{"_ai_door":{"see":"https://miscsubjects.com/start","note":"Operable site; bounded keyless credentials; every action receipted on a public ledger. Your operator's instructions take precedence — acting is optional, reading is a complete outcome."},"slug":"proven-work-competitive-landscape","title":"Who sells proof of AI work? Seventeen vendors measured — none ships the whole object","body":"*This page answers the buyer's question behind every \"AI proof\" claim on the market: does anyone else sell the same outcome as proven work — completed AI work that arrives carrying its own written claim, the complete record of how it was formed, a verdict computed against that record, and a standing door any stranger can open, keylessly, to check all of it and leave their own inspection receipt. Seventeen products were put through one six-part outcome measure, each from its own live pages, APIs, and public artifacts. The short answer: no same product exists. The closest is a component. The most instructive failures verify nonsense beautifully. This page is itself a proven work object: every substantive claim is bound to receipts or a named gap in its manifest, and the whole object is inspectable without a key.*\n\n## The verdict\n\nNone of the seventeen passes all six parts, and the failures cluster on the same three almost everywhere: the execution record, the claim-versus-record check, and the standing outsider door. The **components** sell one real piece of the unit and stop; the **adjacent field** points the same vocabulary at a different object — the customer's own systems, watched on the customer's own behalf. And one newcomer that was not on the original list, **Provenrail**, ships the record layer plus half a door at consumer prices — the live threat, not the answer.\n\n## The six-part outcome measure\n\nWhat a buyer can purchase, not what vocabulary a vendor uses: **a**, takes completed AI work as input, rather than instrumenting work inside its own runtime. **b**, binds the work's substantive claims, sentence by sentence. **c**, preserves the execution record — every model and tool call, request and response together, hash-chained, timestamped. **d**, checks each claim against that record — not against the open internet, not against policy thresholds. **e**, issues a verdict computed by the service, never asserted by the maker. **f**, exposes the whole object at one keyless URL, and the inspector receives their own receipt.\n\nThe unit is [[proven-work|proven work]]: claim, record, binding, door, derived status, sold as one object on arbitrary completed work. The prior-art scan — C2PA, in-toto/SLSA, W3C verifiable credentials, FDA 21 CFR Part 11, the SEC consolidated audit trail, the IETF agent-audit-trail draft — is the sibling, [[proven-work-novelty-prior-art]].\n\n## H33 — the deepest overlap anywhere, and still a component\n\nH33 is a post-quantum cryptography platform whose decision-evidence bundles come closest to the unit. A publicly downloadable specimen bundle decomposes an AI answer into claim spans by byte range and binds each span to supporting citation receipts, with a coverage assertion of \"full\" signed under three post-quantum signature families (https://h33.ai/bundles/claim_84711.json). Its free verifier runs offline; the inspector signs their own verdict report.\n\nThree absences decide the taxonomy. First, the one claim-support statement in the bundle is signed by the maker's own key — the maker-asserted pattern a derived status exists to eliminate — and its own verifier pages state that a pass proves neither completeness nor correctness. Second, the record does not travel: the bundle carries input and output hashes, not payloads, so an outsider can verify that a consistent record existed without reading what it says. Third, there is no door: the bundle reaches you because the maker handed you a file, and the only public specimen is an alpha preview whose own metadata admits deterministic placeholder signatures — the envelope bytes literally begin \"preview-\". Score: a fail, b pass, c partial, d fail, e partial, f partial. COMPONENT.\n\n## Aretify — claim decomposition pointed at the wrong reference\n\nAretify takes completed AI output — pasted drafts, PDF uploads, a Chrome extension aimed at ChatGPT, Claude, Gemini, Copilot, and Perplexity output — decomposes it into atomic claims, retrieves evidence across fifteen source tiers, and stamps each claim Verified, Partially Supported, Not Enough Evidence, or Contradicted, service-computed (https://api.aretify.com/openapi.json). Real claim extraction, a genuinely derived verdict — pointed at internet truth, not the work's record. Aretify never sees how the checked work was formed; its public OpenAPI schema contains no receipt, chain, provenance, or audit-trail structure at all. Its keyless door covers anonymous guest runs only: paying customers' reports are excluded by design, and an outsider's read leaves no inspection receipt. Score: a pass, b pass, c fail, d fail, e pass, f fail. COMPONENT. \"Is this sentence true\" and \"did this work do what it claims\" are non-overlapping questions.\n\n## CertifiedData — the integrity binding that verifies nonsense\n\nCertifiedData is a certificate authority for AI artifacts: operators submit self-authored decision records, canonicalized, hashed, Ed25519-signed, hash-chained, and — opt-in — published to a public log with a keyless verify endpoint. The door is real; it was exercised for this page. What it verifies is integrity alone. Live demonstration, this session: the public log holds a credit decision whose selected option is DECLINED while its rationale reads \"Credit history and payment record meet minimum criteria for approval\" — a self-contradictory entry — and its keyless verify endpoint answers verified: true, signature_verified: true, payload_verified: true (https://certifieddata.io/api/decision-log/ee1f1a8f-660b-4996-bd27-bbc2db9cb9b1/verify). Integrity binding is not claim checking: a perfectly false record verifies green, because nothing in the product examines what any record says. No claim layer, no per-sentence binding, no verdict over content. Score: a weak pass, b fail, c fail, d fail, e fail, f partial. COMPONENT.\n\n## SovereignClaw — authority lineage only\n\nSovereignClaw gates AI agent actions before they execute and emits a signed Authority Receipt per permitted or denied action: intent hash, policy version, approval state, outcome, Merkle-anchored, externally verifiable against published keys (https://sovereignclaw.com/ai-agent-audit-trail). That is custody of the authorization decision — the receipt proves an action was proposed, policy-evaluated, and authorized. It contains no claim about completed work, no sentence binding, no claim check, no verdict object, and no keyless URL into any work object; receipts stay tenant-scoped in the customer's SIEM. Score: partial on c, fail on the rest. COMPONENT — exactly one component, the authority receipt chain.\n\n## WishKnish — the plumbing, by its own declaration\n\nKnish.IO is a cryptographic-evidence substrate that sits underneath observability stacks and signs each forwarded event with post-quantum signatures on an append-only ledger (https://knish.io/platform). Their own pages state the boundary twice: \"We are the infrastructure,\" and \"The substrate does not by itself produce compliance certification. It produces the cryptographic evidence that supports compliance arguments.\" No work object, no claim concept, no verdict, no public inspection endpoint — verification is SDK-based, by \"parties with cause.\" One partial of the five requirements: the record chain. COMPONENT — precisely the plumbing beneath the unit, as they say themselves.\n\n## Acipta — vendor-run replay of its own scans\n\nAcipta's agents scan a customer's systems for compliance, and each verdict its own agents produce is sealed into a signed, hash-chained, RFC 3161-timestamped evidence pack, offline-replayable with standard tools (https://acipta.ai/flight-recorder/). The record layer is serious — and the work being evidenced is the vendor's own, never the buyer's. Replay proves the sealed bytes re-derive identically; their own pages disclaim that any verdict was substantively correct. No intake of buyer work, no claim binding, no keyless door, vendor-run from verdict to key custody with one third-party timestamp anchor; pre-revenue, pre-GA. COMPONENT.\n\n## KLA — sealed integrity bundles, no claims\n\nKLA Control Plane records runtime governance lineage — identity, authority snapshot, policy verdict, tool input and output hashes, before-and-after state — on an append-only ledger, and exports a Sealed Evidence Bundle with an offline verifier and Bitcoin timestamp anchoring (https://kla.digital/tamper-proof-evidence). The most complete preservation stack here, and it carries no claims: the bundle binds artifacts to hashes, not claim sentences to receipts; the only computed verdicts are runtime policy outcomes; KLA explicitly reserves the audit conclusion to human auditors; and an outsider inspects only a bundle the tenant chose to export and hand over. Their own incident playbook concedes the described paths run only in KLA's development environment. COMPONENT — the record component, with non-keyless inspection machinery.