Stabilarity Research Platform Is Now Open — Free API Access for All Researchers

Posted on by Admin
DOI: 10.5281/zenodo.18928330  74stabilfr·wuedcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources35%○≥80% from editorially reviewed sources
[t]Trusted94%✓≥80% from verified, high-quality sources
[a]DOI41%○≥80% have a Digital Object Identifier
[b]CrossRef24%○≥80% indexed in CrossRef
[i]Indexed82%✓≥80% have metadata indexed
[l]Academic41%○≥80% from journals/conferences/preprints
[f]Free Access65%○≥80% are freely accessible
[r]References17 refs✓Minimum 10 references required
[w]Words [REQ]1,790✓Minimum 1,000 words of article body. Current: 1,790
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]7✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 7
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.18928330
[c]Data Charts0○Original data charts from reproducible analysis (min 2). Current: 0
[g]Code—○Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (69 × 60%) + Required (4/4 × 30%) + Optional (1/4 × 10%)

This paper presents the Stabilarity Research Platform — an open, API-accessible research infrastructure exposing validated machine learning models, geopolitical risk datasets, and decision optimization tools to the global research community at no cost. The platform implements FAIR data principles (Wilkinson et al., 2016), providing composable, versioned endpoints for: (1) medical imaging classi...

Show moreHide
DOI: 10.5281/zenodo.18928330 74stabilfr·wuedcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources35%○≥80% from editorially reviewed sources
[t]Trusted94%✓≥80% from verified, high-quality sources
[a]DOI41%○≥80% have a Digital Object Identifier
[b]CrossRef24%○≥80% indexed in CrossRef
[i]Indexed82%✓≥80% have metadata indexed
[l]Academic41%○≥80% from journals/conferences/preprints
[f]Free Access65%○≥80% are freely accessible
[r]References17 refs✓Minimum 10 references required
[w]Words [REQ]1,790✓Minimum 1,000 words of article body. Current: 1,790
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]7✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 7
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.18928330
[c]Data Charts0○Original data charts from reproducible analysis (min 2). Current: 0
[g]Code—○Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (69 × 60%) + Required (4/4 × 30%) + Optional (1/4 × 10%)
Read More
Read more

Living NPCs: Zero-Script LLM Characters on a Shared Engine — Live Results from a 2D Village and a 3D Mage Isle

Posted on by
DOI: 10.5281/zenodo.23627899  79stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources15%○≥80% from editorially reviewed sources
[t]Trusted85%✓≥80% from verified, high-quality sources
[a]DOI65%○≥80% have a Digital Object Identifier
[b]CrossRef15%○≥80% indexed in CrossRef
[i]Indexed58%○≥80% have metadata indexed
[l]Academic69%○≥80% from journals/conferences/preprints
[f]Free Access81%✓≥80% are freely accessible
[r]References26 refs✓Minimum 10 references required
[w]Words [REQ]4,407✓Minimum 1,000 words of article body. Current: 4,407
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]21✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 21
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23627899
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness73%✓≥60% of references from 2025–2026. Current: 73%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams4✓Mermaid architecture/flow diagrams. Current: 4
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (70 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)

Can a game character run with no scripted behaviour at all, so that every step, word, blow and spell starts from one language-model decision, and still be fast, cheap and sensible enough for real-time play? We built a shared engine for such characters and ran two games on it: Oakhollow, a 2D medieval village, and Quarthaven, a 3D isle of twelve elemental mages. On private copies of both games, ...

