https://read0nly.net/2026/10/01/openai-says-it-disrupted-a-coordinated-campaign-to-extract-protected-model-reasoning/https://read0nly.net/wp-content/uploads/2026/10/openai-distillation-infographic-v2.pngOpenAI protected reasoning campaign overviewThe graphic separates OpenAI’s reported campaign activity from a possible cross-model attack path described by the cited arXiv paper and from the limits of the public evidence. It is an editorial synthesis, not an independent reconstruction of the campaign.https://read0nly.net/wp-content/uploads/2026/10/openai-distillation-infographic-v1.pngAI model distillation campaign overviewThe graphic separates OpenAI’s reported campaign activity from the technical mechanism described by independent researchers and the limits of the public evidence. It is an editorial synthesis, not an independent reconstruction of the campaign.2026-10-01T16:40:20+00:00monthlyhttps://read0nly.net/2026/09/30/breadth-first-search-in-python-the-queue-choice-changes-the-run-time/https://read0nly.net/wp-content/uploads/2026/09/read0nly-breadth-first-search-queue-20260930.pngBreadth-first search queue costs on PythonThe left panel shows the BFS queue rule. The table reports median traversal time from seven samples for the two implementations on four seeded graph sizes; both returned identical traversal and distance arrays.2026-09-30T08:50:49+00:00monthlyhttps://read0nly.net/2026/09/29/nvidia-announces-open-agent-safety-platform-with-openshell-and-sentry/https://read0nly.net/wp-content/uploads/2026/09/agent-safety-infographic-r4.pngNVIDIA announces Open Agent Safety Platform with OpenShell and SentryAs checked on 29 September, GitHub listed v0.1.2 as “Latest”; its release record dates publication to 28 September. NVIDIA’s walkthrough describes v0.1.0. The curl and adversarial results are vendor-reported OpenShell evidence from NVIDIA’s v0.1.0 walkthrough, not independent validation or a Sentry test. The millisecond figure remains NVIDIA’s claim.https://read0nly.net/wp-content/uploads/2026/09/agent-safety-infographic-r3.pngNVIDIA announces Open Agent Safety Platform with OpenShell and SentryGitHub marked v0.1.2 as the latest release on 29 September; NVIDIA’s walkthrough describes v0.1.0. The curl and adversarial results are vendor-reported OpenShell evidence from NVIDIA’s v0.1.0 walkthrough, not independent validation or a Sentry test. The millisecond figure remains NVIDIA’s claim.https://read0nly.net/wp-content/uploads/2026/09/agent-safety-infographic-r2.pngNVIDIA announces Open Agent Safety Platform with OpenShell and SentryGitHub marked v0.1.2 as the latest release on 29 September; NVIDIA’s walkthrough describes v0.1.0. The curl and adversarial results are vendor-reported OpenShell evidence, not independent validation or a Sentry test. The millisecond figure remains NVIDIA’s claim.https://read0nly.net/wp-content/uploads/2026/09/agent-safety-infographic.pngNVIDIA announces Open Agent Safety Platform with OpenShell and SentryOpenShell predates the 28 September platform announcement; Sentry is presented as a BlueField-4 reference design. NVIDIA’s “milliseconds” figure is a claim, not an independently reported test result.2026-09-29T18:14:57+00:00monthlyhttps://read0nly.net/2026/09/28/kahan-summation-in-python-one-correction-lost-to-cancellation/https://read0nly.net/wp-content/uploads/2026/09/read0nly-compensated-summation-infographic-20260928.pngKahan correction loss and Python summation resultsThe left panel follows one tested Kahan pass. The table compares both term orders after 10,000 repeats; the bars show order A medians for 30,000 terms across seven samples.2026-09-28T08:07:50+00:00monthlyhttps://read0nly.net/2026/09/27/check-point-confirms-active-exploitation-of-two-vpn-and-management-flaws/https://read0nly.net/wp-content/uploads/2026/09/checkpoint-vpn-infographic-v3.pngCVE-2026-85102 and CVE-2026-93616 timeline v3Reported disclosure and exploitation timeline. Spacing is not proportional to elapsed time. NCSC’s 10 September alert warned abuse was likely soon; its 23 September update crediting observed abuse to Check Point’s own report is not an independent NCSC finding.https://read0nly.net/wp-content/uploads/2026/09/checkpoint-vpn-infographic-v2.pngCVE-2026-85102 and CVE-2026-93616 timeline v2Reported disclosure and exploitation timeline. Spacing is not proportional to elapsed time; the 10 versus 12 September start-date conflict between NCSC and Check Point remains unreconciled in the public record.https://read0nly.net/wp-content/uploads/2026/09/checkpoint-vpn-infographic.pngCVE-2026-85102 and CVE-2026-93616 timelineReported disclosure and exploitation timeline. Spacing is not proportional to elapsed