cargo test --locked -p semaprax --test agent_runtime_v1 everyday_agent (13 of 126 tests in that binary; see Evidence):

• canonical .spx source and an exact ProgramRoot (semaprax check reports a stable content digest for both the project and the Agent module alone);
• bounded, typed fs.read / fs.write-new (no-clobber) filesystem interaction against a real temporary directory, through the same with_authenticated_project / execute_filesystem_command Project route std.fs's own tests use;
• bounded JSON validation, member lookup, and classification over a fixed flat manifest;
• a canonical, bounded, deterministic report;
• a source-declared Agent (@id("everyday.agent")), compiled and selected by stable ID (compile_source_agent_lifecycle);
• explicit authorization before the external boundary (everyday.agent.fn.authorize), proven to stop before the boundary (read.calls == 0 on refusal, not merely a refused status);
• durable per-operation checkpoints and crash/restart recovery without a duplicate external read, across all five CrashPoints (BeforeIntent/AfterIntent/AfterEffect/AfterSettlement/BeforeDelivery), with genuine "uncertain, no blind retry" semantics at the two boundaries where the read may or may not have happened;
• checkpoint tamper detection (truncation, journal reorder, one injected byte), resume() refusing a revoked policy_epoch, and resume() reporting a display-only source rename as source_drift (ProgramRoot drift);
• replayable evidence (DurableRun::evidence()) binding identities, digests, and counts, asserted not to carry the fixture bytes, the authorization seal, or the task payload.

Request/fixture schema

The fixed, flat, three-record manifest (fixtures/input.json):

{"schema":"everyday-agent-manifest.v1","record_count":3,
 "rec0_id":"m1","rec0_tag":"ok","rec0_payload":"alpha",
 "rec1_id":"m2","rec1_tag":"warn","rec1_payload":"beta",
 "rec2_id":"m3","rec2_tag":"ok","rec2_payload":"gamma"}

record_count is pinned at exactly 3 (see Bounds for why); each record has an identifier, a Copy-scalar tag ("ok" accepts, anything else including "warn" is rejected), and an owned text payload bounded at 16 bytes. This is a fixed fixture schema, not the configurable input_manifest/output_report/max_input_bytes/digest-bearing work request the issue's illustrative scenario sketches — building that general request layer on top of std.fs plus std.data.json.doc is exactly what hits SPX-G171 (below).

Output report schema

The canonical, 7-byte report: {"a":<accepted_digit>}. Because record_count is pinned at 3, a reader recovers rejected_count as 3 - accepted_count without a second field. For the fixture above: {"a":2} (two "ok" tags accepted, one "warn" tag rejected).

Checkpoint/recovery rules

Unmodified Agent Lifecycle v1 durable machine (bind_durable_agent/DurableAgent, Agent operation checkpoint v2): one settled observation per intent, one external read per completed run, uncertain dispatch (AfterIntent/AfterEffect) left DurableStatus::Unknown until explicit host reconciliation, revoked policy_epoch refused before any further external effect, and display-only source rename reported as source_drift — the last of eight checks in CheckpointBinding::drift (src/agent_lifecycle/durable/checkpoint.rs:185-207), so the assertion proves the display-only case is not silently absorbed by an earlier, broader drift check.

Evidence schema

DurableRun::evidence() is one canonical JSON line binding: terminal status/reason, boundary_crossings, an evidence_digest (sha256:-prefixed), and the checkpoint chain's own digest (AgentCheckpoint::decode(..).digest() equals the store's recorded digest). It is asserted to omit the authorization seal ("AZ"), the manifest-review identity strings, and the task payload — evidence binds facts, not payload.

It does not bind descriptor/provider/carrier digests, because there is no real descriptor/provider/carrier call in this product (see Nonclaims).

Bounds

Two are real, reproducible language/tooling limits this product ran into, re-verified against current main, not arbitrary round numbers:

Bound	Value	Site	Reproduction
Workspace Semantic Graph builder_bytes (SPX-G171)	67,108,864 bytes	MAX_BUILDER_BYTES, src/workspace_graph.rs:59	The former 18,874,368-byte ceiling rejected this product's std.data.json.doc + std.fs closure at a 23,741,424-byte forecast. The raised ceiling admits that bounded closure; oversized projections still fail closed at the same diagnostic.
bytes_copy/bytes_set/writer_write_u8 interprocedural site count (SPX-T267)	16 sites	MAX_BYTES_COPY_SITES, src/byte_data_capacity.rs:13	Extend report_writer past the 7-byte report: bytes_copy path reaches 17 sites; limit is 16.

The interprocedural site count remains why record_count = 3 and the 7-byte, one-field report are this product's checked bounds with a real fs.write. The graph ceiling was raised for the measured, bounded multi-package workload; future increases still require a measured capacity case rather than an application-local workaround — this is the same class of ceiling already blocking #124 and forcing the reference application (#194) to reimplement std.db/std.http locally, so this product is a third, independently-reproduced case to decide that question against rather than a new one.

Supported validation target

Reference interpreter only (semaprax check/semaprax test, and the cargo test harness above, which runs on the host toolchain — no native C11 or Core Wasm Agent-stage execution is claimed by this product). Native and Wasm public-generic provider execution is explicitly out of scope (below).

Nonclaims

This product does not deliver, and does not claim to deliver:

• A real public-generic descriptor/provider/carrier call. The Agent's execute effect is backed by the test harness's own std::fs::read of the real fixture, standing in for "the host executes the explicitly selected generated consumer." The generated-consumer assets under src/public_generic_consumer/ are explicitly test-only template fixtures bound to a fixture provider literally named unsupported-unpublished. #162 and #163 (both P0 and open at the time this profile was written) own turning that into a real, supported call; this product does not pre-empt or imply their outcome.
• Native or Wasm provider execution of that call, for the same reason.
• Evidence binding descriptor/provider/carrier/result-carrier digests, since there is no real descriptor/provider/carrier here to bind.
• The Agent's operations (initialize/observe/authorize/reduce) themselves performing fs.read/fs.write. Only execute is an effect fn, realized by the host — this is the existing Agent-runtime authority boundary, not a limit this product introduces.
• The iterative, multi-turn AGENT-ITERATIVE-LIFECYCLE-V2 profile or AGENT-STATE-MIGRATION-V3 (both are iterative-lifecycle (V2) facilities; this product uses the non-iterative Lifecycle v1 durable machine).
• Hostile cross-paired-provider rejection, since there is no real provider pairing here to cross.
• The Agent's own .spx operations reading real JSON — the manifest is passed to the Agent only as an opaque Bytes byte count for its own bookkeeping. The genuine JSON validation/classification lives entirely in everyday.manifest, run separately.
• General distributed exactly-once execution.

Evidence

• semaprax check examples/everyday-agent-project — verified, exit 0.
• semaprax test examples/everyday-agent-project — project tests passed, exit 0.
• semaprax check examples/everyday-agent-project/src/agent.spx — verified standalone, exit 0.
• cargo test --locked -p semaprax --test agent_runtime_v1 everyday_agent — 13 passed, 0 failed (all cases named above).
• cargo test --locked -p semaprax --test agent_runtime_v1 (full binary, regression check) — 126 passed, 0 failed.

All local; no hosted CI run is claimed for this product.