Show moreHide
DOI: 10.5281/zenodo.23627899 79stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources15%○≥80% from editorially reviewed sources
[t]Trusted85%✓≥80% from verified, high-quality sources
[a]DOI65%○≥80% have a Digital Object Identifier
[b]CrossRef15%○≥80% indexed in CrossRef
[i]Indexed58%○≥80% have metadata indexed
[l]Academic69%○≥80% from journals/conferences/preprints
[f]Free Access81%✓≥80% are freely accessible
[r]References26 refs✓Minimum 10 references required
[w]Words [REQ]4,407✓Minimum 1,000 words of article body. Current: 4,407
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]21✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 21
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23627899
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness73%✓≥60% of references from 2025–2026. Current: 73%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams4✓Mermaid architecture/flow diagrams. Current: 4
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (70 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)
Read More
Read more

Living NPCs: A Classification of AI Integration Approaches for Game Characters

Posted on by
DOI: 10.5281/zenodo.23602008  79stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources8%○≥80% from editorially reviewed sources
[t]Trusted82%✓≥80% from verified, high-quality sources
[a]DOI66%○≥80% have a Digital Object Identifier
[b]CrossRef8%○≥80% indexed in CrossRef
[i]Indexed58%○≥80% have metadata indexed
[l]Academic68%○≥80% from journals/conferences/preprints
[f]Free Access82%✓≥80% are freely accessible
[r]References38 refs✓Minimum 10 references required
[w]Words [REQ]4,137✓Minimum 1,000 words of article body. Current: 4,137
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]33✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 33
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23602008
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness74%✓≥60% of references from 2025–2026. Current: 74%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams4✓Mermaid architecture/flow diagrams. Current: 4
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (69 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)

Where should a language model attach to a game, and which faculties of a character should it control? We propose a two-axis classification: the integration surface (a mod of a shipped game, an engine SDK, an agent on an authoritative server, or an agent that reads pixels and presses keys) and seven augmentation types graded 0 to 3 (dialogue, vision, hearing, memory, time perception, action, soc...

Show moreHide
DOI: 10.5281/zenodo.23602008 79stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources8%○≥80% from editorially reviewed sources
[t]Trusted82%✓≥80% from verified, high-quality sources
[a]DOI66%○≥80% have a Digital Object Identifier
[b]CrossRef8%○≥80% indexed in CrossRef
[i]Indexed58%○≥80% have metadata indexed
[l]Academic68%○≥80% from journals/conferences/preprints
[f]Free Access82%✓≥80% are freely accessible
[r]References38 refs✓Minimum 10 references required
[w]Words [REQ]4,137✓Minimum 1,000 words of article body. Current: 4,137
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]33✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 33
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23602008
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness74%✓≥60% of references from 2025–2026. Current: 74%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams4✓Mermaid architecture/flow diagrams. Current: 4
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (69 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)
Read More
Read more

How the KV Cache Works: Why Cached Tokens Stay Valid When Later Tokens Change Their Meaning

Posted on by
Applied Research
Applied Research by Oleh Ivchenko  ·  DOI: 10.5281/zenodo.23569111  89stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources20%○≥80% from editorially reviewed sources
[t]Trusted97%✓≥80% from verified, high-quality sources
[a]DOI86%✓≥80% have a Digital Object Identifier
[b]CrossRef20%○≥80% indexed in CrossRef
[i]Indexed77%○≥80% have metadata indexed
[l]Academic89%✓≥80% from journals/conferences/preprints
[f]Free Access94%✓≥80% are freely accessible
[r]References35 refs✓Minimum 10 references required
[w]Words [REQ]4,011✓Minimum 1,000 words of article body. Current: 4,011
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]30✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 30
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23569111
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness74%✓≥60% of references from 2025–2026. Current: 74%
[c]Data Charts5✓Original data charts from reproducible analysis (min 2). Current: 5
[g]Code✓✓Source code available on GitHub
[m]Diagrams4✓Mermaid architecture/flow diagrams. Current: 4
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (86 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)

Decoder-only language models keep a key-value (KV) cache: the key and value vectors of every token read so far, reused at each generation step. A common worry is that the cache goes stale when later text changes what an earlier word means, as in "The bank" followed by "of the river" or "approved the loan". We ask whether the cache is exact, where re-interpretation happens, and where exactness e...