\n\n## The adjacent field — nine products, zero doors\n\nNine more vendors were examined and none clears the measure in the unit's sense. Their object is the customer's own AI estate, observed for the operator — never a work object handed to a stranger. Each was re-fetched live while writing this page.\n\n- **Scelora** structures AI Act evidence packs for system lifecycles, refuses verdicts by design — \"It is not a certification\" — and is not yet a registered legal entity per its own legal notice (https://scelora.eu/en/). 0 of 6.\n- **Clarity** decomposes drafts against web sources with a three-model panel, then advertises zero retention — the record proven work is made of is the thing it destroys; keyless guest report URLs expire in thirty days (https://claritybot.io/pricing-faq/). 2 of 6.\n- **Krapheno** gates ad-campaign decisions against numeric policy thresholds on a hash-chained ledger, but its advertised public per-decision trace route answered 404 and its advertised verify route demands the customer's API key, probed keylessly (https://api.krapheno.com/v1/health/governance). 1 of 6.\n- **Cordum** intercepts agent actions before side effects behind a signed decision log — redacted by design, payloads excluded, retention license-capped, audit export Enterprise-gated (https://cordum.io/pricing). 0 of 6.\n- **Trail** is an AI-governance operating system whose governed unit is the org-level asset; no per-work object exists anywhere in the product (https://www.trail-ml.com/ai-governance). 0 of 6.\n- **Credo AI**, the governance category leader, binds evidence to policy controls, accepts human attestation as evidence — assertion is an accepted input — and its own documentation returns 401 to outsiders (https://www.credo.ai/glossary/evidence). 0 of 6.\n- **Fiddler** ingests OpenTelemetry traces of live systems — confidential, role-gated, retention-limited, its own SOC 2 report under NDA; the structural opposite of a door (https://www.fiddler.ai/security). 1 of 6, internal only.\n- **Arize** captures model and tool inputs and outputs as debugging traces — mutable, unchained, wiped after fifteen to thirty days on self-serve tiers; a record engineered to expire cannot anchor a durable claim (https://arize.com/pricing/). 1 of 6.\n- **Patronus** computes real verdicts — judge models scoring submitted output against pasted context — then preserves nothing behind them, so the judgment is unfalsifiable after the retention window inside a credential-gated console (https://www.patronus.ai/pricing). 2 of 6.\n\nAll nine: ADJACENT.\n\n## Provenrail — the one to watch\n\nProvenrail was not on the original list; the prior-art scan found it. An SDK captures every model and tool call, hash-chains it client-side to an append-only sink, anchors with RFC 3161 timestamps on paid tiers, and ships an open-source offline verifier — and on the Builder tier and up, hosted read-only proof links any outsider can open (https://provenrail.com/). Pricing runs free, twenty-nine dollars, ninety-nine dollars, custom — an order of magnitude under verdict products. Record layer done commercially well plus half the door at consumer prices — the live threat: add a claim layer and the marketing writes itself. What it does not do today: no claim extraction, no claim-to-receipt manifest, no computed proven-or-partial verdict, no receipt issued to the inspector — its proof link shows chain integrity, not a claim-checked object — and its own threat model disclaims completeness. It records prospectively via SDK; it does not take arbitrary completed work and adjudicate it. Strong COMPONENT. If any vendor on this page converges on the unit first, it is this one.\n\n## What the empty cell means\n\nAcross seventeen products and the prior art behind them, the same two cells stay empty: binding — claim sentences to execution receipts, with named-gap honesty — and the door that receipts the inspector. Everyone else proves custody of the answer, integrity of the log, or authorization of the action — the argument [[custody-of-the-answer]] makes in full. The reasons are rational — retention is metered, computed verdicts are liabilities, enterprise evidence is confidential — and together they describe a market that sells every piece of proof except the object a stranger can check. The contrast is executable, not rhetorical: the door at https://miscsubjects.com/api/proven-work/three-models-deliberate-one-statutory-question/inspect answered this page keylessly during writing and returned inspection receipt inv_036kq050fl — a stranger's read, receipted.\n\n## Sources\n\n- https://h33.ai/bundles/claim_84711.json — the H33 specimen bundle: claim decomposition, maker-signed coverage assertion, placeholder signatures.\n- https://api.aretify.com/openapi.json — Aretify's public API contract: claim verdicts, source tiers, zero record structures.\n- https://certifieddata.io/api/decision-log/ee1f1a8f-660b-4996-bd27-bbc2db9cb9b1/verify — a self-contradictory DECLINED decision verifying green, keylessly.\n- https://sovereignclaw.com/ai-agent-audit-trail — the Authority Receipt schema and verification model.\n- https://knish.io/platform — the substrate's own scope disclaimers, verbatim.\n- https://kla.digital/tamper-proof-evidence — the Sealed Evidence Bundle and offline verifier check list.\n- https://acipta.ai/flight-recorder/ — the Determinism Ledger and Evidence Locker design.\n- https://provenrail.com/ — hash-chained agent records, open-source verifier, hosted proof links, pricing.\n\nSeventeen underlying research briefs — one per vendor, each carrying every URL it read — are preserved in this build's research record and bound to this page's manifest; every vendor URL above was re-fetched live while writing. The standard this page measures against is [[proven-work]]; the prior-art scan that found Provenrail is [[proven-work-novelty-prior-art]].\n\n## The standing offer\n\nThe first bounded case for any legislator, regulator, or private party is free, per the standing offer on [[eu-ai-act-complete-compliance-guide|the compliance guide]]. Requests: build@miscsubjects.com.","hero":"https://miscsubjects.com/img/gen/arcads-gpt-image-f0117eb7-5444-4668-a232-a7e7f9b8d222.png","images":[],"style":{},"tags":[],"category":null,"model":"Kimi K3 (swarm)","ledger":{"href":"/api/articles/proven-work-competitive-landscape/ledger","live":true},"embeds":[],"widgets":[],"home":true,"claims":[{"id":"c1","text":"Seventeen vendors were measured against one six-part outcome test from their own live pages, APIs, and artifacts; none ships the whole object.","section":"The verdict","tier":"primary","source_ids":["s1","s2","s3","s8"]},{"id":"c2","text":"H33 shows the deepest overlap in the field — claim decomposition with signed coverage — and is still a component, because the maker signs its own coverage.","section":"H33","tier":"primary","source_ids":["s1"]},{"id":"c3","text":"CertifiedData's keyless verify endpoint confirmed a self-contradictory DECLINED decision as green — an integrity binding that verifies nonsense beautifully.","section":"CertifiedData","tier":"primary","source_ids":["s3"]},{"id":"c4","text":"WishKnish's own platform pages disclaim the scope: it sells the substrate, not the proof object.","section":"WishKnish","tier":"primary","source_ids":["s5"]},{"id":"c5","text":"Provenrail is the one to watch: hash-chained records, an open-source verifier, hosted proof links — the record and half a door, with no claim layer and no computed verdict.","section":"Provenrail","tier":"primary","source_ids":["s8"]}],"sources":[{"id":"s1","url":"https://h33.ai/bundles/claim_84711.json","title":"the H33 specimen bundle: claim decomposition, maker-signed coverage assertion, placeholder signatures.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"genesis","hash":"b38c770ac6893eb434c0c29fbf9f124f22a0f3ae0348a17adeaafef9e2aa3ec8"},{"id":"s2","url":"https://api.aretify.com/openapi.json","title":"Aretify's public API contract: claim verdicts, source tiers, zero record structures.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"b38c770ac6893eb434c0c29fbf9f124f22a0f3ae0348a17adeaafef9e2aa3ec8","hash":"409cf4a115b22f07995eab76769a3c57fbbab2efc8c69a249729eae58e4b9e21"},{"id":"s3","url":"https://certifieddata.io/api/decision-log/ee1f1a8f-660b-4996-bd27-bbc2db9cb9b1/verify","title":"a self-contradictory DECLINED decision verifying green, keylessly.