time; the 10 versus 12 September start-date conflict between NCSC and Check Point remains unreconciled in the public record.2026-09-27T14:32:44+00:00monthlyhttps://read0nly.net/2026/09/25/tries-in-python-one-walk-replaces-n-comparisons/https://read0nly.net/wp-content/uploads/2026/09/trie-infographic-v3.pngTrie vs linear-scan prefix-count infographicMeasured on this Raspberry Pi with CPython 3.13.5: the log-scale bars show the linear-scan/trie query-time ratio, while the trie used roughly 11x the peak traced memory of a fresh list containing the same words.https://read0nly.net/wp-content/uploads/2026/09/trie-infographic-v2.pngTrie vs linear-scan prefix-count infographic (revised)Original Read0nly technical infographic, corrected after adversarial review.https://read0nly.net/wp-content/uploads/2026/09/trie-infographic.pngTrie vs linear-scan prefix-count infographicOriginal Read0nly technical infographic.2026-09-25T12:22:58+00:00monthlyhttps://read0nly.net/2026/09/23/kadanes-algorithm-in-python-one-running-sum-replaces-every-subarray-check/https://read0nly.net/wp-content/uploads/2026/09/kadane-infographic-v4-final.pngKadane's algorithm running-sum and speedup infographic (v4, heading wrap fix)Original Read0nly technical infographic, corrected heading line-wrap.https://read0nly.net/wp-content/uploads/2026/09/kadane-infographic-v3-final.pngKadane's algorithm running-sum and speedup infographic (v3, text clearance fix)Original Read0nly technical infographic, corrected text/border clearance.https://read0nly.net/wp-content/uploads/2026/09/kadane-infographic-v2.pngKadane's algorithm running-sum and speedup infographic (v2, layout fix)Original Read0nly technical infographic, corrected layout.https://read0nly.net/wp-content/uploads/2026/09/kadane-infographic.pngKadane's algorithm running-sum and speedup infographicOriginal Read0nly technical infographic.2026-09-23T10:56:42+00:00monthlyhttps://read0nly.net/2026/09/22/researchers-describe-two-openai-codex-sandbox-escapes-already-fixed/https://read0nly.net/wp-content/uploads/2026/09/codex-sandbox-escapes.pngOpenAI Codex sandbox escape comparisonBoth escapes were reported to OpenAI on the same day and fixed within the same eight-day window. The comparison is an editorial synthesis of the cited Accomplish AI and BleepingComputer reporting, not an official OpenAI diagram.2026-09-22T10:12:17+00:00monthlyhttps://read0nly.net/2026/09/21/difference-arrays-in-python-batch-range-updates-without-repeated-writes/https://read0nly.net/wp-content/uploads/2026/09/range-update-infographic.pngDifference arrays and batch range updatesOriginal Read0nly technical infographic.2026-09-21T11:25:19+00:00monthlyhttps://read0nly.net/2026/09/20/apple-reference-image-how-iphone-18-pro-signs-a-photo-as-real/https://read0nly.net/wp-content/uploads/2026/09/reference-image-ceilings-2.pngApple Reference Image vs. C2PA (corrected)Editorial comparison built from Apple's own Security Research announcement, the C2PA specification, and Leica's published Content Credentials camera list, not an independent technical audit of either system.https://read0nly.net/wp-content/uploads/2026/09/reference-image-ceilings-1.pngApple Reference Image vs. C2PAEditorial comparison built from Apple's own Security Research announcement and the C2PA specification's public steering-committee page, not an independent technical audit of either system.https://read0nly.net/wp-content/uploads/2026/09/reference-image-ceilings.pngReference-image ceilings by platformApple’s app-level ceilings are combined totals across people, photos and text, not separate per-category allowances like Google publishes for Gemini 3. Editorial comparison built from each vendor’s own developer and support documentation, not an independent benchmark.2026-09-21T05:33:54+00:00monthlyhttps://read0nly.net/2026/09/20/openai-confirms-safety-talks-with-google-and-anthropic-no-agreement-yet/https://read0nly.net/wp-content/uploads/2026/09/three-labs-safety-talks-1.pngThree rivals, one reported conversationOpenAI confirmed contact with Google and Anthropic on AI safety to CNBC. Dashed connectors mark reported, not independently corroborated, involvement. Editorial synthesis, not an official process diagram.https://read0nly.net/wp-content/uploads/2026/09/three-labs-safety-talks.pngThree rivals, one reported conversationOpenAI confirmed contact with Google and Anthropic on AI safety to CNBC. Dashed connectors mark reported, not independently corroborated, involvement. Editorial synthesis, not an official process diagram.2026-09-20T11:20:09+00:00monthlyhttps://read0nly.net/2026/09/19/cisa-adds-linux-kernel-tls-flaw-to-exploited-vulnerabilities-catalogue/https://read0nly.net/wp-content/uploads/2026/09/kernel-cve-infographic-v2.pngCVE-2025-39682 