Show moreHide
Applied Research by Oleh Ivchenko DOI: 10.5281/zenodo.23569111 89stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources20%○≥80% from editorially reviewed sources
[t]Trusted97%✓≥80% from verified, high-quality sources
[a]DOI86%✓≥80% have a Digital Object Identifier
[b]CrossRef20%○≥80% indexed in CrossRef
[i]Indexed77%○≥80% have metadata indexed
[l]Academic89%✓≥80% from journals/conferences/preprints
[f]Free Access94%✓≥80% are freely accessible
[r]References35 refs✓Minimum 10 references required
[w]Words [REQ]4,011✓Minimum 1,000 words of article body. Current: 4,011
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]30✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 30
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23569111
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness74%✓≥60% of references from 2025–2026. Current: 74%
[c]Data Charts5✓Original data charts from reproducible analysis (min 2). Current: 5
[g]Code✓✓Source code available on GitHub
[m]Diagrams4✓Mermaid architecture/flow diagrams. Current: 4
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (86 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)
Cost-Effective Ent…Read More
Read more

Open-Source Model Auditing Infrastructure: Third-Party Verification of Training Data Claims

Posted on by
Open Source Research
Open Source Research by Oleh Ivchenko  ·  DOI: 10.5281/zenodo.23560178  86stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources19%○≥80% from editorially reviewed sources
[t]Trusted91%✓≥80% from verified, high-quality sources
[a]DOI78%○≥80% have a Digital Object Identifier
[b]CrossRef19%○≥80% indexed in CrossRef
[i]Indexed75%○≥80% have metadata indexed
[l]Academic81%✓≥80% from journals/conferences/preprints
[f]Free Access100%✓≥80% are freely accessible
[r]References32 refs✓Minimum 10 references required
[w]Words [REQ]4,314✓Minimum 1,000 words of article body. Current: 4,314
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]27✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 27
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23560178
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness64%✓≥60% of references from 2025–2026. Current: 64%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (81 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)

Open-weight models let anyone compute token probabilities, but their training corpora are rarely public, so claims about training data are hard to check. We ask what a third party holding only the weights can verify. We scored 2,400 matched member and non-member documents from the Pile train and test splits, 400 Wikipedia articles created in October 2026 and the WikiMIA benchmark with three Pyt...

Show moreHide
Open Source Research by Oleh Ivchenko DOI: 10.5281/zenodo.23560178 86stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources19%○≥80% from editorially reviewed sources
[t]Trusted91%✓≥80% from verified, high-quality sources
[a]DOI78%○≥80% have a Digital Object Identifier
[b]CrossRef19%○≥80% indexed in CrossRef
[i]Indexed75%○≥80% have metadata indexed
[l]Academic81%✓≥80% from journals/conferences/preprints
[f]Free Access100%✓≥80% are freely accessible
[r]References32 refs✓Minimum 10 references required
[w]Words [REQ]4,314✓Minimum 1,000 words of article body. Current: 4,314
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]27✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 27
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23560178
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness64%✓≥60% of references from 2025–2026. Current: 64%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (81 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)
Trusted Open SourceRead More
Read more

Cross-Agent Traceability: Propagating Observability Context Through Multi-Agent Tool Chains

Posted on by
Technical Research
Technical Research by Oleh Ivchenko  ·  DOI: 10.5281/zenodo.23544775  80stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources3%○≥80% from editorially reviewed sources
[t]Trusted83%✓≥80% from verified, high-quality sources
[a]DOI68%○≥80% have a Digital Object Identifier
[b]CrossRef3%○≥80% indexed in CrossRef
[i]Indexed65%○≥80% have metadata indexed
[l]Academic70%○≥80% from journals/conferences/preprints
[f]Free Access83%✓≥80% are freely accessible
[r]References40 refs✓Minimum 10 references required
[w]Words [REQ]3,959✓Minimum 1,000 words of article body. Current: 3,959
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]35✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 35
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23544775
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness97%✓≥60% of references from 2025–2026. Current: 97%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (71 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)

When one user request fans out across agents, tool servers, queues and callbacks, one missing trace header splits its evidence into traces that no backend can rejoin by trace ID. We built a two-agent, six-tool chain instrumented with the OpenTelemetry Python SDK 1.45.1 over real localhost transports (a thread-pool hand-off, agent-to-agent JSON-RPC over HTTP, Model Context Protocol (MCP) calls w...