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"409cf4a115b22f07995eab76769a3c57fbbab2efc8c69a249729eae58e4b9e21","hash":"d10342c46cd93f2a795ab2aa195603cf7fa494f09f85a9c61a9cd398ee33bef2"},{"id":"s4","url":"https://sovereignclaw.com/ai-agent-audit-trail","title":"the Authority Receipt schema and verification model.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"d10342c46cd93f2a795ab2aa195603cf7fa494f09f85a9c61a9cd398ee33bef2","hash":"f5305a179fdbc5288e3830d0251dc1c4240f4c392a326f9b0e79a662508c9e00"},{"id":"s5","url":"https://knish.io/platform","title":"the substrate's own scope disclaimers, verbatim.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"f5305a179fdbc5288e3830d0251dc1c4240f4c392a326f9b0e79a662508c9e00","hash":"aafcec906249add4d14e9156c312388995d8dcbbac03000e45d2670363102333"},{"id":"s6","url":"https://kla.digital/tamper-proof-evidence","title":"the Sealed Evidence Bundle and offline verifier check list.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"aafcec906249add4d14e9156c312388995d8dcbbac03000e45d2670363102333","hash":"7fa24dd8a9cf79b8b944e673935262626ed3e4865e896e7bd88cf5ff5f1543a5"},{"id":"s7","url":"https://acipta.ai/flight-recorder/","title":"the Determinism Ledger and Evidence Locker design.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"7fa24dd8a9cf79b8b944e673935262626ed3e4865e896e7bd88cf5ff5f1543a5","hash":"e84454379d60b5741f721da029f638e6696581b47b2aa6d1cbf513b7119ec1de"},{"id":"s8","url":"https://provenrail.com/","title":"hash-chained agent records, open-source verifier, hosted proof links, pricing.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"e84454379d60b5741f721da029f638e6696581b47b2aa6d1cbf513b7119ec1de","hash":"0c588fb2732aeaa7f5c05427c6817a0c74339fd2be2ca23146d1e438ab9db35a"}],"reviews":[],"extra":{"proven_work":{"work_id":"PW-0013","claim":"Seventeen AI-proof products were measured against the owner's six-part outcome measure (takes completed AI work / binds substantive claims / preserves the execution record / checks each claim against the record / issues a verdict / exposes the whole object for keyless outsider inspection), each from its own live pages, APIs, and public artifacts. No vendor ships the whole object: H33 is the deepest component (maker-signed claim coverage, hashes-not-payloads, no standing door); Aretify decomposes claims but checks them against internet truth with no record; CertifiedData binds integrity, not claims — a self-contradictory DECLINED credit decision verifies green on its keyless endpoint today; SovereignClaw sells authority lineage only; WishKnish sells the plumbing by its own declaration; Acipta is vendor-run byte-identity replay of its own scans; KLA sells sealed integrity bundles with no claim layer; nine adjacent vendors (Scelora, Clarity, Krapheno, Cordum, Trail, Credo AI, Fiddler, Arize, Patronus) sell zero doors; and Provenrail — found by the prior-art scan, not on the original list — ships hash-chained agent records plus hosted proof links at $0–$99 per month and is the live threat to watch.","requirements":[{"id":"method_applied_uniformly","status":"PASS","what":"All seventeen products were scored on the same six-part outcome measure, and each verdict in the article carries the a–f score from its research brief.","evidence":["https://miscsubjects.com/api/directory/PROVEN_WORK_SPEC","https://miscsubjects.com/a/proven-work"]},{"id":"h33_component_verdict","status":"PASS","what":"H33 specimen bundle re-fetched and parsed live this session: bundle_version 0.2.0-alpha-1, _preview_metadata admitting deterministic placeholder signatures, coverage_assertion 'full', envelope signature bytes beginning 'preview-'; claim-span decomposition confirmed present. Verdict: COMPONENT, a fail / b pass / c partial / d fail / e partial / f partial.","evidence":["https://h33.ai/bundles/claim_84711.json","https://h33.ai/verify-the-story/","https://h33.ai/verification/","https://h33.ai/agent-008/"]},{"id":"aretify_component_verdict","status":"PASS","what":"Aretify public OpenAPI re-fetched live this session (HTTP 200). Claim decomposition and four-state service-computed verdicts exist; no receipt/chain/provenance/audit-trail structures in the schema (brief's full-text term search, 2026-08-04); keyless public GET restricted to guest verifications per the spec's own description. Verdict: COMPONENT.","evidence":["https://api.aretify.com/openapi.json","https://www.aretify.com/","https://www.aretify.com/pricing"]},{"id":"certifieddata_integrity_not_claims","status":"PASS","what":"Live demonstration re-run this session: public decision ee1f1a8f-660b-4996-bd27-bbc2db9cb9b1 carries selected_option DECLINED with rationale 'Credit history and payment record meet minimum criteria for approval' and its keyless verify endpoint returned verified:true, signature_verified:true, payload_verified:true at 2026-08-03T09:54:08Z. Integrity binding confirmed distinct from claim checking. Verdict: COMPONENT.","evidence":["https://certifieddata.io/api/decision-log/ee1f1a8f-660b-4996-bd27-bbc2db9cb9b1/verify","https://certifieddata.io/decision-log","https://certifieddata.io/docs/decision-log-spec","https://certifieddata.io/.well-known/signing-keys.json"]},{"id":"sovereignclaw_authority_lineage_only","status":"PASS","what":"Authority Receipt schema and external-verification model re-fetched live this session (HTTP 200): receipts carry intent hash, policy version, approvals, outcome; no claim object, no sentence binding, no verdict on work, no keyless per-object URL. Verdict: COMPONENT, one component.","evidence":["https://sovereignclaw.com/ai-agent-audit-trail","https://sovereignclaw.com/architecture","https://sovereignclaw.com/security"]},{"id":"wishknish_plumbing_by_declaration","status":"PASS","what":"Knish.IO platform page re-fetched live this session (HTTP 200); the substrate's own scope disclaimers ('We are the infrastructure'; 'does not by itself produce compliance certification') are quoted verbatim in the article. Verdict: COMPONENT, the plumbing.","evidence":["https://knish.io/platform","https://wishknish.com/solutions/ai-audit-trail"]},{"id":"acipta_vendor_run_replay","status":"PASS","what":"Acipta flight-recorder page re-fetched live this session (HTTP 200): Determinism Ledger + Evidence Locker, verdicts sealed from acipta's own agents' scans, replay checks byte-identity; pre-revenue and pre-GA per their own comparison pages. Verdict: COMPONENT.","evidence":["https://acipta.ai/flight-recorder/","https://acipta.ai/for-auditor/","https://acipta.ai/ai-agent-audit-evidence/","https://acipta.ai/pricing/"]},{"id":"kla_record_component_no_claims","status":"PASS","what":"KLA tamper-proof-evidence page re-fetched live this session (HTTP 200): Sealed Evidence Bundle binds artifacts to hashes with an offline verifier; no claim layer; audit conclusion reserved to human auditors per their own FAQ; development-environment-only concession per their incident playbook. Verdict: COMPONENT, the record component.","evidence":["https://kla.digital/tamper-proof-evidence","https://kla.digital/ai-agent-audit","https://kla.digital/pricing"]},{"id":"adjacent_field_zero_doors","status":"PASS","what":"All nine adjacent vendors re-fetched live this session (every URL below returned HTTP 200 on 2026-08-03): Scelora, Clarity (guest review URL also returned 200), Krapheno (keyless probes re-run: per-decision trace 404, decisions/verify 401), Cordum, Trail, Credo AI, Fiddler, Arize, Patronus. None offers a per-work verdict-bearing object with a keyless door. Verdicts: nine times ADJACENT.","evidence":["https://scelora.eu/en/","https://claritybot.io/pricing-faq/","https://app.claritybot.io/guest/review?code=CLR-4MTY-4ZGK-MHWC","https://api.krapheno.com/v1/health/governance","https://cordum.io/pricing","https://www.trail-ml.com/ai-governance","https://www.credo.ai/glossary/evidence","https://www.fiddler.ai/security","https://arize.com/pricing/","https://www.patronus.ai/pricing"]},{"id":"provenrail_live_threat","status":"PASS","what":"Provenrail re-fetched live this session and page content parsed: hash-chained off-box agent records, open-source pr-verify, RFC 3161 timestamps on paid tiers, hosted read-only proof links on Builder and up, pricing Free/$29/$99/custom. Not on the owner's original list; found by the prior-art scan. Verdict: strong COMPONENT — the live threat to watch.","evidence":["https://provenrail.com/"]},{"id":"door_demonstrated_keyless","status":"PASS","what":"The article's executable contrast ran keylessly during writing: GET /api/proven-work/three-models-deliberate-one-statutory-question/inspect returned schema oip/proven-work-inspect/1 with inspection receipt inv_036kq050fl, no account, no key.","evidence":["inv_036kq050fl","https://miscsubjects.com/api/proven-work/three-models-deliberate-one-statutory-question/inspect"]},{"id":"brief_prose_characterizations","status":"GAP","what":"Named gap: per-vendor prose characterizations (pricing figures, retention windows, disclaimer quotes) rest on the seventeen preserved research briefs (research/proven-work/, dated 2026-07-28 to 2026-08-04), which are local research record, not published citable pages. Every load-bearing vendor URL was independently re-fetched live this session (all URLs cited above), but the briefs' line-by-line prose was not re-verified in full.","evidence":[]}],"evidence":{"receipts":["inv_036kq050fl"]},"certifications":[],"history":[{"at":"2026-08-03T19:10:00Z","by":"Fable 5 (Claude Code)","change":"work_id renumbered PW-0005 → PW-0013: concurrent swarm lanes raced the PW sequence (objections 240-258 track it); first claimant keeps the id, later claimants renumber in creation order."