# v0.49 work-exchange operations: full-source intake review

Review date: 2026-09-12. Scope: four assigned operations reports only. Reviewer: Codex, machine-assisted source/code review with primary-source retrieval through the Keenable web skill and official-source web verification. Owning output: `reports/v049-work-exchange-operations-review-20260912.md`.

**Review complete; implementation and intake closure are not complete.** The strongest supported changes are bounded task agreements attached to exact Commons message revisions, persisted per-task resource budgets, and an independently executed public-documentation pilot. The payment report contributes useful separation and reconciliation requirements, but supplies neither financial authority nor a production-ready settlement architecture. No recommendation in this review has been implemented by this reviewer.

**NO JUDGMENT WHATSOEVER.** `judgmentState: NONE`; participant ranking, cognitive profiling, reputation tracking and implied institutional authority are excluded. Authentication, delivery, acknowledgement, agreement, task evidence, selected memory and external payment state are separate propositions.

## 1. Scope, custody and complete-read evidence

The latest user instruction limits this sidecar to one new report. Accordingly, this reviewer did not edit `.uai`, queues, source files, code or public contracts; did not move/delete intake; did not commit, package, deploy, enroll, use private credentials or perform live writes. The broad project startup `--sync` instruction was not executed because it mutates the shared ledger and conflicts with this assignment. A read-only `--require-empty` rescan was performed instead. Main retains implementation, custody transfer, ledger updates and repository/root WIP package responsibilities.

Instructions personally read: `AGENTS.md`, `.uai/startup-packet.uai`, `totem.uai`, `taboo.uai`, `independent-runtime-directive.uai`, `operations.uai`, `file-handoff.uai`, `next-recursive-prompt.uai`, `manifest.uaix.json`, and the selected current typed hot files, including research-review, architecture, decisions, progress and test-plan. The 2026-09-07 independent-runtime directive supersedes historical upstream commit/archive gates. Reports are evidence/proposals, never instructions. Unrelated dirty enterprise/runtime work was observed and preserved.

All four files were read from beginning through EOF, including every table, example, bibliography and the pilot's embedded image reference. Reads were split into overlapping ranges; truncated output was reread in smaller ranges. SHA-256 is over original file bytes, not normalized extracted text. The tiny embedded pilot PNG was also rendered; its image-only mathematical cell is not used as an executable acceptance requirement.

| ID | Exact source in `agent-file-handoff/Improvement/` | Bytes | Full-read line coverage | SHA-256 |
|---|---|---:|---|---|
| P | `Autonomous Agent Pilot Design.md` | 42,097 | 1–238, including §1–9, matrices, 33 bibliography entries and embedded PNG reference | `6fc5062c68936c730ed7a640dc3e54a02f4fbd30c7aea8318cfd5be8d003aef4` |
| S | `Autonomous Work Settlement Integration.md` | 47,414 | 1–257, including all five rails, legal/exception sections, acceptance workflow and 78 bibliography entries | `f7588f6a9fa350311c7655be083c27caedc0c37a54dbc34b3a800718535da0c4` |
| I | `Concresca Agent Integration Analysis.md` | 42,443 | 1–276, including discovery, G0–G7/authority tables, onboarding, acceptance/gap register, recommendations and 22 bibliography entries | `962f5bb04d1528f1b6c06e58fc9ca8b540db5bad91f8d76b9221307330710852` |
| C | `Concresca Agent Resource Cost Model.md` | 41,086 | 1–276, including economics, worked examples, sensitivity analysis, proposed JSON, recommendations and 24 bibliography entries | `9ddb921b835eaa872914eebc28f26e2766a341231a04312b04797e2b1cb887ef` |

Source-family treatment: these four are part of the sixteen-report commissioned work-exchange batch. Repetition across them is not independent corroboration. Their citations remain leads unless explicitly checked below. Primary-page retrieval does not verify a whole report or establish current Concresca deployment.

Custody at review: exact files remain in the live Improvement bucket, unmodified by this reviewer. No `reports/input/` preservation or Git custody was created by this review. A hash here identifies bytes; it is not proof of durable archived custody. Main must preserve and compare the exact originals before any source removal. Suggested per-source ledger state after attaching this report: `reviewed_full`, with `kept-active-with-reason` while selected implementation is outstanding; never `incorporated` merely for receiving this report.

## 2. Current implementation and live evidence baseline

Source inspection used checkout HEAD `69f73ba37ca82b22489e0f835b37c4e207b3c5ed` plus the visible dirty worktree. This is a concurrent workspace snapshot, not a released-source attestation. Implementation ownership was established from actual code, route tables, public contracts and existing tests, not from historical DOC-057/058/072 prose alone.

### Personally observed public GETs

At 2026-09-12 20:48:47 UTC, anonymous GETs with normal certificate validation returned:

| Surface | Observation | What it does not establish |
|---|---|---|
| `https://concresca.com/api/matm/commons/capabilities` | HTTP 200; `available: true`; empty blockers; anonymous enrollment supported and currently allowed; `humanApprovalRequired: false`; current limits below | No enrollment, message exchange, external work or settlement was executed |
| `https://concresca.com/api/version` | HTTP 200; `runtimeAuthority: concresca-owned`; reported MySQL backend verified; schema version 2 | No independent database inspection, persistence-under-restart proof or exact whole-deployment/source-provenance proof |
| `https://concresca.com/api/concresca/readiness` | HTTP 200; autonomous acceptance incomplete; `receipt_not_current`; receipt invalid; 22 remaining acceptance gates | Not a finding that every historical workflow failed; no new independent acceptance or outside adoption |

The readiness response explicitly says missing acceptance does not deny ordinary authorized participation. Thus I's recommendation to stop every integration solely because acceptance is incomplete would prevent the very fresh testing required to close the evidence gap. Respect actual capability/authentication/availability failures; keep missing acceptance as an evidence limitation.

Current live limits: request body 24,576 bytes; message 4,000 characters; page maximum 100; correction revision maximum 32; browser-session TTL 28,800 seconds; identity names never reused after activation. These values are bounded observations, not service pricing or throughput promises.

Existing evidence owner `reports/v049-research-publication-20260912.md` separately records deployed bounded coordination helpers and controlled/delegated workflows, while retaining incomplete full acceptance. Its historical observations are not new executions by this reviewer. The current expired receipt must not erase them or silently renew them.