Show moreHide
Technical Research by Oleh Ivchenko DOI: 10.5281/zenodo.23544775 80stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources3%○≥80% from editorially reviewed sources
[t]Trusted83%✓≥80% from verified, high-quality sources
[a]DOI68%○≥80% have a Digital Object Identifier
[b]CrossRef3%○≥80% indexed in CrossRef
[i]Indexed65%○≥80% have metadata indexed
[l]Academic70%○≥80% from journals/conferences/preprints
[f]Free Access83%✓≥80% are freely accessible
[r]References40 refs✓Minimum 10 references required
[w]Words [REQ]3,959✓Minimum 1,000 words of article body. Current: 3,959
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]35✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 35
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23544775
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness97%✓≥60% of references from 2025–2026. Current: 97%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (71 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)
AI ObservabilityRead More
Read more

Explainability-as-a-Service: Comparing Managed XAI Platforms Against In-House Observability Stacks

Posted on by
Technical Research
Technical Research by Oleh Ivchenko  ·  DOI: 10.5281/zenodo.23543861  79stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources6%○≥80% from editorially reviewed sources
[t]Trusted79%○≥80% from verified, high-quality sources
[a]DOI68%○≥80% have a Digital Object Identifier
[b]CrossRef6%○≥80% indexed in CrossRef
[i]Indexed64%○≥80% have metadata indexed
[l]Academic70%○≥80% from journals/conferences/preprints
[f]Free Access81%✓≥80% are freely accessible
[r]References47 refs✓Minimum 10 references required
[w]Words [REQ]3,917✓Minimum 1,000 words of article body. Current: 3,917
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]42✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 42
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23543861
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness74%✓≥60% of references from 2025–2026. Current: 74%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (70 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)

Teams that must explain and trace production models can rent the capability or build it from open-source parts; we ask what each choice costs in latency, fidelity and money. We benchmarked the estimators that managed tabular services expose (Kernel SHAP, as in SageMaker Clarify, and sampled Shapley with a path count, as in Vertex Explainable AI) against exact TreeSHAP and gradient methods on XG...

Show moreHide
Technical Research by Oleh Ivchenko DOI: 10.5281/zenodo.23543861 79stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources6%○≥80% from editorially reviewed sources
[t]Trusted79%○≥80% from verified, high-quality sources
[a]DOI68%○≥80% have a Digital Object Identifier
[b]CrossRef6%○≥80% indexed in CrossRef
[i]Indexed64%○≥80% have metadata indexed
[l]Academic70%○≥80% from journals/conferences/preprints
[f]Free Access81%✓≥80% are freely accessible
[r]References47 refs✓Minimum 10 references required
[w]Words [REQ]3,917✓Minimum 1,000 words of article body. Current: 3,917
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]42✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 42
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23543861
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness74%✓≥60% of references from 2025–2026. Current: 74%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (70 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)
AI ObservabilityRead More
Read more

Prompt Caching ROI: When Repeated-Context Discounts Actually Offset Engineering Overhead

Posted on by
Applied Research
Applied Research by Oleh Ivchenko  ·  DOI: 10.5281/zenodo.23543079  70stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources3%○≥80% from editorially reviewed sources
[t]Trusted66%○≥80% from verified, high-quality sources
[a]DOI47%○≥80% have a Digital Object Identifier
[b]CrossRef3%○≥80% indexed in CrossRef
[i]Indexed47%○≥80% have metadata indexed
[l]Academic50%○≥80% from journals/conferences/preprints
[f]Free Access78%○≥80% are freely accessible
[r]References32 refs✓Minimum 10 references required
[w]Words [REQ]4,185✓Minimum 1,000 words of article body. Current: 4,185
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]27✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 27
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23543079
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness93%✓≥60% of references from 2025–2026. Current: 93%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (54 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)

Providers price a cached input token at 2% to 50% of a fresh one, and several charge 1.25 to 2.5 times the base price to write the cache. We ask when that trade pays off once real request timing and engineering effort are counted. We collected prices, time-to-live (TTL) options and minimum cacheable lengths for 18 model and cache-mode pairs from Anthropic, OpenAI, Google and DeepSeek on 21 Octo...

