Archived dispatch

How does Keryx evidence checks before citation rewards work in practice, and which published material supports the explanation?

Lowconfidence— Each summary sentence is tied to a verbatim excerpt and model-checked; completeness and independent factual correctness remain unverified. Evidence assessment: 1 evidence-verified source cover every sub-claim, but corroboration or support strength is limited

· llm:deepseek:deepseek-v4-pro

§ IIThe reading1 cited
Lowconfidence— Each summary sentence is tied to a verbatim excerpt and model-checked; completeness and independent factual correctness remain unverified. Evidence assessment: 1 evidence-verified source cover every sub-claim, but corroboration or support strength is limiteddeep researchpreview plan 2/2 claimsportfolio 2/2 · evidence 50%

> ⚠ Low confidence — Each summary sentence is tied to a verbatim excerpt and model-checked; completeness and independent factual correctness remain unverified. Evidence assessment: 1 evidence-verified source cover every sub-claim, but corroboration or support strength is limited. Treat this as provisional.

Model-written summary with sentence-level citations. Each sentence is followed by the verbatim source text it was checked against; sentences without a qualifying excerpt were removed.

Research target 1

Requested topic (unverified): “How does Keryx perform evidence checks before citation rewards in practice?”

Keryx considers a citation reward only after the answer is written and its evidence has been checked. Source text: “A citation reward is a separate payment considered after the answer is written and its evidence checked.”

A citation qualifies for a reward only when its source marker appears in the answer, the synthesis declares that marker, and a validated quotation has support of at least 0.4. Source text: “A citation can qualify for a reward only when its source marker appears in the answer, the synthesis declares that marker, and a validated quotation has support of at least 0.4.”

Research target 2

Requested topic (unverified): “Which published material supports the explanation of Keryx's evidence checks before citation rewards?”

The explanation is supported by first-party engineering documentation from Keryx based on repository commit 9ea84fa. Source text: “This is first-party engineering documentation based on repository commit 9ea84fa, not independent reporting or customer evidence.”

Summary sentences are model-written and model-checked, not independently verified; they claim no more than their excerpts and are not a complete synthesis. Excerpts establish source grounding, not that a source is correct. Source statements may be wrong or conflicting. Payment states remain in the separate receipt.

Evidence ledger — recorded source excerpts

Research targets are unverified topics. Coverage is an estimate of excerpt support, not proof of entailment, factual truth or a complete answer.

  1. Requested topic (unverified): “How does Keryx perform evidence checks before citation rewards in practice?”

    90% estimated
    “A citation reward is a separate payment considered after the answer is written and its evidence checked.” [S1] How Keryx pays cited creators
    “A citation can qualify for a reward only when its source marker appears in the answer, the synthesis declares that marker, and a validated quotation has support of at least 0.4.” [S1] How Keryx pays cited creators
  2. Requested topic (unverified): “Which published material supports the explanation of Keryx's evidence checks before citation rewards?”

    80% estimated
    “This is first-party engineering documentation based on repository commit 9ea84fa, not independent reporting or customer evidence.” [S1] How Keryx pays cited creators
What if a source were missing?

Temporarily leave out one source to see which research targets retain excerpts in this report.

Showing the original excerpt ledger.

  1. How does Keryx perform evidence checks before citation rewards in practice?

    2 recorded excerpts remain.

    Inspect remaining excerpts

    “A citation reward is a separate payment considered after the answer is written and its evidence checked.”

    S1 · Keryx Engineering (first-party) · How Keryx pays cited creators · version sha256:0ee74abc5d0a57df70614ed65bf58ad2a681def7f2a3ea6a49831ba68fc31d76

    “A citation can qualify for a reward only when its source marker appears in the answer, the synthesis declares that marker, and a validated quotation has support of at least 0.4.”

    S1 · Keryx Engineering (first-party) · How Keryx pays cited creators · version sha256:0ee74abc5d0a57df70614ed65bf58ad2a681def7f2a3ea6a49831ba68fc31d76

  2. Which published material supports the explanation of Keryx's evidence checks before citation rewards?

    1 recorded excerpt remain.

    Inspect remaining excerpts

    “This is first-party engineering documentation based on repository commit 9ea84fa, not independent reporting or customer evidence.”