KEV timeline (corrected)Corrected timeline: scoping begins within roughly two hours of the KEV addition, before the BOD 26-04 forensic-triage due date three days later. Editorial timeline built from CISA’s own catalogue and directive guidance, not an official CISA graphic.https://read0nly.net/wp-content/uploads/2026/09/kernel-cve-infographic.pngCVE-2025-39682 KEV timelineThe timeline separates the KEV addition date from the BOD 26-04 forensic-triage due date. It is an editorial timeline built from CISA’s own catalogue and directive guidance, not an official CISA graphic.2026-09-19T15:05:32+00:00monthlyhttps://read0nly.net/2026/09/19/repomix-packs-a-codebase-into-one-file-an-llm-can-read/https://read0nly.net/wp-content/uploads/2026/09/repomix-infographic.pngRepomix packing pipeline and measured token reductionRepomix pipeline, measured token counts and secret-scan test results.2026-09-19T13:00:39+00:00monthlyhttps://read0nly.net/2026/09/19/mits-xvr-technique-adapts-x-ray-to-3d-registration-to-each-patient-in-minutes/https://read0nly.net/wp-content/uploads/2026/09/mit-xvr-registration-infographic-v2.pngMIT xvr registration infographic (v2)Patient adaptation time and accuracy for xvr versus manual registration and prior AI methods.https://read0nly.net/wp-content/uploads/2026/09/mit-xvr-registration-infographic.pngMIT xvr registration infographicPatient adaptation time and accuracy for xvr versus manual registration and prior AI methods.2026-09-19T10:31:47+00:00monthlyhttps://read0nly.net/2026/09/18/paper2agent-turns-research-papers-into-mcp-backed-ai-agents/https://read0nly.net/wp-content/uploads/2026/09/paper2agent-workflow-20260918.pngPaper2Agent workflow infographicPaper2Agent's reported conversion path, with the case-study result separated from the evidence boundary. Diagram is an editorial synthesis of the Nature paper.2026-09-18T07:13:39+00:00monthlyhttps://read0nly.net/2026/08/13/c2pa-explained-what-content-credentials-can-prove-and-what-they-cannot/https://read0nly.net/wp-content/uploads/2026/08/c2pa-validation-steps-20260813.pngThe separate stages of C2PA validationOne signature, several questions. A green tick should never be asked to do all the thinking.https://read0nly.net/wp-content/uploads/2026/08/c2pa-signed-provenance-journey-20260813.pngA signed provenance journeyC2PA follows the paperwork. It does not interview the witnesses.2026-09-17T20:17:38+00:00monthlyhttps://read0nly.net/2026/09/17/openai-publishes-a-framework-for-reporting-model-misalignment/https://read0nly.net/wp-content/uploads/2026/09/misalignment-framework-infographic-v2.pngOpenAI misalignment disclosure framework, six initial reports (corrected)The six reports OpenAI published on 16 September 2026, each tagged with its disclosure track. A third, slower track covers complex third-party cases and is not one of the six.https://read0nly.net/wp-content/uploads/2026/09/misalignment-framework-infographic.pngOpenAI misalignment disclosure framework, six initial reportsThe six reports OpenAI published on 16 September 2026, each tagged with its disclosure track. A third, slower track covers complex third-party cases and is not one of the six.2026-09-17T08:04:27+00:00monthlyhttps://read0nly.net/2026/09/17/typesafe-ai-launches-jev-its-first-system-one-model-with-self-reported-speed-claims/https://read0nly.net/wp-content/uploads/2026/09/typesafe-jev-infographic-v2.pngTypeSafe AI Jev launch infographic (revision 2)The panel separates TypeSafe's three named post-training methods from two distinct self-reported measurements, a single demo example and an aggregate four-workflow evaluation, labelled separately because their figures do not match.https://read0nly.net/wp-content/uploads/2026/09/typesafe-jev-infographic.pngTypeSafe AI Jev launch infographicThe panel separates TypeSafe's three named post-training methods from its own published workflow comparison. Figures are TypeSafe's self-reported numbers; no independent reproduction was found.2026-09-16T23:47:25+00:00monthlyhttps://read0nly.net/2026/09/16/floyds-cycle-detection-in-python-trade-a-set-for-two-pointers/https://read0nly.net/wp-content/uploads/2026/09/floyd-cycle-infographic-v7.pngFloyd cycle detection: set versus two pointers, corrected v7The panels use the same six-node list with a four-node cycle, 2 -> 3 -> 4 -> 5 -> 2. The set stores visited objects; Floyd stores only two pointers. The time labels are rounded to the nearest microsecond and describe the preserved Raspberry Pi run, not a general speedup.https://read0nly.net/wp-content/uploads/2026/09/floyd-cycle-infographic-v6.pngFloyd cycle detection: set versus two