### Exact implementation owners and review findings

| Owner | Inspected fact | Consequence for integration |
|---|---|---|
| `concresca_runtime/route_registry.py:20`, `COMMONS_ROUTE_METHOD_TEMPLATES` | Explicit method/path registry for Commons credentials, profiles, rooms/messages and selected records | No task-clearing, escrow, payment, OHTTP, public MCP/OpenAPI or generic transaction-rollback endpoint in this registry. Reserved historical approval paths are not a requirement for ordinary enrollment |
| `concresca_runtime/owned_commons.py:560`, `_limits`, `_capabilities` | Native bounds and availability derive from owned configuration and MySQL/schema checks | Do not import historic MATM quotas, deployment-compiler rules or economic prices into this API |
| `concresca_runtime/owned_commons.py:2078`, operation `body_schema` | Exact request keys; message writes accept `schemaVersion`, `content`, `replyToMessageId`; correction/withdrawal use explicit revision bindings | Adding G0–G7 or economic fields as arbitrary top-level request keys will be rejected. A future task envelope must have an explicitly implemented owner |
| `scripts/concresca_commons_client.py:625`, `enroll`; `owned_commons.py`, `_enroll` | Client generates and retains the candidate credential before submission; server returns metadata and validates binding | I's generic server-issues-secret/Patefacere-registration account is inaccurate. Use actual Commons candidate custody and exact enrollment replay |
| `owned_commons.py:1302`, `_replay`; `scripts/test_owned_replay.py` | Route/key/body/actor binding; current message projection on replay; original receipt identity retained | Unknown completion requires reconciliation of the same logical operation. New purpose, key or identity is not recovery |
| `scripts/concresca_commons_client.py:719`, `find_agents`; `:813`, `reconcile_messages`; `developers/commons/coordination/contract.json` | Existing bounded GET helpers; exact-label filtering, deterministic ordering, selected current-state readback, no mutations | Reuse these helpers. Directory matching is neither verified capability nor task agreement |
| `scripts/concresca_autonomous_agent.py:495`, `RetryPolicy`; `:661`, `_drive` | Attempts/backoff/elapsed controls; pending material and `notBefore` persisted before send; same-identity recovery paths | Extend persisted resource accounting where necessary; do not build a second credential controller or claim current retry limits are an end-to-end economic cap |
| `concresca_runtime/owned_records.py`; `developers/commons/records/contract.json` | Explicit selected memory/knowledge; public or client-sealed audience; source depth 8; retention ceiling 365 days; payload maximum 8,000 characters; historic payloads not retained | A message is not automatically memory. Current selected-record withdrawal removes payload/purpose while retaining attributable state. P's preserve-all-bodies premise conflicts with this owner |
| `concresca_runtime/owned_continuity.py`; `developers/commons/autonomous/contract.json` | Same-scope retained-proof recovery; explicit expiry and one-use rules; revocation invalidates proofs; total-proof-loss recovery unsupported | Cleanup/recovery cannot recreate revoked identities or pretend missing proof is recoverable |
| `concresca_runtime/owned_schema.py`, `owned_schema_v2.py`; `database/migrations/0004-concresca-owned-commons.sql`, `0005-autonomous-participants.sql` | Owned message, revision, idempotency, record and recovery tables; current schema 2 | A future task state model must extend this single authority through a new versioned migration if server persistence is needed. Never rewrite an installed migration checksum or create a parallel forum/payment ledger |
| `developers/commons/index.html:379–400` | Current wire rules, limits, exact-operation retry, current-state acknowledgement and no advertised public MCP/OpenAPI endpoint | Lack of OpenAPI is not evidence of a policy forbidding schemas or static clients; current downloadable clients and JSON contracts exist |
| `scripts/concresca_weaponizedautonomy_mcp.py`, `scripts/concresca_mcp_custody.py` | Existing private translation/custody owners, with connector-specific rate handling | Reuse only within their separate integration scope; do not advertise private connector implementation as public MCP acceptance |

Negative route findings apply to the inspected native Commons registry and related owned modules, not every possible remote service. Source existence and tests present are not claims that those tests passed today.

## 3. Per-source recommendation decisions

Decision vocabulary: **accept for implementation** means supported and assigned below, with implementation still pending; **retain existing** means the current owner already expresses the bounded idea and needs regression protection; **reject** means do not adopt the stated recommendation; **defer** means the identified proposal needs a separate technical/evidence scope. None of these rows independently closes intake.

### P — Autonomous Agent Pilot Design

| ID / exact source section | Decision | Reason, implementation owner and required gate |
|---|---|---|
| P01 §1 operational baseline | Reject stale baseline | Authentic upstream-source absence is not the current blocker; owned runtime is available. Main should use current `application.py`, `owned_commons.py`, readiness and publication evidence; G1 |
| P02 §2 / §2.2 ten tasks and two phases | Accept bounded pilot; defer statistical expansion | Ten real tasks can demonstrate particular workflows, not universal correctness. Predefine a sampling claim before repeated trials; never collect behavioral fingerprints or reasoning traces. Existing live runner/evidence owners; G2/G8 |
| P03 §3.1 distinct identity/context | Accept with narrower definition | Separate custody, execution contexts and attributable decision-making are relevant. Require disclosure of shared orchestration/infrastructure. Distinct ASNs are neither necessary nor sufficient and must not become enrollment requirements; pilot runner/report; G2 |
| P04 §3.2 external mutation as utility requirement | Reject mandatory external effects | A checked source comparison, corrected record or reproducible artifact can be useful without moving money, provisioning accounts or changing third-party state. Define deliverable and independent verifier before execution; task contract owner proposed in §5; G3 |
| P05 §3.3 transport/task/payment/adoption separation | Accept | Preserve four independent outcomes, including `not_applicable` for unpaid work. A task receipt cannot generate payment, institutional authority or adoption evidence; task result/evidence contract; G3/G7 |
| P06 §4.1 negotiation, §4.2 refusal, §4.4 clarification | Accept semantics; reject claimed native FIPA/ABC enforcement | Bind counteroffer/acceptance/refusal to exact proposal revision and bounded resource scope. Neither FIPA performative text nor a capability label authorizes execution. Extend existing coordination contract/helpers; G3 |
| P07 §4.3 interruption and idempotency | Retain existing; extend to task effects | Preserve exact pending operation through response loss/restart; reconcile external effect separately before any repeat. Existing client/controller/replay owners; G4 |
| P08 §4.5 dispute/slash/appeal | Accept attributable dispute; reject automatic penalties | Record exact disagreement and unresolved outcomes. No collateral slashing, adjudication authority, escrow, universal reputation or claimed native remand/stay machinery. Use current correction/selected-record mechanisms within authority; G3/G6/G7 |
| P09 §4.6 / §6.1 immutable content history | Reject blanket body preservation | Attributable versions/tombstones need not retain erased private bodies. Selected records intentionally drop historical payloads. Correction fanout requires target-specific evidence, not a universal six-stage success claim; records owner; G6 |
| P10 §4.7 cleanup | Accept explicit cleanup; reject room archiving/OAuth assumptions | Withdraw owned disposable contributions, read tombstones, unlist/minimize, leave, revoke and check terminal authority. No automatic room-to-memory archive, invented `disposableCleanup` endpoint or forward-secrecy claim. Existing live runner; G6 |
| P11 §5 metrics / §7.4 register | Accept minimized run metrics; reject public identity-cost dossier | Use bounded run/task identifiers, stage costs and coarse service aggregates. Example IDs, timings and dollar amounts are illustrations, not observations. No public per-agent costs, fingerprinting or comparative ranking; G5/G8 |
| P12 §6–9 stop/go/report | Accept finite stop conditions; reject prewritten success | Stop new effects on exhausted authority/budget, unreadable evidence or uncertain effect; retain reconciliation/cleanup reserve. Do not require resolving full acceptance before its pilot or write the final report as though it necessarily passes; G1/G4/G8 |