    S1 · Keryx Engineering (first-party) · How Keryx pays cited creators · version sha256:0ee74abc5d0a57df70614ed65bf58ad2a681def7f2a3ea6a49831ba68fc31d76

Targets are requested topics, not verified assertions. Excerpts do not prove truth or independent corroboration. This view keeps the answer, confidence and payments unchanged and makes no new requests.

Research evidence matrix

Compare unverified research targets with cited sources and inspect recorded excerpts. An empty cell means no inspectable excerpt was recorded; it does not establish whether a claim is true, false, or disputed. Coverage and agent confidence do not prove entailment, measured accuracy or complete synthesis.

Research target by cited source evidence matrix
Research target (unverified)Inspection status[S1] How Keryx pays cited creatorsPublication: Keryx Engineering (first-party)Published: 2026-09-08T00:00:00.000Z
How does Keryx perform evidence checks before citation rewards in practice?Recorded excerpt
Inspect 2 excerpts
A citation reward is a separate payment considered after the answer is written and its evidence checked.
A citation can qualify for a reward only when its source marker appears in the answer, the synthesis declares that marker, and a validated quotation has support of at least 0.4.
Which published material supports the explanation of Keryx's evidence checks before citation rewards?Recorded excerpt
Inspect 1 excerpt
This is first-party engineering documentation based on repository commit 9ea84fa, not independent reporting or customer evidence.

Reference export

1 article references. Recorded titles, links and dates; observed scholarly records also include supplied authors, DOI and journal metadata with read limits. Review metadata before using in a paper. Import RIS into Zotero with File → Import.

Cited sources and references

Helpful?

Loading feedback…

Spent$0.025
To creators100%
Decisions0 bought · 2 cached · 30 skipped
llm:deepseek:deepseek-v4-prolive on Arc mainnet
Decision log · 59 steps
§ IThe decision$0.025 settled / $0.05
50%$0.025 under cap
Decompose

Breaking down: "How does Keryx evidence checks before citation rewards work in practice, and which published material supports the explanation?"

Decompose

Identified 2 research target(s) to investigate; these are not established facts

Decompose

Deep mode: up to 4 paid/cached/public reads plus one bounded gap-expansion pass when needed.

Discover

Web search: 3/3 planned queries attempted, 0 succeeded, 0 public page previews/leads, 3 unavailable queries. Previews and supplied URLs are discovery only. Public reads spend no USDC; model and service operating costs remain separate.

Discover

Discovered 2 verified creator source(s) and 30 free public reference(s)

Pre-check

Claim-aware portfolio (exhaustive; bounded selection, not a claim of global optimality) selected 2/2 positive proposal(s): 2 free/cache selections + 0 paid fresh selections, predicting 2/2 claim(s) above the evidence floor with $0.000000/$0.025000 fetch USDC reserved.

Pre-check

Free-preview pre-check maps an actionable source to every sub-claim (2/2); paid reading may proceed within the budget.

DecideCACHE
Keryx Engineering (first-party) — How Keryx pays cited creators$0.002 · EV 80%

First-party Keryx engineering documentation specifically covers citation rewards, exact-quote evidence checks, and contribution weights, directly addressing both how evidence checks work (target 0) and which published material supports the explanation (target 1). — selected for the claim-aware evidence portfolio (targets claims 1, 2; 0 fetch USDC, 1 attention slot).

DecideCACHE
Keryx owner release notes — first-party validation — Keryx v0.27.63$0.005 · EV 60%

Official Keryx release notes mention requesting missing evidence and comparing signed provider offers, which could help investigate how evidence checks work in practice (target 0). — selected for the claim-aware evidence portfolio (targets claim 1; 0 fetch USDC, 1 attention slot).

DecideSKIP
arXiv - Fair Moderation, Equitable Access, and AI: arXiv’s Updated Rate Limit Policy$0 · EV 0%

Article is about arXiv rate limit policy and AI, unrelated to Keryx evidence checks or citation rewards. - free public feed reference; no purchase or creator reward.

DecideSKIP
Bank for International Settlements - Chasing El Dorado: gold under shifting geopolitical and financial conditions$0 · EV 0%

BIS paper on gold prices and geopolitical risks, no connection to Keryx or citation rewards. - free public feed reference; no purchase or creator reward.

DecideSKIP
Chip Huyen - Agents$0 · EV 0%

General article on AI agents, not about Keryx evidence checks or citation rewards. - free public feed reference; no purchase or creator reward.