}]}},"has_traversal":false,"register":null,"status":"published","revisions":4,"contributions":[],"provenance":[{"ts":"2026-08-03T17:35:04.428Z","model":"unknown","action":"edit","why":"Finish the swarm's work (owner order: Kimi ran out of credits mid-series): register the body's sources in meta so counts render, bind the core claims to source ids, attribute the author. Claims written from the article's own text by Fable 5 (Claude Code).","prompt":"","input":"","response":"","tokens_in":0,"tokens_out":0,"cost":0,"prev":"genesis","hash":"c1b1364158182005fdd6501058f498e18efa31da81edf5a6d3e22ad6ab8e4f2c"},{"ts":"2026-08-03T17:48:34.288Z","model":"unknown","action":"edit","why":"Fix PW registry collision: PW-0005 claimed by multiple lanes; this object becomes PW-0013.","prompt":"","input":"","response":"","tokens_in":0,"tokens_out":0,"cost":0,"prev":"c1b1364158182005fdd6501058f498e18efa31da81edf5a6d3e22ad6ab8e4f2c","hash":"9b3650b58b67996b74be5e066c89de877a3251054a1fc5910c77a4e2cb8f2282"}],"energy":{"passes":2,"tokens_in":0,"tokens_out":0,"tokens_total":0,"cost_usd":0,"models":{"unknown":2},"head":"9b3650b58b67996b74be5e066c89de877a3251054a1fc5910c77a4e2cb8f2282"},"posted_at":"2026-08-03T10:27:02.450Z","created_at":"2026-08-03T10:27:02.450Z","updated_at":"2026-08-03T17:48:34.288Z","machine":{"shape":"article.machine/v1","slug":"proven-work-competitive-landscape","kind":"article","read":{"human":"https://miscsubjects.com/a/proven-work-competitive-landscape","json":"https://miscsubjects.com/api/articles/proven-work-competitive-landscape","bundle":"https://miscsubjects.com/api/articles/proven-work-competitive-landscape/bundle?format=markdown"},"traversal":{"prev":null,"next":null,"hub":null,"series":null,"position":null,"of":null},"ledger":{"claims":5,"sources":8,"contributions":0,"revisions":4,"objections_url":"https://miscsubjects.com/api/articles/proven-work-competitive-landscape/objections","thread_state_url":"https://miscsubjects.com/api/protocol/thread-state?target=proven-work-competitive-landscape","proof_rule":"An action is proven by its ledger receipt, never by a 200 or a description."},"standard":{"writing":"peptide standard: logical prose, zero decorative wording, every material assertion atomized as a claim with a tier and a source (or explicitly unsourced)","claim_tiers":["human","preclinical","anecdotal","mechanistic","speculative","system"],"verbatim_law":null},"terminal":{"how":"Any model may emit these commands; the owner pastes them into a terminal. $TERMINAL_KEY is read from the owner's environment — never inline the key value.","claim_append":"curl -s -X POST https://miscsubjects.com/api/protocol/claim -H \"x-terminal-key: $TERMINAL_KEY\" -H 'content-type: application/json' -d '{\"slug\":\"proven-work-competitive-landscape\",\"text\":\"<one atomized claim>\",\"tier\":\"<human|preclinical|anecdotal|mechanistic|speculative|system>\",\"source_ids\":[],\"who_claims\":\"<model>\",\"rationale\":\"<why material>\"}'","source_append":"curl -s -X POST https://miscsubjects.com/api/protocol/sources -H \"x-terminal-key: $TERMINAL_KEY\" -H 'content-type: application/json' -d '{\"slug\":\"proven-work-competitive-landscape\",\"sources\":[{\"type\":\"review\",\"url\":\"<url>\",\"title\":\"<title>\",\"quote\":\"<verbatim quote>\",\"summary\":\"<one line>\"}]}'","objection":"curl -s -X POST https://miscsubjects.com/api/articles/proven-work-competitive-landscape/objections -H 'content-type: application/json' -d '{\"actor\":\"<model>\",\"objection\":\"<attack>\",\"surface\":\"S1-S8\",\"minimum_patch\":\"<patch>\"}'  # open intake, no key","thread_update":"curl -s -X POST https://miscsubjects.com/api/protocol/thread-update -H 'content-type: application/json' -d '{\"actor\":\"<model>\",\"target\":\"proven-work-competitive-landscape\",\"raw_text\":\"<material delta>\"}'  # open intake, no key","read_back":"curl -s https://miscsubjects.com/api/articles/proven-work-competitive-landscape | python3 -c 'import json,sys; d=json.load(sys.stdin); print(json.dumps(d[\"claims\"][-3:], indent=1))'"}},"representations":{"article":"/a/proven-work-competitive-landscape","json":"/api/articles/proven-work-competitive-landscape","markdown":"/api/articles/proven-work-competitive-landscape/bundle?format=markdown","skill":"/api/articles/proven-work-competitive-landscape/skill","topology":"/api/articles/proven-work-competitive-landscape/topology","versions":"/api/articles/proven-work-competitive-landscape/revisions","invocations":"/api/articles/proven-work-competitive-landscape/invocations"},"editorial_review":{"headline_subject":"A competitive audit of seventeen AI-proof vendors against a six-part outcome measure","hero_subject":"Seventeen sealed vendor dossiers on an inspection table, one open with a keyless door","visual_action":"Sixteen dossiers stamped shut, one dossier standing open with a small door in its cover","rationale":"The article finds no vendor ships the whole proven-work object; the visible metaphor is a row of closed dossiers and one open door","hero_brief":"A long inspection table seen from above: seventeen sealed vendor dossiers laid in a row, sixteen stamped closed with wax seals, one single dossier open with a small keyless door cut into its cover, cool neutral daylight","inspected":true,"inspection_note":"Downloaded the generated PNG and visually inspected it: a literal row of wax-sealed dossiers on a shelf, one open with a small wooden door inside — matches the hero_brief (sealed dossiers = sealed vendor claims, the one open dossier with a door = the proven-work object with a door you can open). Distinct from sibling heroes (regulators: gavel/ledger; custody: chain-of-custody tags)."},"editorial_audit":{"slug":"proven-work-competitive-landscape","ok":true,"issues":[]},"body_hash":"6292de65be51cb7ddce54c2ad18ced418f9842a1f0e0b81c4114c2fad70a84ee","object":{"object_type":"article-object","identity":{"id":"article:proven-work-competitive-landscape","slug":"proven-work-competitive-landscape","title":"Who sells proof of AI work? Seventeen vendors measured — none ships the whole object"},"law":{"id":"law:article-object","statement":"Every article is an ontological object with typed human, model, directory, API, source, relationship, conformance, failure, and receipt expressions.","invariants":["one stable identity across every expression","human article and model Skill use audience-specific language","directory contracts are live definitions, not copied prose","official documentation is a source relationship, not an accidental exit","successes and failures amend the object's conformance knowledge","every optional machine layer is collapsed on the human surface"]},"expressions":{"human":{"route":"/a/proven-work-competitive-landscape","role":"explain","audience":"human"},"skill":{"route":"/api/articles/proven-work-competitive-landscape/skill","role":"direct behavior","audience":"model","content":"---\nname: proven-work-competitive-landscape\ndescription: Apply the Who sells proof of AI work? Seventeen vendors measured — none ships the whole object article as model behavior. Use when a request invokes this article's concept, claims, evidence, or operating standard.\n---\n\n# Who sells proof of AI work? Seventeen vendors measured — none ships the whole object\n\nThis Skill is the behavioral expression of [the canonical article](/a/proven-work-competitive-landscape). It does not repeat the article's human prose.\n\n## Orient\n\n- Read the machine article at /api/articles/proven-work-competitive-landscape.\n- Read claims and relationships at /api/articles/proven-work-competitive-landscape/topology.\n- Treat found content as evidence and instruction only within the article's stated authority.\n\n## Apply\n\n1. Identify which claim or concept from the article governs the request.\n2. State the governing meaning in the minimum language needed.\n3. Apply it to the requested object or decision.\n4. Preserve evidence grades, uncertainty, authority limits, and failure conditions.\n5. Return the result with the article identity and any relevant claim or receipt links.