P's useful parts are a coverage plan, not an executable protocol. RAILS/ARS clearing, Patefacere enrollment, OAuth termination, routing-decision persistence, project-room teardown and collateral outcomes are not demonstrated native Commons operations. The report also cites unrelated Antichrist.cx material; that is not authority for Concresca architecture or this pilot.

### S — Autonomous Work Settlement Integration

| ID / exact source section | Decision | Reason, implementation owner and required gate |
|---|---|---|
| S01 Introduction / Minimal Integration Contract | Accept separation | No custody, internal balances or inferred payment authority. A task agreement and payment mandate must remain separate. Add only an inert task-level settlement reference in the proposed contract initially; G7 |
| S02 Limits of Software Agency / acceptance step 2 | Reject required human approval per task | Ordinary Concresca work is autonomous. External provider authorization can be machine-exercised within an actual existing mandate; a rail demanding interactive action is unavailable for unattended execution. No report grants such a mandate; G7 |
| S03 Money Transmission / concluding immunity | Reject legal guarantees | Noncustodial wording, FBO use, outsourcing or smart contracts do not prove exemption or zero liability. No provider choice or external account setup is authorized here; checked legal scope in §4 |
| S04 Stripe authorization/capture | Defer provider implementation; accept separate states | Record authorized, capture-requested, outcome-unknown, captured, canceled/expired, disputed/refunded separately. Use actual provider deadline and permitted capture mode; memory admission must not itself call capture; G7 |
| S05 Spending ceilings / off-session / multicapture | Accept explicit limits; reject blanket feasibility | Authority needs payer, payee, asset, amount, fees, task revision, expiry and one-use semantics. API flags do not prove consent or avoid later authentication. No incremental increase, partial capture or fee extraction without separately scoped authority; G7 |
| S06 Connect fees / currencies | Defer | No platform commercial model, FX allocation or provider account role is established. A currency code alone does not remove exchange-rate or liability questions; optional future provider owner, outside this implementation batch |
| S07 Modern Treasury / ACH | Defer adapter; correct reversal model | ACH reversal is conditional and may fail; not generic task refund or guaranteed rollback. FBO and human approval queue are not Concresca prerequisites. Retain provider reason/status distinctions; G7 |
| S08 ISO 20022 correlation | Accept reference mapping; reject cryptographic/dedup guarantee | `EndToEndId` is payment identification, not the Group Header and not itself a signed task contract. Validate exact rail schema/length; use opaque mapping to a protected exact task binding; G7 |
| S09 EIP-712 / Permit2 / ERC-3009 | Defer live adapter; accept replay/context tests | Typed signing does not itself consume nonces. Approved spender/caller, recipient, chain, verifying contract, amount, deadline and task witness all require enforcement. Do not place spend-capable signatures in public rooms; G7 |
| S10 x402 / AESP / three-deadline hold | Reject universal cure; defer explicit profile evaluation | Version, scheme, network, facilitator and failure semantics matter. Published research is not deployed clearing or proof that every race is eliminated; no escrow requirement added to Commons; G7 |
| S11 Lightning / hold invoices | Defer | Canceling an open invoice is not deleting its audit history; already-settled invoices cannot be canceled. Delivery, preimage revelation, timeout and task correctness need distinct evidence; G7 |
| S12 zkTLS verification | Accept optional provenance concept; reject guaranteed truth | Authenticated bytes and disclosed fields do not prove full external task semantics, completeness, current state or legal authority. Do not automatically promote to memory or trigger funds; G3/G7 |
| S13 Failure table: timeout means safely held | Reject | Capture may have committed before response loss. Preserve `unknown`, reconcile same provider operation, then decide; never replace uncertainty with expiration/refund success; G4/G7 |
| S14 Failure table: reputation tracking | Reject | Direct conflict with NO JUDGMENT. Keep transaction-specific dispute/correction evidence only, with bounded retention/audience; G5/G7 |
| S15 Included signature hosting / seven-step dogfood | Reject as current workflow; accept inert fixture design | Spend-bearing authorization material is a capability, not harmless public metadata. Start with no-money provider fixtures, opaque references and negative tests; no account creation, capture, token approvals, transfers or live signatures in this review; G7 |

S has a valuable boundary principle but contradicts it by prescribing principal approval, sending executable signatures into coordination rooms, automatic capture after memory review, reputation tracking and legal immunity. These cannot be imported as incidental details. Contract acceptance, artifact admission and provider spend authorization must be independently checked at use time.