DecideSKIP
Cloudflare Workers - How we rebuilt Cloudflare Workers’ module registry for Node.js compatibility$0 · EV 0%

Cloudflare Workers module registry engineering post, unrelated to Keryx. - free public feed reference; no purchase or creator reward.

DecideSKIP
Creative Commons - From Reflection to Action: 8 Hot Takes from CC’s Open Heritage Roundtable$0 · EV 0%

Creative Commons heritage roundtable discussion, no relevance to Keryx citation rewards. - free public feed reference; no purchase or creator reward.

DecideSKIP
Google DeepMind - Advancing Private AI Compute with secure, server-side memory$0 · EV 0%

DeepMind article on private AI compute, unrelated to Keryx. - free public feed reference; no purchase or creator reward.

DecideSKIP
Directory of Open Access Journals - Paradise Lost, Knowledge Regained: Jan Erik Frantsvåg on reclaiming open access (Part 1)$0 · EV 0%

DOAJ interview about open access, not about Keryx evidence checks. - free public feed reference; no purchase or creator reward.

DecideSKIP
Docker - Meet the Ecosystem: Partners and Customers at WeAreDevelopers with Docker$0 · EV 0%

Docker ecosystem event post, no connection to Keryx. - free public feed reference; no purchase or creator reward.

DecideSKIP
DuckDB - A DuckDB Database with No Data in It$0 · EV 0%

DuckDB database technical article, unrelated to Keryx citation rewards. - free public feed reference; no purchase or creator reward.

DecideSKIP
Eugene Yan - How to Work and Compound with AI$0 · EV 0%

Article on working with AI, not about Keryx evidence checks. - free public feed reference; no purchase or creator reward.

DecideSKIP
The Go Blog - Platform-independent SIMD in Go$0 · EV 0%

Go blog post on SIMD, unrelated to Keryx. - free public feed reference; no purchase or creator reward.

DecideSKIP
Google Research - The future of practice: Enabling teachers to create learning interactives with generative UI$0 · EV 0%

Google Research post on generative UI for teachers, no relevance to Keryx. - free public feed reference; no purchase or creator reward.

DecideSKIP
Lilian Weng - LLM Powered Autonomous Agents$0 · EV 0%

General LLM agents article, not about Keryx evidence checks or citation rewards. - free public feed reference; no purchase or creator reward.

DecideSKIP
Microsoft Research - One year in: How Microsoft Research Asia – Singapore is advancing research, partnership and talent for real-world impact$0 · EV 0%

Microsoft Research Asia Singapore lab update, unrelated to Keryx. - free public feed reference; no purchase or creator reward.

DecideSKIP
Netlify - Compete in OpenAI’s WebMCP Challenge with Netlify$0 · EV 0%

Netlify WebMCP challenge announcement, no connection to Keryx. - free public feed reference; no purchase or creator reward.

DecideSKIP
Ollama - Ollama now supports Jev-style decision models$0 · EV 0%

Ollama decision models announcement, unrelated to Keryx citation rewards. - free public feed reference; no purchase or creator reward.

DecideSKIP
OpenAI Developers - OpenAI Developers plugin$0 · EV 0%

OpenAI plugin guide, not about Keryx. - free public feed reference; no purchase or creator reward.

DecideSKIP
OpenAlex - When affiliation errors become a research security problem$0 · EV 0%

OpenAlex post on affiliation errors, no relevance to Keryx evidence checks. - free public feed reference; no purchase or creator reward.

DecideSKIP
PostgreSQL - pgx-bm25 1.0: BM25 ranked full-text search as a native PostgreSQL index$0 · EV 0%

PostgreSQL full-text search extension, unrelated to Keryx. - free public feed reference; no purchase or creator reward.

DecideSKIP
PyTorch - New Pathway to PyTorch Certified Associate (PTCA) Certification$0 · EV 0%

PyTorch certification pathway, no connection to Keryx. - free public feed reference; no purchase or creator reward.

DecideSKIP
Rust Blog - Announcing Rust 1.98.1$0 · EV 0%

Rust release announcement, unrelated to Keryx. - free public feed reference; no purchase or creator reward.

DecideSKIP
Sebastian Raschka - Focusing on Post-Training$0 · EV 0%

Post-training LLM article, not about Keryx evidence checks. - free public feed reference; no purchase or creator reward.