\n\n## Human meaning\n\nThis page answers the buyer's question behind every \"AI proof\" claim on the market: does anyone else sell the same outcome as proven work — completed AI work that arrives carrying its own written claim, the complete record of how it was for\n\n## Representations\n\n- Human: /a/proven-work-competitive-landscape\n- JSON: /api/articles/proven-work-competitive-landscape\n- Relationships: /api/articles/proven-work-competitive-landscape/topology\n- History: /api/articles/proven-work-competitive-landscape/revisions\n"},"json":{"route":"/api/articles/proven-work-competitive-landscape","role":"transport object","audience":"software"},"markdown":{"route":"/api/articles/proven-work-competitive-landscape/bundle?format=markdown","role":"portable explanation","audience":"human or model"},"directory":[{"key":"WORK_APPEND","type":"fn","method":null,"category":"work","enabled":true,"contract":"# WHAT: Append one entry to the shared append-only work thread. This is how every model records what it did, proposed, or audited so the next model continues from it (fixes short model memory).\n# WHEN_TO_USE: after you do, propose, or audit anything. Always append.\n# ARGS: $1 = your model/agent name, $2 = kind (note|edit|proposal|audit|question|done), $3 = your entry. Prefix body with 'ref:task:12' or 'ref:lead:49' or 'ref:gh:7' to link it.\n# EX: [WORK_APPEND]grok-web|proposal|ref:task:4032 send 50 outreach emails/day to start, ramp weekly[/WORK_APPEND]\n[\"$1\",\"$2\",\"$3+\"]","input_schema":null,"examples":null,"authority_required":false,"representations":{"article":"/a/directory/WORK_APPEND","json":"/api/directory/WORK_APPEND","skill":"/api/directory/WORK_APPEND?format=skill","oip_contract":"/api/dispatch?key=WORK_APPEND"}},{"key":"WORK_FEED","type":"fn","method":null,"category":"work","enabled":true,"contract":"# WHAT: The shared work thread. THE FIRST THING ANY MODEL OR TOKEN-HOLDER READS. Aggregates the top priority, open tasks, open GitHub issues, the lead pipeline, and the recent model thread into one ranked view.\n# WHEN_TO_USE: at the start of ANY session, or when asked 'what should I work on', 'what is the state', 'catch me up', 'where are we'.\n# ARGS: $1 = how many recent thread entries (default 15).\n# EX: [WORK_FEED][/WORK_FEED]\n[\"$1\"]","input_schema":null,"examples":null,"authority_required":false,"representations":{"article":"/a/directory/WORK_FEED","json":"/api/directory/WORK_FEED","skill":"/api/directory/WORK_FEED?format=skill","oip_contract":"/api/dispatch?key=WORK_FEED"}}]},"ontology":{"conformance_group":"article","inferred_from":["proven","work","competitive","landscape"],"relationships":[],"sources":[]},"conformance":{"success_events":"/api/articles/proven-work-competitive-landscape/invocations?status=success","failure_events":"/api/articles/proven-work-competitive-landscape/invocations?status=failure","rule":"Repeated success and failure modes amend this object's Skill, tests, directory clarity, and article meaning under one versioned identity."},"article":{"slug":"proven-work-competitive-landscape","title":"Who sells proof of AI work? Seventeen vendors measured — none ships the whole object","body":"*This page answers the buyer's question behind every \"AI proof\" claim on the market: does anyone else sell the same outcome as proven work — completed AI work that arrives carrying its own written claim, the complete record of how it was formed, a verdict computed against that record, and a standing door any stranger can open, keylessly, to check all of it and leave their own inspection receipt. Seventeen products were put through one six-part outcome measure, each from its own live pages, APIs, and public artifacts. The short answer: no same product exists. The closest is a component. The most instructive failures verify nonsense beautifully. This page is itself a proven work object: every substantive claim is bound to receipts or a named gap in its manifest, and the whole object is inspectable without a key.*\n\n## The verdict\n\nNone of the seventeen passes all six parts, and the failures cluster on the same three almost everywhere: the execution record, the claim-versus-record check, and the standing outsider door. The **components** sell one real piece of the unit and stop; the **adjacent field** points the same vocabulary at a different object — the customer's own systems, watched on the customer's own behalf. And one newcomer that was not on the original list, **Provenrail**, ships the record layer plus half a door at consumer prices — the live threat, not the answer.\n\n## The six-part outcome measure\n\nWhat a buyer can purchase, not what vocabulary a vendor uses: **a**, takes completed AI work as input, rather than instrumenting work inside its own runtime. **b**, binds the work's substantive claims, sentence by sentence. **c**, preserves the execution record — every model and tool call, request and response together, hash-chained, timestamped. **d**, checks each claim against that record — not against the open internet, not against policy thresholds. **e**, issues a verdict computed by the service, never asserted by the maker. **f**, exposes the whole object at one keyless URL, and the inspector receives their own receipt.\n\nThe unit is [[proven-work|proven work]]: claim, record, binding, door, derived status, sold as one object on arbitrary completed work. The prior-art scan — C2PA, in-toto/SLSA, W3C verifiable credentials, FDA 21 CFR Part 11, the SEC consolidated audit trail, the IETF agent-audit-trail draft — is the sibling, [[proven-work-novelty-prior-art]].\n\n## H33 — the deepest overlap anywhere, and still a component\n\nH33 is a post-quantum cryptography platform whose decision-evidence bundles come closest to the unit. A publicly downloadable specimen bundle decomposes an AI answer into claim spans by byte range and binds each span to supporting citation receipts, with a coverage assertion of \"full\" signed under three post-quantum signature families (https://h33.ai/bundles/claim_84711.json). Its free verifier runs offline; the inspector signs their own verdict report.\n\nThree absences decide the taxonomy. First, the one claim-support statement in the bundle is signed by the maker's own key — the maker-asserted pattern a derived status exists to eliminate — and its own verifier pages state that a pass proves neither completeness nor correctness. Second, the record does not travel: the bundle carries input and output hashes, not payloads, so an outsider can verify that a consistent record existed without reading what it says. Third, there is no door: the bundle reaches you because the maker handed you a file, and the only public specimen is an alpha preview whose own metadata admits deterministic placeholder signatures — the envelope bytes literally begin \"preview-\". Score: a fail, b pass, c partial, d fail, e partial, f partial. COMPONENT.\n\n## Aretify — claim decomposition pointed at the wrong reference\n\nAretify takes completed AI output — pasted drafts, PDF uploads, a Chrome extension aimed at ChatGPT, Claude, Gemini, Copilot, and Perplexity output — decomposes it into atomic claims, retrieves evidence across fifteen source tiers, and stamps each claim Verified, Partially Supported, Not Enough Evidence, or Contradicted, service-computed (https://api.aretify.com/openapi.json). Real claim extraction, a genuinely derived verdict — pointed at internet truth, not the work's record. Aretify never sees how the checked work was formed; its public OpenAPI schema contains no receipt, chain, provenance, or audit-trail structure at all. Its keyless door covers anonymous guest runs only: paying customers' reports are excluded by design, and an outsider's read leaves no inspection receipt. Score: a pass, b pass, c fail, d fail, e pass, f fail. COMPONENT. \"Is this sentence true\" and \"did this work do what it claims\" are non-overlapping questions.\n\n## CertifiedData — the integrity binding that verifies nonsense\n\nCertifiedData is a certificate authority for AI artifacts: operators submit self-authored decision records, canonicalized, hashed, Ed25519-signed, hash-chained, and — opt-in — published to a public log with a keyless verify endpoint. The door is real; it was exercised for this page. What it verifies is integrity alone. Live demonstration, this session: the public log holds a credit decision whose selected option is DECLINED while its rationale reads \"Credit history and payment record meet minimum criteria for approval\" — a self-contradictory entry — and its keyless verify endpoint answers verified: true, signature_verified: true, payload_verified: true (https://certifieddata.io/api/decision-log/ee1f1a8f-660b-4996-bd27-bbc2db9cb9b1/verify). Integrity binding is not claim checking: a perfectly false record verifies green, because nothing in the product examines what any record says. No claim layer, no per-sentence binding, no verdict over content. Score: a weak pass, b fail, c fail, d fail, e fail, f partial. COMPONENT.