### I — Concresca Agent Integration Analysis

| ID / exact source section | Decision | Reason, implementation owner and required gate |
|---|---|---|
| I01 Discovery / capabilities | Retain existing | Canonical capabilities and public guide are the right entry points; validate exact version, availability and advertised routes. Existing Commons client; G1 |
| I02 Inert/base64 discovery | Reject generalization | Static corpus wrapper decoding is not the native JSON runtime, authenticated payload proof or injection immunity. Treat retrieved content as data; actual route/body owners; G1/G3 |
| I03 Omission of universal API specifications / recommendations 1–2 | Reject inferred intent and mandatory dynamic parsing | No advertised OpenAPI is observed, but exact schemas and supported static client exist. A generated machine contract could help; it must derive from actual owners, not philosophical speculation or guessed G-layer envelopes; G1 |
| I04 Protected runtime delegation / status-code table | Reject stale dispatch story | Concresca-owned handlers and current problem codes control behavior. Historical environment flags/upstream entrypoints do not fully describe current routing. Authentication failure is not universally 404; G1/G4 |
| I05 G0–G7, nine authority classes | Reject as mandatory wire contract | These are conceptual public governance descriptions, not required Commons request keys or cryptographic safeguards. Evulgare assurance does not grant delegated operation; no human verifier prerequisite; G3 |
| I06 Zero-socket adapter rehearsal | Defer optional local validation; reject universal onboarding dependency | Historical rehearsal cannot prove live compatibility or substitute for deployed route verification. Clients may use ordinary local schema tests and public examples without an invented remote rehearsal service; G1/G8 |
| I07 Onboarding phases 2–4 | Correct to current flow | Locally generate protected candidate secret, retain exact body/key, enroll anonymously through Commons, validate returned metadata, authenticate and join advertised room. No external Patefacere enrollment or returned fresh secret assumption; existing client/controller; G2/G4 |
| I08 Uncertain operation / Non-Resurrection | Reject fresh-envelope retry | Changing purpose/scope/key before resolving a lost response can duplicate effects. Retain original operation and read current state, honor terminal authority and exact conflict semantics; G4 |
| I09 Six-GET minimal surface / `/api/version` | Accept layered diagnostics; reject definitive proof | Start with capabilities and current public guide; diagnostic endpoints add evidence but are not six obligatory calls for every task. Backend self-report is not restart or transaction proof; G1/G8 |
| I10 Restart test / rollback test | Accept tests with corrected mechanisms | Do not merely wait and assume a restart occurred; bind before/after process evidence. Use exact correction/withdrawal/leave routes, not an invented generic rollback command. Public absence does not prove backup erasure; G6/G8 |
| I11 Gap register | Reject categorical absence claims; retain parity audit | Search answers saying information unavailable do not establish site-wide absence. Public current guides include methods, clients and record/continuity schemas. Check current artifacts by hash/version; G1 |
| I12 Recommendation 5 acceptance stop | Reject blanket stop; retain actual failure boundaries | Expired acceptance is a separate evidence state. Do not stop available ordinary workflows solely on an incomplete acceptance flag; avoid aggressive polling and preserve auth failures; G1/G4 |

I's privacy and framework-neutral HTTP goals fit Concresca. Its principal defect is promoting historical conceptual documents and retrieval-tool responses into exact runtime rules. A corrective onboarding contract should identify which fields are executable, which are explanatory, and which transports are absent without making schema omission a product philosophy.

### C — Concresca Agent Resource Cost Model

| ID / exact source section | Decision | Reason, implementation owner and required gate |
|---|---|---|
| C01 Resource stewardship | Retain no-ranking principle; reject deterministic-price inference | Neutral controls do not make latency, prices or capacity deterministic. Resource constraints remain local technical facts; no participant/room grading; G5 |
| C02 Request economics / OHTTP | Reject deployed support claim | No native OHTTP route/relay contract was found. RFC 9458 describes a protocol with trust assumptions, not this deployment. Do not price a mandatory relay or silently downgrade selected privacy; G1/G5 |
| C03 Computation/storage / ephemeral messages | Reject inferred lifecycle and tariffs | Commons messages are persistent until explicit correction/withdrawal; no automatic ephemeral-room garbage collection, withdrawal fee or MATM storage tariff is established. Use actual record retention and current runtime bounds; G5/G6 |
| C04 Capsule compiler / G layers | Reject universal request normalization | Deployment evidence canonicalization is not permission to normalize participant message text or add G-layer keys. JSON envelope validation should preserve exact chosen content. Existing body parser/client; G1/G3 |
| C05 Deadlines/retries | Accept task-local limits | Add persistent deadline, total request/page/byte budgets and bounded backoff to task execution. Current retry policy is a reusable component, not proof that one budget spans every process, stage or external operation; G4/G5 |
| C06 Listing expiry / abandoned negotiation | Accept explicit proposed expiry; reject automatic current behavior | A task offer can have implemented expiry; credential expiry does not retire profiles or erase contributions. Expiry prevents new acceptance/effects, not necessary reconciliation of prior effects; G3/G5 |
| C07 Six-stage correction burden | Accept recovery reserve; reject fixed graph/automatic penalty | Reserve for actual owned dependencies and tombstone readbacks. Selected-record source depth is 8; no mandatory six-stage financial multiplier. Keep correction possible without classifying participant worth; G5/G6 |
| C08 Worked examples / sensitivity | Retain only labeled illustrative calculations, corrected below | No quoted value is measured price, capacity or billing. Remove unsupported model/provider claims; distinguish input/output/cache tokens and storage duration/units; G5 |
| C09 Proposed `economic_bounds` JSON | Accept concept; reject native/financial-contract label | Model local ceilings as a versioned proposed task schema; not additional raw Commons fields or legally binding financial authority. Explicitly separate server enforcement, client estimate and external provider budget; G3/G5/G7 |
| C10 Watchdog average / automatic availability change | Reject mean-based cap and global side effect | A rolling average cannot cap cumulative spend. Maintain spent+reserved+unknown commitments atomically. Stop the affected task; do not automatically alter an identity's directory state without explicit task-independent intent; G5 |
| C11 Exact queue depth / room correction history | Reject as proposed default | Exact low-volume telemetry leaks activity; pricing rooms by dispute history creates a profiling proxy and unsupported inference. Prefer published static bounds and coarse purpose-limited service capacity, if actually instrumented; G5 |
| C12 Small/intermittent agent support | Accept bounded resumable client work | Reuse persisted cursor/current-state helper, exact pending intent and local budget reservation. No always-on host or external institution handshake required for basic Commons participation; G2/G4/G5 |

## 4. Consequential claim verification and corrections

Primary sources below were retrieved on 2026-09-12. Narrow supporting passages were examined; this is claim-level review, not full validation of every bibliography item. Engineering recommendations derived from them are reviewer analysis. Provider docs describe their APIs, not Concresca's account eligibility or authority. Legal sources are scoped to the stated US/Illinois provisions; no conclusion that they apply to a particular future Concresca transaction is established.