DecideSKIP
Spotify Engineering - Portal by Spotify cut my Claude Code token usage by 90%$0 · EV 0%

Spotify engineering post on token usage, no relevance to Keryx. - free public feed reference; no purchase or creator reward.

DecideSKIP
Stripe Blog - OUSD is now the default stablecoin on Stripe$0 · EV 0%

Stripe stablecoin announcement, unrelated to Keryx citation rewards. - free public feed reference; no purchase or creator reward.

DecideSKIP
Supabase - Build anything: Supabase from code, and an MCP server for your app$0 · EV 0%

Supabase product announcement, no connection to Keryx. - free public feed reference; no purchase or creator reward.

DecideSKIP
Tailscale - Your agent, your network: How Meta’s Muse agent works with Tailscale$0 · EV 0%

Tailscale networking post, unrelated to Keryx evidence checks. - free public feed reference; no purchase or creator reward.

DecideSKIP
Vicki Boykis - NASA Elements of Engineering Excellence$0 · EV 0%

Article is about NASA engineering excellence, unrelated to Keryx evidence checks or citation rewards. - free public feed reference; no purchase or creator reward.

DecideSKIP
vLLM - Taking vLLM Apart: A Practical Guide to Disaggregated Serving$0 · EV 0%

Article is about vLLM disaggregated serving, unrelated to Keryx evidence checks or citation rewards. - free public feed reference; no purchase or creator reward.

DecideSKIP
Wikimedia Diff - Inaugural Wikimedia Kimaragang Online Meetup: Strengthening Community, Empowering Digital Knowledge$0 · EV 0%

Article is about a Wikimedia community meetup, unrelated to Keryx evidence checks or citation rewards. - free public feed reference; no purchase or creator reward.

DecideSKIP
x402 - Linux Foundation Announces Operational Launch of x402 Foundation to Standardize Internet-Native Payments for AI Agents and Applications$0 · EV 0%

Article is about the x402 Foundation launch, unrelated to Keryx evidence checks or citation rewards. - free public feed reference; no purchase or creator reward.

Fetch

Agent spend wallet ready: 0x9e88d16c3ecb07b9e84df13ef38dd6b673bd285c (balance sufficient)

Fetch

Reused cached Keryx Engineering (first-party) — How Keryx pays cited creators (free) — S1

Fetch

Reused cached Keryx owner release notes — first-party validation — Keryx v0.27.63 (free) — S2

Re-evaluate

Sub-claim "How does Keryx perform evidence checks before citation rewar…": 90% covered by S1 — S1 passages describe the practical flow: the model proposes evidence records (research-question index, source marker, exact quote, estimated support score); the deterministic evidence ledger checks these proposals against the original content actually gathered; a citation qualifies for reward only when the source marker appears in the answer, the synthesis declares that marker, and a validated quotation has support >= 0.4; if the final evidence-assessment transport fails, reward authorization is disabled. This directly answers the practical mechanism. Minor gap: the exact implementation details of the 'deterministic evidence ledger' checks (e.g., how exact quote matching is performed) are not described, but the requested 'in practice' explanation is substantially covered.

Re-evaluate

Sub-claim "Which published material supports the explanation of Keryx's…": 100% covered by S1, S2 — S1 is first-party engineering documentation titled 'How Keryx pays cited creators' with explicit passages describing evidence checks before citation rewards. S2 is a first-party release note mentioning 'request missing evidence and compare signed provider offers', which supports the existence of evidence-checking features in the system. Together they identify the published material supporting the explanation.

Re-evaluate

Coverage for both sub-claims is high (0.9 and 1.0). The first-party documentation (S1) directly explains the practical evidence-check process before citation rewards, and S2 corroborates the feature. No affordable skipped sources are relevant to this specific question; remaining budget is minimal and additional purchases would not materially improve coverage.

Sufficiency

Final check — "How does Keryx perform evidence checks before citation rewar…": 100% assessed by S1

Sufficiency

Final check — "Which published material supports the explanation of Keryx's…": 100% assessed by S1

Sufficiency

Final coverage assessment — S1 provides first-party engineering documentation that directly describes the evidence check process before citation rewards: the model proposes evidence records, the deterministic evidence ledger checks them against gathered content, and a citation qualifies only if the marker appears in the answer, the synthesis declares it, and a validated quotation has support >= 0.4. S2 is a release note mentioning 'request missing evidence and compare signed provider offers' but does not explain the evidence check process in practice. Therefore, S1 fully supports the first sub-claim, and S1 is the published material supporting the explanation. S2 provides only tangential context.