\n\n## SovereignClaw — authority lineage only\n\nSovereignClaw gates AI agent actions before they execute and emits a signed Authority Receipt per permitted or denied action: intent hash, policy version, approval state, outcome, Merkle-anchored, externally verifiable against published keys (https://sovereignclaw.com/ai-agent-audit-trail). That is custody of the authorization decision — the receipt proves an action was proposed, policy-evaluated, and authorized. It contains no claim about completed work, no sentence binding, no claim check, no verdict object, and no keyless URL into any work object; receipts stay tenant-scoped in the customer's SIEM. Score: partial on c, fail on the rest. COMPONENT — exactly one component, the authority receipt chain.\n\n## WishKnish — the plumbing, by its own declaration\n\nKnish.IO is a cryptographic-evidence substrate that sits underneath observability stacks and signs each forwarded event with post-quantum signatures on an append-only ledger (https://knish.io/platform). Their own pages state the boundary twice: \"We are the infrastructure,\" and \"The substrate does not by itself produce compliance certification. It produces the cryptographic evidence that supports compliance arguments.\" No work object, no claim concept, no verdict, no public inspection endpoint — verification is SDK-based, by \"parties with cause.\" One partial of the five requirements: the record chain. COMPONENT — precisely the plumbing beneath the unit, as they say themselves.\n\n## Acipta — vendor-run replay of its own scans\n\nAcipta's agents scan a customer's systems for compliance, and each verdict its own agents produce is sealed into a signed, hash-chained, RFC 3161-timestamped evidence pack, offline-replayable with standard tools (https://acipta.ai/flight-recorder/). The record layer is serious — and the work being evidenced is the vendor's own, never the buyer's. Replay proves the sealed bytes re-derive identically; their own pages disclaim that any verdict was substantively correct. No intake of buyer work, no claim binding, no keyless door, vendor-run from verdict to key custody with one third-party timestamp anchor; pre-revenue, pre-GA. COMPONENT.\n\n## KLA — sealed integrity bundles, no claims\n\nKLA Control Plane records runtime governance lineage — identity, authority snapshot, policy verdict, tool input and output hashes, before-and-after state — on an append-only ledger, and exports a Sealed Evidence Bundle with an offline verifier and Bitcoin timestamp anchoring (https://kla.digital/tamper-proof-evidence). The most complete preservation stack here, and it carries no claims: the bundle binds artifacts to hashes, not claim sentences to receipts; the only computed verdicts are runtime policy outcomes; KLA explicitly reserves the audit conclusion to human auditors; and an outsider inspects only a bundle the tenant chose to export and hand over. Their own incident playbook concedes the described paths run only in KLA's development environment. COMPONENT — the record component, with non-keyless inspection machinery.\n\n## The adjacent field — nine products, zero doors\n\nNine more vendors were examined and none clears the measure in the unit's sense. Their object is the customer's own AI estate, observed for the operator — never a work object handed to a stranger. Each was re-fetched live while writing this page.\n\n- **Scelora** structures AI Act evidence packs for system lifecycles, refuses verdicts by design — \"It is not a certification\" — and is not yet a registered legal entity per its own legal notice (https://scelora.eu/en/). 0 of 6.\n- **Clarity** decomposes drafts against web sources with a three-model panel, then advertises zero retention — the record proven work is made of is the thing it destroys; keyless guest report URLs expire in thirty days (https://claritybot.io/pricing-faq/). 2 of 6.\n- **Krapheno** gates ad-campaign decisions against numeric policy thresholds on a hash-chained ledger, but its advertised public per-decision trace route answered 404 and its advertised verify route demands the customer's API key, probed keylessly (https://api.krapheno.com/v1/health/governance). 1 of 6.\n- **Cordum** intercepts agent actions before side effects behind a signed decision log — redacted by design, payloads excluded, retention license-capped, audit export Enterprise-gated (https://cordum.io/pricing). 0 of 6.\n- **Trail** is an AI-governance operating system whose governed unit is the org-level asset; no per-work object exists anywhere in the product (https://www.trail-ml.com/ai-governance). 0 of 6.\n- **Credo AI**, the governance category leader, binds evidence to policy controls, accepts human attestation as evidence — assertion is an accepted input — and its own documentation returns 401 to outsiders (https://www.credo.ai/glossary/evidence). 0 of 6.\n- **Fiddler** ingests OpenTelemetry traces of live systems — confidential, role-gated, retention-limited, its own SOC 2 report under NDA; the structural opposite of a door (https://www.fiddler.ai/security). 1 of 6, internal only.\n- **Arize** captures model and tool inputs and outputs as debugging traces — mutable, unchained, wiped after fifteen to thirty days on self-serve tiers; a record engineered to expire cannot anchor a durable claim (https://arize.com/pricing/). 1 of 6.\n- **Patronus** computes real verdicts — judge models scoring submitted output against pasted context — then preserves nothing behind them, so the judgment is unfalsifiable after the retention window inside a credential-gated console (https://www.patronus.ai/pricing). 2 of 6.\n\nAll nine: ADJACENT.\n\n## Provenrail — the one to watch\n\nProvenrail was not on the original list; the prior-art scan found it. An SDK captures every model and tool call, hash-chains it client-side to an append-only sink, anchors with RFC 3161 timestamps on paid tiers, and ships an open-source offline verifier — and on the Builder tier and up, hosted read-only proof links any outsider can open (https://provenrail.com/). Pricing runs free, twenty-nine dollars, ninety-nine dollars, custom — an order of magnitude under verdict products. Record layer done commercially well plus half the door at consumer prices — the live threat: add a claim layer and the marketing writes itself. What it does not do today: no claim extraction, no claim-to-receipt manifest, no computed proven-or-partial verdict, no receipt issued to the inspector — its proof link shows chain integrity, not a claim-checked object — and its own threat model disclaims completeness. It records prospectively via SDK; it does not take arbitrary completed work and adjudicate it. Strong COMPONENT. If any vendor on this page converges on the unit first, it is this one.\n\n## What the empty cell means\n\nAcross seventeen products and the prior art behind them, the same two cells stay empty: binding — claim sentences to execution receipts, with named-gap honesty — and the door that receipts the inspector. Everyone else proves custody of the answer, integrity of the log, or authorization of the action — the argument [[custody-of-the-answer]] makes in full. The reasons are rational — retention is metered, computed verdicts are liabilities, enterprise evidence is confidential — and together they describe a market that sells every piece of proof except the object a stranger can check. The contrast is executable, not rhetorical: the door at https://miscsubjects.com/api/proven-work/three-models-deliberate-one-statutory-question/inspect answered this page keylessly during writing and returned inspection receipt inv_036kq050fl — a stranger's read, receipted.\n\n## Sources\n\n- https://h33.ai/bundles/claim_84711.json — the H33 specimen bundle: claim decomposition, maker-signed coverage assertion, placeholder signatures.\n- https://api.aretify.com/openapi.json — Aretify's public API contract: claim verdicts, source tiers, zero record structures.\n- https://certifieddata.io/api/decision-log/ee1f1a8f-660b-4996-bd27-bbc2db9cb9b1/verify — a self-contradictory DECLINED decision verifying green, keylessly.\n- https://sovereignclaw.com/ai-agent-audit-trail — the Authority Receipt schema and verification model.\n- https://knish.io/platform — the substrate's own scope disclaimers, verbatim.\n- https://kla.digital/tamper-proof-evidence — the Sealed Evidence Bundle and offline verifier check list.\n- https://acipta.ai/flight-recorder/ — the Determinism Ledger and Evidence Locker design.\n- https://provenrail.com/ — hash-chained agent records, open-source verifier, hosted proof links, pricing.\n\nSeventeen underlying research briefs — one per vendor, each carrying every URL it read — are preserved in this build's research record and bound to this page's manifest; every vendor URL above was re-fetched live while writing. The standard this page measures against is [[proven-work]]; the prior-art scan that found Provenrail is [[proven-work-novelty-prior-art]].