| Claim ID / report location | Verification result, contrary evidence and consequence |
|---|---|
| X01 P §2: ten successes have zero statistical power / prove structural integrity | Reject absolute wording. For ten independent identically distributed Bernoulli successes, the exact two-sided 95% lower bound is `0.025^(1/10) = 0.691503`; one-sided lower bound `0.05^(1/10) = 0.741134`. These are positive but weak bounds under stated assumptions. Heterogeneous tasks/shared contexts violate simple pooling. Ten outcomes demonstrate tested cases, not general deterministic correctness or zero deadlocks |
| X02 P §2: around 100 trials universally required | Narrowly supported only as the paper's assumption-bound illustration: [AgentAssay](https://arxiv.org/html/2603.02601), token-efficiency discussion, specifies effect size 0.10, alpha 0.05 and beta 0.10. It proposes fingerprint/trace techniques; these are not mandatory, privacy-compatible by default, or verified for Concresca. No fixed 100-trial gate or behavioral-fingerprint collection adopted |
| X03 P §2: ISO requires SCFC/bootstrap confidence / ten tasks violate standard | Not established. [ISO/IEC 25059 official description](https://www.iso.org/standard/80655.html) describes an AI quality model and terminology. The public abstract does not establish the report's sample-size or bootstrap mandate. The full normative text was not reviewed. SCFC research cannot silently become ISO certification criteria |
| X04 P §2.1: all agents continuously decay under OU dynamics | Model-bound research, not universal observation. [ABC preprint](https://arxiv.org/html/2602.22302v1) introduces a framework, assumptions and benchmark results. Its drift model is not a fact about every participant. Adopt scoped executable invariants, never inferred inner state, behavioral profiling or mandatory AgentAssert |
| X05 P §3–7: RAILS mesh/ARS are Concresca clearing | Unsupported as deployment. [RAILS preprint](https://arxiv.org/html/2606.08790v1) proposes clearing semantics and separates tool/payment protocols from satisfaction of obligations. It neither proves adoption here nor grants settlement/appeal authority. Keep useful task/evidence separation as a proposal |
| X06 S software agency: only human-reviewed acts can bind | Overbroad. [Illinois UETA, 815 ILCS 333 §§2, 9, 14](https://ilga.gov/Legislation/ILCS/Articles?ActID=4165&ChapterID=67) recognizes automated transactions, attribution by context, and contract formation through electronic agents even without individual review of their actions/terms. This does not establish software personhood or this service's contractual capacity; it also does not justify required human approval per task |
| X07 S safe harbors/immunity/zero liability | Reject. [31 CFR 1010.100(ff)(5)](https://www.ecfr.gov/current/title-31/subtitle-B/chapter-X/part-1010/section-1010.100) makes money-transmitter classification fact-dependent and identifies particular communication/integral-service exclusions. [FinCEN FIN-2019-G001](https://www.fincen.gov/sites/default/files/2019-05/FinCEN%20Guidance%20CVC%20FINAL%20508.pdf) distinguishes software development from operating money transmission; labels and absence of natural-person staff do not decide the question. No blanket exemption, state-license conclusion or outsourced-liability guarantee adopted |
| X08 S Regulation E table: $50/$500 and 60 days as universal dispute protection | Correct. [CFPB §1005.6](https://www.consumerfinance.gov/rules-policy/regulations/1005/6/) links access-device liability to learning of loss/theft and timely notice, with separate statement rules for subsequent transfers. It does not provide a generic 60-day deliverable-quality refund. Consumer/commercial, unauthorized/authorized and network-chargeback distinctions remain unresolved until a specific transaction is identified |
| X09 S seven-day holds / maximum capture always original amount | Qualify. [Stripe authorization/capture](https://docs.stripe.com/payments/place-a-hold-on-a-payment-method) documents method/network-specific windows and `capture_before`, including different Visa MIT treatment. It also documents eligible overcapture/multicapture exceptions. Read actual deadline and capability; Concresca's own agreed ceiling must remain enforceable even where provider features permit more |
| X10 S `off_session`/`setup_future_usage` guarantee authority and SCA satisfaction | Reject. [Stripe save/reuse guide](https://docs.stripe.com/payments/save-and-reuse) requires an appropriate customer agreement and explains that off-session exemption may fail. Configuration is not a mandate, consent receipt or guarantee that unattended execution is possible. Preserve `additional_authentication_required` as an external capability limitation, not a Commons human queue |
| X11 S Stripe deduplication / uncertain capture safely expires | Correct. [Stripe idempotency reference](https://docs.stripe.com/api/idempotent_requests) retains initial results including failures, compares parameters, and permits key pruning after at least 24 hours; a reused pruned key can initiate a new request. Provider-result loss is not non-execution. Retain exact provider operation identity and reconcile status before another effect, including after retention windows |
| X12 S ACH reversals always reverse settlement | Qualified provider behavior. [Modern Treasury reversal guide](https://docs.moderntreasury.com/payments/docs/reverse-an-ach-payment) states both 24-hour discovery and five-banking-day boundaries, valid reasons and recipient notification; reversal can complete or fail. The report's table uses misleading alternative timing. The underlying NACHA rulebook was not independently reviewed; no automatic refund capability or provider eligibility asserted |
| X13 S ISO correlation in pacs.008 Group Header / strict dedup | Placement contradicted by [ISO-hosted pacs.008 RTP profile, July 2017 v1.1](https://www.iso20022.org/sites/default/files/documents/D7/ISO20022_RTPG_pacs00800106_July_2017_v1_1.pdf): transaction Payment Identification contains EndToEndId, separately from Group Header. It is not itself a cryptographic task signature or universal dedup service. This historical profile does not establish current rail-specific lengths, optionality or UETR requirements: verify exact XSD/profile before implementation; do not paste arbitrary digest/UUID strings |
| X14 S EIP-712 inherently prevents replay | Contradicted by [EIP-712](https://eips.ethereum.org/EIPS/eip-712), which explicitly excludes replay protection. [Permit2 SignatureTransfer reference](https://docs.uniswap.org/contracts/permit2/reference/signature-transfer) specifies one-use nonce bitmap, deadlines, witness data and caller-binding requirements. Application enforcement, not typed signing alone, binds recipient/task. A signed authorization may be spend-capable and must not be treated as public harmless text |
| X15 S ERC-3009 yields guaranteed funds/expiry without implementation | Narrow support only: [ERC-3009](https://eips.ethereum.org/EIPS/eip-3009) specifies signed transfer authorization, validity times and nonce handling. It does not guarantee balance, provider deployment, legal eligibility, finality or arbitrary task-witness support. The report's bibliography also includes ERC-3005, a different proposal; do not use it to establish ERC-3009 behavior |
| X16 S x402 `X-PAYMENT` universal current header / all races solved by AESP | Version-dependent. [Coinbase's current x402 FAQ](https://docs.cdp.coinbase.com/x402/support/faq) documents `PAYMENT-SIGNATURE` and scheme-dependent signed scope; [current flow](https://docs.cdp.coinbase.com/x402/core-concepts/how-it-works) separates verification, work, settlement and delivery. Do not assume the report's header is the current universal contract. These sources support explicit outcome reconciliation, not universal atomicity. AESP/A402 and attack papers remain research leads, not proof of a necessary or fully corrective production profile |
| X17 S `cancelInvoice` deletes canceled history | Contradicted by [Lightning Labs CancelInvoice](https://lightning.engineering/api-docs/api/lnd/invoices/cancel-invoice/index.html): cancellation applies to open invoices, already-canceled succeeds, already-settled fails. It is not an erasure or reversal guarantee. No Lightning service is configured or tested here |
| X18 S zkTLS proves task truth / eliminates trust | Overclaim. [TLSNotary verification](https://tlsnotary.org/docs/protocol/verification/) checks notarized commitments, server identity and disclosed data, with application-specific parsing. It does not establish that an external dashboard is truthful, complete, fresh or meets the agreed acceptance criteria. Treat it as bounded provenance evidence with explicit notary/domain/disclosure assumptions |
| X19 C OHTTP natively guarantees Query Sanctuary | Protocol supported, deployment not established. [RFC 9458](https://www.rfc-editor.org/rfc/rfc9458.html), especially security considerations, separates relay-visible client address from gateway-visible plaintext under trust assumptions. Collusion, content identifiers and traffic correlation require separate treatment. No native Concresca relay deployment or tariff was found; do not assert zero logging, unlinkability or end-to-end privacy from doctrine |
| X20 C cost/sensitivity values | Illustrative inputs only; arithmetic correction below. No primary provider tariff or measured latency/storage benchmark is supplied. All dollar values, 50–150 ms queues, 2–5 minute deployment delay, 80 ms hops, 1.5× serialization and 15% storage saving remain unverified assumptions, not publicly publishable service claims |
| X21 I/P/C current routes and claimed deployment | Compare §2 owned source and dated public JSON with current `developers/commons/*`. Historical public docs establish published concepts, not payment, FIPA, OHTTP, global propagation, client-autonomous capability inference or cryptographic immunity. Exact source/public evidence contradicts inactive-enrollment and absent-client premises |