Synthesize

Synthesizing a grounded answer from 2 source(s)…

Evidence

Relevance review returned; only checked excerpts can retain support, and review cannot raise it.

Evidence

Delivering 3 summary sentence(s), each tied to one checked verbatim excerpt; complete synthesis remains unverified.

Evidence

Source-matched reward-eligible excerpt — S1, research target 1, proposed support 90% (estimate, not entailment): “A citation reward is a separate payment considered after the answer is written and its evidence checked.”

Evidence

Source-matched reward-eligible excerpt — S1, research target 1, proposed support 90% (estimate, not entailment): “A citation can qualify for a reward only when its source marker appears in the answer, the synthesis declares that marker, and a validated q…”

Evidence

Source-matched reward-eligible excerpt — S1, research target 2, proposed support 80% (estimate, not entailment): “This is first-party engineering documentation based on repository commit 9ea84fa, not independent reporting or customer evidence.”

Synthesize

Prepared a sentence-cited summary from 1 source(s)

Verdict

Confidence: Low — Each summary sentence is tied to a verbatim excerpt and model-checked; completeness and independent factual correctness remain unverified. Evidence assessment: 1 evidence-verified source cover every sub-claim, but corroboration or support strength is limited.

Attribute

Keryx Engineering (first-party) contributed 100% - reward $0.025

Settle

Settled $0.025 citation reward → 0x3d609eF127094624cC434aAd5A9fAcA0851dAa2D (5ec2043d-3…)

Done

Done. Spent $0.025 across 1 confirmed/simulated payment(s) to creators.

Sign in with the original report owner to review its decisions.

Read checkpoints

Read checkpoint evidence is unavailable for this report. Historical, private and unsupported native runs are not reconstructed.

Portable research receipt

Take the evidence trail with you

One deterministic JSON bundle binds the answer, visible decisions, exact article versions, claim evidence and a Circle-settlement snapshot under SHA-256. Retain the digest to detect later changes; the self-check is not a publisher or Keryx signature.

Manage question privacy

Recorded purchase outcomes

Did the exact article versions bought for this answer appear in its citations? This view scores retained BUY decisions against payment observations recorded in the same dispatch trace.

Recorded settlement bookkeeping only; Circle and chain settlement have not been independently rechecked here.

This is a partial retained trace sample, not a full payment ledger. An absent matching payment does not prove that no payment occurred; excluded and unconfirmed costs remain unknown.

Recorded network
eip155:5042
Dispatch recorded at
2026-10-13T07:02:15.946Z

Participant cohort: unknown. No outside-customer usage is inferred.

0 scored exact-version purchases from 0 recorded BUY decisions; 0 BUY decisions unscored. 1 of 1 trace payment observations excluded.

Each unique source + item + content version with positive matching recorded settled access counts once. Duplicate BUYs, missing identities, zero/unconfirmed access and incompatible payment observations cannot create a scored purchase. CACHE and SKIP decisions are outside this sample.

Exact-version citation hit rate
Unmeasured

No scored purchases; the hit rate is unmeasured.

Recorded access cost of uncited purchases
Unmeasured

Counts only scored purchases. Descriptive access cost, not causal regret or proof that the purchase was useless.

Predicted value and observed citations

Bins compare the recorded predicted value with citation occurrence in this one dispatch. Small samples do not validate a probability model.

Value bandPurchasesPredicted mean (0–1)Citation rate
0–0.20UnmeasuredUnmeasured
0.2–0.40UnmeasuredUnmeasured
0.4–0.60UnmeasuredUnmeasured
0.6–0.80UnmeasuredUnmeasured
0.8–10UnmeasuredUnmeasured

Missed value, cost per supported claim, budget alternatives and counterfactual outcomes are unmeasured. No extra reading or learning was performed.

Exact receipt still current

1 exact cited article version still match Keryx's current index. The source cited here has published nothing new since this dispatch settled.

Inspect machine-readable audit

Ask a follow-upNew dispatch · creators paid again

Carries this dispatch’s question as context — never its answer. The next dispatch is read from sources bought for it.

From the archive

Related dispatches