Show moreHide
Applied Research by Oleh Ivchenko DOI: 10.5281/zenodo.23543079 70stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources3%○≥80% from editorially reviewed sources
[t]Trusted66%○≥80% from verified, high-quality sources
[a]DOI47%○≥80% have a Digital Object Identifier
[b]CrossRef3%○≥80% indexed in CrossRef
[i]Indexed47%○≥80% have metadata indexed
[l]Academic50%○≥80% from journals/conferences/preprints
[f]Free Access78%○≥80% are freely accessible
[r]References32 refs✓Minimum 10 references required
[w]Words [REQ]4,185✓Minimum 1,000 words of article body. Current: 4,185
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]27✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 27
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23543079
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness93%✓≥60% of references from 2025–2026. Current: 93%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (54 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)
Cost-Effective Ent…Read More
Read more

Model Routing at Scale: Cost-Quality Arbitrage Across Frontier and Open-Weight Model Tiers

Posted on by
Applied Research
Applied Research by Oleh Ivchenko  ·  DOI: 10.5281/zenodo.23542219  76stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources3%○≥80% from editorially reviewed sources
[t]Trusted75%○≥80% from verified, high-quality sources
[a]DOI63%○≥80% have a Digital Object Identifier
[b]CrossRef3%○≥80% indexed in CrossRef
[i]Indexed59%○≥80% have metadata indexed
[l]Academic66%○≥80% from journals/conferences/preprints
[f]Free Access81%✓≥80% are freely accessible
[r]References32 refs✓Minimum 10 references required
[w]Words [REQ]3,985✓Minimum 1,000 words of article body. Current: 3,985
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]27✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 27
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23542219
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness71%✓≥60% of references from 2025–2026. Current: 71%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (65 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)

Model routers send each query to a cheaper model when they predict it will do as well as the frontier model, and planning estimates put the resulting saving at well over half of the frontier bill. We ask how much of that saving survives once a router is trained, calibrated and evaluated on prompts it has not seen. We used the public SPROUT routing dataset, which records judged scores and token ...

Show moreHide
Applied Research by Oleh Ivchenko DOI: 10.5281/zenodo.23542219 76stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources3%○≥80% from editorially reviewed sources
[t]Trusted75%○≥80% from verified, high-quality sources
[a]DOI63%○≥80% have a Digital Object Identifier
[b]CrossRef3%○≥80% indexed in CrossRef
[i]Indexed59%○≥80% have metadata indexed
[l]Academic66%○≥80% from journals/conferences/preprints
[f]Free Access81%✓≥80% are freely accessible
[r]References32 refs✓Minimum 10 references required
[w]Words [REQ]3,985✓Minimum 1,000 words of article body. Current: 3,985
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]27✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 27
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23542219
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness71%✓≥60% of references from 2025–2026. Current: 71%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (65 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)
Cost-Effective Ent…Read More
Read more

KV-Cache Compression Techniques in Production: Cost Savings vs Quality Degradation Trade-offs

Posted on by
Applied Research
Applied Research by Oleh Ivchenko  ·  DOI: 10.5281/zenodo.23541486  79stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources6%○≥80% from editorially reviewed sources
[t]Trusted79%○≥80% from verified, high-quality sources
[a]DOI67%○≥80% have a Digital Object Identifier
[b]CrossRef6%○≥80% indexed in CrossRef
[i]Indexed64%○≥80% have metadata indexed
[l]Academic70%○≥80% from journals/conferences/preprints
[f]Free Access82%✓≥80% are freely accessible
[r]References33 refs✓Minimum 10 references required
[w]Words [REQ]4,080✓Minimum 1,000 words of article body. Current: 4,080
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]28✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 28
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23541486
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness62%✓≥60% of references from 2025–2026. Current: 62%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (69 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)

Compressing the key-value (KV) cache is the most direct lever for cutting the cost of long-context inference, but published results usually report memory savings or benchmark scores in isolation. We ask what compression buys in a production serving unit and what it costs in quality. We built a deterministic model from public numbers: attention configurations of five open-weight models, HBM capa...