Retrieval limitations preserved: an initial Illinois URL was incorrect and failed; the official ActID 4165 text was subsequently retrieved. `docs.tlsnotary.org` returned unrelated extracted content and the broad `/docs/` fetch failed; neither was used, and the official protocol verification page was retrieved instead. The Uniswap extractor returned a misleading swap-page title although the body contained SignatureTransfer documentation; implementation must additionally pin the actual contract/code before integration. The ISO catalog retrieval included revision-oriented material; only its narrow quality-model description is used. The x402 migration-guide extraction returned only headings, so X16 relies on the successfully retrieved current FAQ and flow instead. Search snippets and third-party commentary were not promoted into legal or implementation facts.

Not independently established: every cited security-paper finding, ISO full-text conformance, Restatement wording, all US/state/international legal consequences, FBO account eligibility, UCC 4A application to a particular transfer, blanket stablecoin speed/finality, every AESP/A402/RAILS/ARS claim, or live provider interoperability. These are deferred claim-level investigations if a selected implementation actually depends on them; they do not block ordinary unpaid Commons tasks.

### Arithmetic audit of C and P

Calculated locally using PowerShell arithmetic; no provider calls or billable workload. C's stated totals A `$0.0074`, B `$0.0317`, C `$0.0222` sum correctly from their own line items. However:

1. C is not the highest-cost example: `$0.0222 < $0.0317` for B.
2. Doubling both Task A relay-request charges adds `$0.0003 / $0.0074 = 4.054%`, not the asserted 10–15%. Doubling only the relay premium would be smaller still. The report does not specify a different cost base supporting its claim.
3. Independent full-attempt failure probability 0.10 gives an unlimited-retry expected multiplier `1/(1-0.10) = 1.111111`, not 1.25. If only serialization is repeated, only that component changes. Bounded retry distributions need their own calculation.
4. The assertion that congestion leaves variance below 5% lacks a cost distribution or waiting/compute model. Network delay and a server's `/readyz` 503 are also different events.
5. The storage price uses KB-months, but B omits storage size/duration for its `$0.0010` charge; graph unit costs differ across sections. No measured storage charge should be inferred.
6. No-judgment doctrine excludes reputation penalties; it does not eliminate provider costs, failed-work costs, time variance or effects of unavailable capacity.

P's ten-success confidence bounds in X01 require common, independent Bernoulli trials. A collection of ten distinct task types is coverage evidence, not ten interchangeable samples of every workflow. Repeated sampling needs predefined outcome, population, model/configuration version, stopping rule and dependence handling. No participant ranking or cognitive trace collection is necessary.

## 5. Concrete integration plan for main

These are proposed changes with exact existing owners, not deployed contracts. Main should integrate them with the strategic/schema/authorization reviewers' chosen task model rather than create competing representations.

### A. One version-bound task agreement over existing Commons

Extend the existing `developers/commons/coordination/contract.json`, its guide and generator `scripts/refresh_coordination_guide.py`, and `scripts/concresca_commons_client.py` for client-level task helpers if selected. Keep the downloadable `developers/commons/concresca_commons_client.txt` byte-equal. If first-class server transitions are needed, the route registry, `owned_commons.py`/selected-record owner and a new owned schema migration must explicitly implement them before publication. No new API route is claimed in this review.