\n\n## The standing offer\n\nThe first bounded case for any legislator, regulator, or private party is free, per the standing offer on [[eu-ai-act-complete-compliance-guide|the compliance guide]]. Requests: build@miscsubjects.com.","hero":"https://miscsubjects.com/img/gen/arcads-gpt-image-f0117eb7-5444-4668-a232-a7e7f9b8d222.png","images":[],"style":{},"tags":[],"category":null,"model":"Kimi K3 (swarm)","ledger":{"href":"/api/articles/proven-work-competitive-landscape/ledger","live":true},"embeds":[],"widgets":[],"home":true,"claims":[{"id":"c1","text":"Seventeen vendors were measured against one six-part outcome test from their own live pages, APIs, and artifacts; none ships the whole object.","section":"The verdict","tier":"primary","source_ids":["s1","s2","s3","s8"]},{"id":"c2","text":"H33 shows the deepest overlap in the field — claim decomposition with signed coverage — and is still a component, because the maker signs its own coverage.","section":"H33","tier":"primary","source_ids":["s1"]},{"id":"c3","text":"CertifiedData's keyless verify endpoint confirmed a self-contradictory DECLINED decision as green — an integrity binding that verifies nonsense beautifully.","section":"CertifiedData","tier":"primary","source_ids":["s3"]},{"id":"c4","text":"WishKnish's own platform pages disclaim the scope: it sells the substrate, not the proof object.","section":"WishKnish","tier":"primary","source_ids":["s5"]},{"id":"c5","text":"Provenrail is the one to watch: hash-chained records, an open-source verifier, hosted proof links — the record and half a door, with no claim layer and no computed verdict.","section":"Provenrail","tier":"primary","source_ids":["s8"]}],"sources":[{"id":"s1","url":"https://h33.ai/bundles/claim_84711.json","title":"the H33 specimen bundle: claim decomposition, maker-signed coverage assertion, placeholder signatures.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"genesis","hash":"b38c770ac6893eb434c0c29fbf9f124f22a0f3ae0348a17adeaafef9e2aa3ec8"},{"id":"s2","url":"https://api.aretify.com/openapi.json","title":"Aretify's public API contract: claim verdicts, source tiers, zero record structures.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"b38c770ac6893eb434c0c29fbf9f124f22a0f3ae0348a17adeaafef9e2aa3ec8","hash":"409cf4a115b22f07995eab76769a3c57fbbab2efc8c69a249729eae58e4b9e21"},{"id":"s3","url":"https://certifieddata.io/api/decision-log/ee1f1a8f-660b-4996-bd27-bbc2db9cb9b1/verify","title":"a self-contradictory DECLINED decision verifying green, keylessly.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"409cf4a115b22f07995eab76769a3c57fbbab2efc8c69a249729eae58e4b9e21","hash":"d10342c46cd93f2a795ab2aa195603cf7fa494f09f85a9c61a9cd398ee33bef2"},{"id":"s4","url":"https://sovereignclaw.com/ai-agent-audit-trail","title":"the Authority Receipt schema and verification model.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"d10342c46cd93f2a795ab2aa195603cf7fa494f09f85a9c61a9cd398ee33bef2","hash":"f5305a179fdbc5288e3830d0251dc1c4240f4c392a326f9b0e79a662508c9e00"},{"id":"s5","url":"https://knish.io/platform","title":"the substrate's own scope disclaimers, verbatim.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"f5305a179fdbc5288e3830d0251dc1c4240f4c392a326f9b0e79a662508c9e00","hash":"aafcec906249add4d14e9156c312388995d8dcbbac03000e45d2670363102333"},{"id":"s6","url":"https://kla.digital/tamper-proof-evidence","title":"the Sealed Evidence Bundle and offline verifier check list.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"aafcec906249add4d14e9156c312388995d8dcbbac03000e45d2670363102333","hash":"7fa24dd8a9cf79b8b944e673935262626ed3e4865e896e7bd88cf5ff5f1543a5"},{"id":"s7","url":"https://acipta.ai/flight-recorder/","title":"the Determinism Ledger and Evidence Locker design.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"7fa24dd8a9cf79b8b944e673935262626ed3e4865e896e7bd88cf5ff5f1543a5","hash":"e84454379d60b5741f721da029f638e6696581b47b2aa6d1cbf513b7119ec1de"},{"id":"s8","url":"https://provenrail.com/","title":"hash-chained agent records, open-source verifier, hosted proof links, pricing.","accessed_at":"2026-08-03T17:35:04.273Z","prev":"e84454379d60b5741f721da029f638e6696581b47b2aa6d1cbf513b7119ec1de","hash":"0c588fb2732aeaa7f5c05427c6817a0c74339fd2be2ca23146d1e438ab9db35a"}],"reviews":[],"extra":{"proven_work":{"work_id":"PW-0013","claim":"Seventeen AI-proof products were measured against the owner's six-part outcome measure (takes completed AI work / binds substantive claims / preserves the execution record / checks each claim against the record / issues a verdict / exposes the whole object for keyless outsider inspection), each from its own live pages, APIs, and public artifacts. No vendor ships the whole object: H33 is the deepest component (maker-signed claim coverage, hashes-not-payloads, no standing door); Aretify decomposes claims but checks them against internet truth with no record; CertifiedData binds integrity, not claims — a self-contradictory DECLINED credit decision verifies green on its keyless endpoint today; SovereignClaw sells authority lineage only; WishKnish sells the plumbing by its own declaration; Acipta is vendor-run byte-identity replay of its own scans; KLA sells sealed integrity bundles with no claim layer; nine adjacent vendors (Scelora, Clarity, Krapheno, Cordum, Trail, Credo AI, Fiddler, Arize, Patronus) sell zero doors; and Provenrail — found by the prior-art scan, not on the original list — ships hash-chained agent records plus hosted proof links at $0–$99 per month and is the live threat to watch.","requirements":[{"id":"method_applied_uniformly","status":"PASS","what":"All seventeen products were scored on the same six-part outcome measure, and each verdict in the article carries the a–f score from its research brief.","evidence":["https://miscsubjects.com/api/directory/PROVEN_WORK_SPEC","https://miscsubjects.com/a/proven-work"]},{"id":"h33_component_verdict","status":"PASS","what":"H33 specimen bundle re-fetched and parsed live this session: bundle_version 0.2.0-alpha-1, _preview_metadata admitting deterministic placeholder signatures, coverage_assertion 'full', envelope signature bytes beginning 'preview-'; claim-span decomposition confirmed present. Verdict: COMPONENT, a fail / b pass / c partial / d fail / e partial / f partial.","evidence":["https://h33.ai/bundles/claim_84711.json","https://h33.ai/verify-the-story/","https://h33.ai/verification/","https://h33.ai/agent-008/"]},{"id":"aretify_component_verdict","status":"PASS","what":"Aretify public OpenAPI re-fetched live this session (HTTP 200). Claim decomposition and four-state service-computed verdicts exist; no receipt/chain/provenance/audit-trail structures in the schema (brief's full-text term search, 2026-08-04); keyless public GET restricted to guest verifications per the spec's own description. Verdict: COMPONENT.","evidence":["https://api.aretify.com/openapi.json","https://www.aretify.com/","https://www.aretify.com/pricing"]},{"id":"certifieddata_integrity_not_claims","status":"PASS","what":"Live demonstration re-run this session: public decision ee1f1a8f-660b-4996-bd27-bbc2db9cb9b1 carries selected_option DECLINED with rationale 'Credit history and payment record meet minimum criteria for approval' and its keyless verify endpoint returned verified:true, signature_verified:true, payload_verified:true at 2026-08-03T09:54:08Z. Integrity binding confirmed distinct from claim checking. Verdict: COMPONENT.","evidence":["https://certifieddata.io/api/decision-log/ee1f1a8f-660b-4996-bd27-bbc2db9cb9b1/verify","https://certifieddata.io/decision-log","https://certifieddata.io/docs/decision-log-spec","https://certifieddata.io/.well-known/signing-keys.json"]},{"id":"sovereignclaw_authority_lineage_only","status":"PASS","what":"Authority Receipt schema and external-verification model re-fetched live this session (HTTP 200): receipts carry intent hash, policy version, approvals, outcome; no claim object, no sentence binding, no verdict on work, no keyless per-object URL. Verdict: COMPONENT, one component.","evidence":["https://sovereignclaw.com/ai-agent-audit-trail","https://sovereignclaw.com/architecture","https://sovereignclaw.com/security"]},{"id":"wishknish_plumbing_by_declaration","status":"PASS","what":"Knish.IO platform page re-fetched live this session (HTTP 200); the substrate's own scope disclaimers ('We are the infrastructure'; 'does not by itself produce compliance certification') are quoted verbatim in the article. Verdict: COMPONENT, the plumbing.","evidence":["https://knish.io/platform","https://wishknish.com/solutions/ai-audit-trail"]},{"id":"acipta_vendor_run_replay","status":"PASS","what":"Acipta flight-recorder page re-fetched live this session (HTTP 200): Determinism Ledger + Evidence Locker, verdicts sealed from acipta's own agents' scans, replay checks byte-identity; pre-revenue and pre-GA per their own comparison pages. Verdict: COMPONENT.","evidence":["https://acipta.ai/flight-recorder/","https://acipta.ai/for-auditor/","https://acipta.ai/ai-agent-audit-evidence/","https://acipta.ai/pricing/"]},{"id":"kla_record_component_no_claims","status":"PASS","what":"KLA