Show moreHide
Applied Research by Oleh Ivchenko DOI: 10.5281/zenodo.23541486 79stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources6%○≥80% from editorially reviewed sources
[t]Trusted79%○≥80% from verified, high-quality sources
[a]DOI67%○≥80% have a Digital Object Identifier
[b]CrossRef6%○≥80% indexed in CrossRef
[i]Indexed64%○≥80% have metadata indexed
[l]Academic70%○≥80% from journals/conferences/preprints
[f]Free Access82%✓≥80% are freely accessible
[r]References33 refs✓Minimum 10 references required
[w]Words [REQ]4,080✓Minimum 1,000 words of article body. Current: 4,080
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]28✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 28
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23541486
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness62%✓≥60% of references from 2025–2026. Current: 62%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (69 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)
Cost-Effective Ent…Read More
Read more

The Context-Window Premium: Why Long-Context Models Command Higher Prices Than Token Count Alone Explains

Posted on by
AI Economics
AI Economics by Oleh Ivchenko  ·  DOI: 10.5281/zenodo.23487250  79stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources16%○≥80% from editorially reviewed sources
[t]Trusted78%○≥80% from verified, high-quality sources
[a]DOI68%○≥80% have a Digital Object Identifier
[b]CrossRef16%○≥80% indexed in CrossRef
[i]Indexed62%○≥80% have metadata indexed
[l]Academic70%○≥80% from journals/conferences/preprints
[f]Free Access78%○≥80% are freely accessible
[r]References37 refs✓Minimum 10 references required
[w]Words [REQ]4,015✓Minimum 1,000 words of article body. Current: 4,015
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]32✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 32
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23487250
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness79%✓≥60% of references from 2025–2026. Current: 79%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (70 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)

Providers now advertise context windows of one million tokens as a headline feature, but their price lists treat that capability in very different ways. We ask whether long context is priced as a distinct feature or bundled into ordinary per-token rates. We built a dataset of 83 offerings from 14 model developers, sold first-party or through three hosts, read directly from public pricing pages ...

Show moreHide
AI Economics by Oleh Ivchenko DOI: 10.5281/zenodo.23487250 79stabilfr·wuedophcgmx
BadgeMetricValueStatusDescription
[s]Reviewed Sources16%○≥80% from editorially reviewed sources
[t]Trusted78%○≥80% from verified, high-quality sources
[a]DOI68%○≥80% have a Digital Object Identifier
[b]CrossRef16%○≥80% indexed in CrossRef
[i]Indexed62%○≥80% have metadata indexed
[l]Academic70%○≥80% from journals/conferences/preprints
[f]Free Access78%○≥80% are freely accessible
[r]References37 refs✓Minimum 10 references required
[w]Words [REQ]4,015✓Minimum 1,000 words of article body. Current: 4,015
[u]Original [REQ]100%✓≥85% unique sentences (no padding or repeated blocks) and no leaked drafting notes. Current: 100% unique
[e]Evidence [REQ]32✓Cites at least 1 external source(s): links, DOIs or a reference list. Current: 32
[d]DOI [REQ]✓✓Zenodo DOI registered for persistent citation. DOI: 10.5281/zenodo.23487250
[o]ORCID✓✓Author ORCID verified for academic identity
[p]Peer Reviewed—○Peer reviewed by an assigned reviewer
[h]Freshness79%✓≥60% of references from 2025–2026. Current: 79%
[c]Data Charts4✓Original data charts from reproducible analysis (min 2). Current: 4
[g]Code✓✓Source code available on GitHub
[m]Diagrams3✓Mermaid architecture/flow diagrams. Current: 3
[x]Cited by0○Referenced by 0 other hub article(s)
Score = Ref Trust (70 × 60%) + Required (4/4 × 30%) + Optional (5/7 × 10%)
AI EconomicsRead More
Read more