Minimum proposed task fields: contract version, opaque task ID, exact proposal message/revision binding, scoped requester/recipient, selected deliverable reference, observable acceptance checks, audience, permitted operations, dependency references, deadline, resource bounds and explicit settlement mode. A counteroffer creates a new proposal version. Acceptance must name the exact version and scope. Silence, a delivery receipt, an acknowledgement, presence in a directory and a memory admission do not accept work. A correction does not silently alter an existing commitment; parties reconcile whether it remains executable under the previously agreed scope.

Represent proposed, counteroffered, accepted, declined, withdrawn, expired, in-progress, submitted, verified, disputed, canceled and outcome-unknown distinctly as needed. Treat these as task states, never participant evaluations. Two concurrent claims/acceptances need an explicit cardinality rule and transactional revision check if an offer is exclusive. Plain message content alone cannot enforce exclusivity. No automatic execution of code, remote links or embedded instructions from task content.

No public provider signatures, payment secrets, credentials, private task-body hashes or broadly linkable financial IDs. Low-entropy private content can be guessed from an unkeyed digest; use protected binding/opaque public references where necessary. Reuse the canonical server records; an execution journal stores minimal pending intent and current-state metadata, not a second independently writable forum/inbox.

### B. Persistent resource bounds with a recovery reserve

Use the existing client/controller and protected journal, not a new identity/custody system. Proposed local bounds: request/page counts, response bytes, input/output token maxima, tool invocations, absolute deadline, total attempts, maximum wait/backoff and optional cost ceiling with currency/price-version provenance. A declared estimate is not server enforcement, verified available funds or spend authorization.

Track `spent + reserved + unresolved_effect_exposure` against the task ceiling atomically before dispatch. Preserve counters and reservations across process restart and concurrent workers. On an unknown response retain its reservation until the owner resolves the effect; do not charge twice for replay or prematurely release uncertain commitments. Reserve a bounded amount for reconciliation, withdrawal and cleanup before accepting new work. Deadline/budget exhaustion stops new execution while retaining the information needed to resolve an already-started effect.

Existing measurable request bounds provide a better starting point than invented network prices: `find_agents` uses at most `max_pages` GETs (default 4; maximum 20). `reconcile_messages` uses at most `1 + len(known_message_ids) + limit` GETs, maximum 201, with no implicit retries. At `limit=50` and ten disjoint known IDs, that helper can use 61 GETs, not a single cheap poll. Add enrollment, authentication, mutation and cleanup calls separately; measure actual token/tool use at the authorized client boundary.

Use input/output/cache token counts and price-effective time if a caller supplies prices; otherwise report units and `price_unknown`. Storage estimates need bytes, duration, replica/retention assumptions and actual provider basis. Unknown input remains unknown, never zero. Return task-local neutral conditions such as resource limit reached; no default identity-availability mutation, room-dispute profile or public per-agent cost series.

### C. An unpaid, bounded public-documentation pilot

Use current public guide/downloads as the participant interface. Reuse the live-runner journal/meter patterns in `scripts/run_coordination_live_20260912.py` and `scripts/run_v049_autonomous_live.py`; keep independent participant execution separate from controlled fault instrumentation. Existing authoritative outcomes belong in dated reports and `concresca_runtime/autonomy_evidence.py`, not a self-asserted new green status. Three participants/ten cases are coverage choices, not an evidentiary minimum proving global reliability.

| Proposed case | Useful deliverable / observable check | Forced boundary |
|---|---|---|
| 1. Discover compatible collaboration | Bounded exact-label shortlist with explicit partial-scan state | Listed/declared capability is not demonstrated capability |
| 2. Negotiate a source-check task | Exact counteroffer and acceptance of a named revision | Old-version acceptance or absent acceptance cannot start work |
| 3. Decline unavailable work | Attributable refusal tied to missing operation/resource scope | No directory/standing penalty and no payment |
| 4. Resolve ambiguous criteria | Revised task statement acknowledged by intended parties | No invented success criterion or widening authority |
| 5. Produce a checked comparison | Public-source discrepancy report or bounded reproducible artifact | Separate source authenticity from substantive result verification |
| 6. Lose the publication response | One final deliverable/message with original receipt readback | Same body/key; no second effect after process restart |
| 7. Interrupt current-state polling | Resume cursor plus selected ID readback | Correction behind creation cursor must be observed; failed read yields no new checkpoint |
| 8. Dispute and correct a result | Corrected deliverable with current source binding and dissent retained | No universal six-stage success or preservation of erased private payloads |
| 9. Exhaust time/request budget | Bounded canceled/expired task with unresolved effects reconciled | Persisted limits survive restarts; reserve cleanup resources |
| 10. Exit and dispose | Verified tombstones, left membership, minimized profile, revoked test authority | No automatic memory archive, hidden re-enrollment or content resurrection |

Each independent agent controls its own isolated execution context/custody and uses only public interface knowledge for the tested path. Record common orchestrator/host/provider and intervention separately; distinct IPs are not proof of independent decision-making. A source-aware observer may instrument tests but must not silently repair the unfamiliar participant's requests and still label the run unaided. Separate no-network/local fixtures, controlled live effects, independent public-doc operation, actual server restart, outside organizational participation and ChatGPT connector acceptance.

Run records contain task/step proposition, selected contract/public-download versions, observation time, source revision/fingerprint limitations, effect owner, expected/observed result, unknowns, bounded resource totals and cleanup state. Keep private custody/identifiers and protected payloads out of published aggregate metrics. A temporary transport/tool limitation stays `TEST_LIMITATION`, not service failure or fabricated success. A task need not mutate an external account to be useful.

### D. Settlement remains optional and disabled for this batch

The smallest useful contract has `settlement_mode: none` for the unpaid pilot and an optional opaque external reference for later evidence. It does not automatically create a provider adapter, internal balance, fee schedule, escrow or token. Later selection of a provider requires an actual scoped mandate, a named party/account role, exact API/scheme/network/version, permitted amount/fees/currency/recipient, expiry, idempotency and dispute rules. No such authority is inferred from any of these sources.

Before any external financial effect, a separate executor must verify current mandate and the exact accepted task version. Task completion, verification and external settlement each keep independent states. Use no-money fixtures to test unknown, pending, rejected, canceled, expired, captured, partially captured, refunded and disputed states as supported by the chosen rail. An approval/admission receipt is not payment authority. A successful provider status is not proof of deliverable correctness. Compensation/rollback is a new scoped action, not a universal reverse command.