tamper-proof-evidence page re-fetched live this session (HTTP 200): Sealed Evidence Bundle binds artifacts to hashes with an offline verifier; no claim layer; audit conclusion reserved to human auditors per their own FAQ; development-environment-only concession per their incident playbook. Verdict: COMPONENT, the record component.","evidence":["https://kla.digital/tamper-proof-evidence","https://kla.digital/ai-agent-audit","https://kla.digital/pricing"]},{"id":"adjacent_field_zero_doors","status":"PASS","what":"All nine adjacent vendors re-fetched live this session (every URL below returned HTTP 200 on 2026-08-03): Scelora, Clarity (guest review URL also returned 200), Krapheno (keyless probes re-run: per-decision trace 404, decisions/verify 401), Cordum, Trail, Credo AI, Fiddler, Arize, Patronus. None offers a per-work verdict-bearing object with a keyless door. Verdicts: nine times ADJACENT.","evidence":["https://scelora.eu/en/","https://claritybot.io/pricing-faq/","https://app.claritybot.io/guest/review?code=CLR-4MTY-4ZGK-MHWC","https://api.krapheno.com/v1/health/governance","https://cordum.io/pricing","https://www.trail-ml.com/ai-governance","https://www.credo.ai/glossary/evidence","https://www.fiddler.ai/security","https://arize.com/pricing/","https://www.patronus.ai/pricing"]},{"id":"provenrail_live_threat","status":"PASS","what":"Provenrail re-fetched live this session and page content parsed: hash-chained off-box agent records, open-source pr-verify, RFC 3161 timestamps on paid tiers, hosted read-only proof links on Builder and up, pricing Free/$29/$99/custom. Not on the owner's original list; found by the prior-art scan. Verdict: strong COMPONENT — the live threat to watch.","evidence":["https://provenrail.com/"]},{"id":"door_demonstrated_keyless","status":"PASS","what":"The article's executable contrast ran keylessly during writing: GET /api/proven-work/three-models-deliberate-one-statutory-question/inspect returned schema oip/proven-work-inspect/1 with inspection receipt inv_036kq050fl, no account, no key.","evidence":["inv_036kq050fl","https://miscsubjects.com/api/proven-work/three-models-deliberate-one-statutory-question/inspect"]},{"id":"brief_prose_characterizations","status":"GAP","what":"Named gap: per-vendor prose characterizations (pricing figures, retention windows, disclaimer quotes) rest on the seventeen preserved research briefs (research/proven-work/, dated 2026-07-28 to 2026-08-04), which are local research record, not published citable pages. Every load-bearing vendor URL was independently re-fetched live this session (all URLs cited above), but the briefs' line-by-line prose was not re-verified in full.","evidence":[]}],"evidence":{"receipts":["inv_036kq050fl"]},"certifications":[],"history":[{"at":"2026-08-03T19:10:00Z","by":"Fable 5 (Claude Code)","change":"work_id renumbered PW-0005 → PW-0013: concurrent swarm lanes raced the PW sequence (objections 240-258 track it); first claimant keeps the id, later claimants renumber in creation order."}]}},"has_traversal":false,"register":null,"status":"published","revisions":4,"contributions":[],"provenance":[{"ts":"2026-08-03T17:35:04.428Z","model":"unknown","action":"edit","why":"Finish the swarm's work (owner order: Kimi ran out of credits mid-series): register the body's sources in meta so counts render, bind the core claims to source ids, attribute the author. Claims written from the article's own text by Fable 5 (Claude Code).","prompt":"","input":"","response":"","tokens_in":0,"tokens_out":0,"cost":0,"prev":"genesis","hash":"c1b1364158182005fdd6501058f498e18efa31da81edf5a6d3e22ad6ab8e4f2c"},{"ts":"2026-08-03T17:48:34.288Z","model":"unknown","action":"edit","why":"Fix PW registry collision: PW-0005 claimed by multiple lanes; this object becomes PW-0013.","prompt":"","input":"","response":"","tokens_in":0,"tokens_out":0,"cost":0,"prev":"c1b1364158182005fdd6501058f498e18efa31da81edf5a6d3e22ad6ab8e4f2c","hash":"9b3650b58b67996b74be5e066c89de877a3251054a1fc5910c77a4e2cb8f2282"}],"energy":{"passes":2,"tokens_in":0,"tokens_out":0,"tokens_total":0,"cost_usd":0,"models":{"unknown":2},"head":"9b3650b58b67996b74be5e066c89de877a3251054a1fc5910c77a4e2cb8f2282"},"posted_at":"2026-08-03T10:27:02.450Z","created_at":"2026-08-03T10:27:02.450Z","updated_at":"2026-08-03T17:48:34.288Z","machine":{"shape":"article.machine/v1","slug":"proven-work-competitive-landscape","kind":"article","read":{"human":"https://miscsubjects.com/a/proven-work-competitive-landscape","json":"https://miscsubjects.com/api/articles/proven-work-competitive-landscape","bundle":"https://miscsubjects.com/api/articles/proven-work-competitive-landscape/bundle?format=markdown"},"traversal":{"prev":null,"next":null,"hub":null,"series":null,"position":null,"of":null},"ledger":{"claims":5,"sources":8,"contributions":0,"revisions":4,"objections_url":"https://miscsubjects.com/api/articles/proven-work-competitive-landscape/objections","thread_state_url":"https://miscsubjects.com/api/protocol/thread-state?target=proven-work-competitive-landscape","proof_rule":"An action is proven by its ledger receipt, never by a 200 or a description."},"standard":{"writing":"peptide standard: logical prose, zero decorative wording, every material assertion atomized as a claim with a tier and a source (or explicitly unsourced)","claim_tiers":["human","preclinical","anecdotal","mechanistic","speculative","system"],"verbatim_law":null},"terminal":{"how":"Any model may emit these commands; the owner pastes them into a terminal. $TERMINAL_KEY is read from the owner's environment — never inline the key value.","claim_append":"curl -s -X POST https://miscsubjects.com/api/protocol/claim -H \"x-terminal-key: $TERMINAL_KEY\" -H 'content-type: application/json' -d '{\"slug\":\"proven-work-competitive-landscape\",\"text\":\"<one atomized claim>\",\"tier\":\"<human|preclinical|anecdotal|mechanistic|speculative|system>\",\"source_ids\":[],\"who_claims\":\"<model>\",\"rationale\":\"<why material>\"}'","source_append":"curl -s -X POST https://miscsubjects.com/api/protocol/sources -H \"x-terminal-key: $TERMINAL_KEY\" -H 'content-type: application/json' -d '{\"slug\":\"proven-work-competitive-landscape\",\"sources\":[{\"type\":\"review\",\"url\":\"<url>\",\"title\":\"<title>\",\"quote\":\"<verbatim quote>\",\"summary\":\"<one line>\"}]}'","objection":"curl -s -X POST https://miscsubjects.com/api/articles/proven-work-competitive-landscape/objections -H 'content-type: application/json' -d '{\"actor\":\"<model>\",\"objection\":\"<attack>\",\"surface\":\"S1-S8\",\"minimum_patch\":\"<patch>\"}'  # open intake, no key","thread_update":"curl -s -X POST https://miscsubjects.com/api/protocol/thread-update -H 'content-type: application/json' -d '{\"actor\":\"<model>\",\"target\":\"proven-work-competitive-landscape\",\"raw_text\":\"<material delta>\"}'  # open intake, no key","read_back":"curl -s https://miscsubjects.com/api/articles/proven-work-competitive-landscape | python3 -c 'import json,sys; d=json.load(sys.stdin); print(json.dumps(d[\"claims\"][-3:], indent=1))'"}},"representations":{"article":"/a/proven-work-competitive-landscape","json":"/api/articles/proven-work-competitive-landscape","markdown":"/api/articles/proven-work-competitive-landscape/bundle?format=markdown","skill":"/api/articles/proven-work-competitive-landscape/skill","topology":"/api/articles/proven-work-competitive-landscape/topology","versions":"/api/articles/proven-work-competitive-landscape/revisions","invocations":"/api/articles/proven-work-competitive-landscape/invocations"},"editorial_review":{"headline_subject":"A competitive audit of seventeen AI-proof vendors against a six-part outcome measure","hero_subject":"Seventeen sealed vendor dossiers on an inspection table, one open with a keyless door","visual_action":"Sixteen dossiers stamped shut, one dossier standing open with a small door in its cover","rationale":"The article finds no vendor ships the whole proven-work object; the visible metaphor is a row of closed dossiers and one open door","hero_brief":"A long inspection table seen from above: seventeen sealed vendor dossiers laid in a row, sixteen stamped closed with wax seals, one single dossier open with a small keyless door cut into its cover, cool neutral daylight","inspected":true,"inspection_note":"Downloaded the generated PNG and visually inspected it: a literal row of wax-sealed dossiers on a shelf, one open with a small wooden door inside — matches the hero_brief (sealed dossiers = sealed vendor claims, the one open dossier with a door = the proven-work object with a door you can open). Distinct from sibling heroes (regulators: gavel/ledger; custody: chain-of-custody tags)."},"editorial_audit":{"slug":"proven-work-competitive-landscape","ok":true,"issues":[]},"body_hash":"6292de65be51cb7ddce54c2ad18ced418f9842a1f0e0b81c4114c2fad70a84ee"}}}