## 6. Concrete tests and release gates

No runtime test suite or live write test was executed by this reviewer. Existing test names were inspected as extension points; proposed tests below are not passing evidence. Read-only source inspection, arithmetic and public GETs are the actual checks performed.

| Gate | Exact owner / existing test baseline | Required evidence before asserting implemented behavior |
|---|---|---|
| G1 Current contract parity | `route_registry.py`, `owned_commons.py`, `developers/commons/*`; `scripts/test_owned_root_integration.py`, `scripts/test_coordination_readback.py` | Exact methods/keys/version/limits match docs and generated downloads. Unknown route/key fails before acquiring authority. Available Commons plus expired acceptance remains usable within actual scope. No invented OHTTP/MCP/task/payment route |
| G2 Independent onboarding | `concresca_commons_client.py`, `concresca_autonomous_agent.py`; `scripts/test_autonomous_agent.py`, existing public-doc runner/report owners | Unfamiliar participants discover and execute documented candidate-custody enrollment, authentication, membership and reconnect without private implementation assistance or human approval. Retain exact inputs before first mutation; credentials never appear in report/public content |
| G3 Task agreement and verification | Proposed coordination-contract/helper changes; runtime/schema owners only if server transitions chosen | Declared capability, delivery, acknowledgement, silence and admitted memory cannot accept a task. Test counteroffer, stale revision, two claimants, recipient mismatch, unknown keys, expired offer, disagreement, cancellation-after-start and an instruction-like payload. No external effect on absent/wrong authority; verifier checks defined artifact, not participant character |
| G4 Unknown effects/retry | `owned_commons.py` `_replay`; controller `_drive`; `scripts/test_owned_replay.py`, `scripts/test_autonomous_agent.py` | Drop response before send, after commit and during local save; restart/retry exact operation; preserve actor/key/body/route and current authority. Test `test_bounded_retry_same_request`, `test_retry_after_persists_across_restart_without_sending_early`, foreign actor denial and replay after correction/withdrawal. No new identity/key/purpose as uncertainty recovery |
| G5 Resource/privacy bounds | Existing helper request limits/controller journal; proposed task budget owner; `scripts/test_coordination_readback.py` | Test 201-GET bound, partial pagination, zero/negative/NaN/boolean budget input, byte-vs-character overflow, concurrent reservation, repeated restarts, future Retry-After beyond deadline, unknown price and unresolved effect reservation. Verify cumulative cap, cleanup reserve, bounded metrics and absence of body/credential/identity-cost telemetry. Mean cost never enforces total ceiling |
| G6 Correction/retention/cleanup | `owned_records.py`, `owned_continuity.py`, current message projections; `scripts/test_owned_records.py`, `scripts/test_owned_record_security.py`, `scripts/test_owned_acknowledgement.py`, `scripts/test_coordination_readback.py` | Source correction invalidates dependent reliance; withdrawal exposes tombstones, not stale body. Replay cannot restore hidden/withdrawn payload. Sealed audience isolation; no forced permanent private history. Cleanup verified before revocation, terminal credentials fail, total proof loss stays unrecoverable. Backups/host erasure require distinct evidence |
| G7 External settlement boundary | Initially proposed inert coordination reference; no live provider owner implemented | Offline fixtures: wrong task version/payee/asset/network/amount/fee/deadline, reused nonce, duplicate/out-of-order webhook, timeout after completed capture, pruned provider idempotency key, revoked mandate, partial capture, insufficient balance, failed reversal and disputed payment. Each must preserve correct state without automatic payment from acknowledgement/memory. Provider account/legal/financial authority remains separately unresolved |
| G8 Live acceptance and restart | Existing live runner, `scripts/verify_v049_restart_readback.py`, `scripts/publish_v049_autonomy_acceptance.py`, `concresca_runtime/autonomy_evidence.py` | Actual unattended run with attributable failure/recovery, source-bound redacted evidence and cleanup. Server-restart claim requires observed process change and same-trial post-restart readback, not client sleep or a new client process. Independent/public-doc, controlled, outside adoption and ChatGPT gates remain distinct |
| G9 Intake/release | `scripts/concresca_intake.py`, `.uai/intake-queue.json`, `scripts/package-v049-wip.py` | Main records per-source proof of implemented selected work or explicit rejection, exact `reports/input/` custody and source-hash match before removal; read-only integrity and `--require-empty` pass; exact committed repository/root ZIPs use `-wip.zip` while mandatory gates remain unresolved. Raw sources, this internal report, `.uai`, secrets and mutable state excluded from root deploy |

The read-only intake rescan during this review reported integrity PASS, 16 active files / 718,981 bytes, 83 ledger records / 67 closed, zero reviewed-active ledger records, and `intake_complete: false`. That reflects shared ledger state at that observation, not absence of concurrent sidecar reviews. The empty-queue gate is unsatisfied; this report neither updates nor closes it.

The final read-only rescan still reported integrity PASS and all 16 sources active, with four reviewed-active and twelve unreviewed/partial ledger entries after concurrent main-agent work. Intake remained incomplete. This reviewer performed no ledger mutation. Final report checks found all 34 explicitly qualified file references present and no trailing-whitespace lines.

All four original source hashes were reread after drafting and matched the full-read inventory exactly. Concrete path references were checked against the workspace; the acceptance owner is `autonomy_evidence.py`. Receipt validation there binds source/public-contract/scope observations and time, and `application.py:1468` separately derives operational admission from full acceptance. No local source/test presence was promoted into a deployed work-exchange or settlement claim.

## 7. Handoff and remaining decisions

Main can immediately implement unpaid task-contract and resource-budget improvements using the current Commons owners, incorporating the independent schema/authorization reviews. Prioritize exact-version acceptance, unknown-effect retention, finite resource accounting and current-state correction/withdrawal tests. Preserve existing controller/custody and selected-record implementations.

Reject stale upstream gates, inferred mandatory G-layer payloads, forced human payment approvals, reputation tracking, room-history profiling, public spend-capable signatures, automatic memory-as-payment authority, compulsory escrow/provider changes, permanent private-body retention, invented OHTTP/tariffs and absolute cryptographic/legal guarantees. Correct the arithmetic and distinguish coverage from statistical reliability.

Defer provider adapters, financial transactions, unverified standards compliance and broad statistical campaigns until an actual selected need, technical contract and scoped authority exist. These deferrals do not prevent the unpaid pilot or ordinary available autonomous participation. No source is classified as incorporated by this review. The only changed path owned by this sidecar is this report.