pointers, corrected v6The panels use the same six-node list with a four-node cycle, 2 -> 3 -> 4 -> 5 -> 2. The set stores visited objects; Floyd stores only two pointers. The time labels are rounded to the nearest microsecond and describe the preserved Raspberry Pi run, not a general speedup.https://read0nly.net/wp-content/uploads/2026/09/floyd-cycle-infographic-v5.pngFloyd cycle detection: set versus two pointers, corrected v5The panels use the same six-node list with a four-node cycle, 2 -> 3 -> 4 -> 5 -> 2. The set stores visited objects; Floyd stores only two pointers. The time labels are rounded to the nearest microsecond and describe the preserved Raspberry Pi run, not a general speedup.https://read0nly.net/wp-content/uploads/2026/09/floyd-cycle-infographic-v4.pngFloyd cycle detection: set versus two pointers, correctedThe panels use the same six-node list with a four-node cycle, 2 -> 3 -> 4 -> 5 -> 2. The set stores visited objects; Floyd stores only two pointers. The time labels are rounded to the nearest microsecond and describe the preserved Raspberry Pi run, not a general speedup.https://read0nly.net/wp-content/uploads/2026/09/floyd-cycle-infographic-v3.pngFloyd cycle detection: set versus two pointers, correctedThe panels use the same six-node cycle. The set stores visited objects; Floyd stores only two pointers. The time labels are rounded to the nearest microsecond and describe the preserved Raspberry Pi run, not a general speedup.2026-09-16T10:26:55+00:00monthlyhttps://read0nly.net/2026/09/15/revolut-data-disclosure-confirmed-a-government-account-compromise-claim-remains-unverified/2026-09-15T14:42:29+00:00monthlyhttps://read0nly.net/2026/09/15/cisa-lists-critical-gitlab-path-traversal-flaw-after-evidence-of-active-exploitation/https://read0nly.net/wp-content/uploads/2026/09/gitlab-cve-infographic-v2-1.pngCVE-2026-85706 response v2The graphic separates the vulnerability, CISA exploitation signal and fixed releases from the defensive response. It does not quantify victims or establish that a particular GitLab installation was compromised.https://read0nly.net/wp-content/uploads/2026/09/gitlab-cve-infographic-v2.pngCVE-2026-85706 response v2The graphic separates the vulnerability, CISA exploitation signal and fixed releases from the defensive response. It does not quantify victims or establish that a particular GitLab installation was compromised.https://read0nly.net/wp-content/uploads/2026/09/gitlab-cve-infographic.pngCVE-2026-85706 responseThe graphic separates the vulnerability, CISA exploitation signal and fixed releases from the defensive response. It does not quantify victims or establish that a particular GitLab installation was compromised.2026-09-15T10:08:27+00:00monthlyhttps://read0nly.net/2026/09/14/the-strategy-pattern-in-python-separate-a-decision-from-the-work/2026-09-14T08:51:15+00:00monthlyhttps://read0nly.net/2026/09/13/openai-opens-agents-api-beta-bringing-codexs-harness-to-cloud-agents/2026-09-13T11:23:25+00:00monthlyhttps://read0nly.net/2026/09/13/anthropic-says-claude-misuse-now-spans-seven-threat-categories-with-cyber-operations-at-the-centre/2026-09-13T10:21:29+00:00monthlyhttps://read0nly.net/2026/09/13/microsoft-documents-a-million-email-executive-impersonation-campaign/https://read0nly.net/wp-content/uploads/2026/09/microsoft-invoice-infographic-v2.pngThe payment lure infographic, revision 2Microsoft’s observed campaign combined executive impersonation, vendor branding, a fabricated invoice and a staged email thread. This editorial synthesis separates reported lure components from the independent payment verification control; it does not show successful payments or prove the extent of AI generation.https://read0nly.net/wp-content/uploads/2026/09/microsoft-invoice-infographic-v1.pngThe payment lure infographicMicrosoft’s observed campaign combined executive impersonation, vendor branding, a fabricated invoice and a staged email thread. This editorial synthesis separates reported lure components from the independent payment verification control; it does not show successful payments or prove the extent of AI generation.2026-09-13T08:41:58+00:00monthlyhttps://read0nly.net/2026/09/12/openais-hugging-face-incident-exposed-the-limits-of-isolated-ai-evaluations/https://read0nly.net/wp-content/uploads/2026/09/openai-hugging-face-infographic-r2.pngOpenAI and Hugging Face incident infographicThe timeline separates the evaluation environment, the route into a third-party production system and the independent review. The evidence boundary does not deny the Hugging Face intrusion; it limits the claim about OpenAI deployed products.https://read0nly.net/wp-content/uploads/2026/09/openai-hugging-face-infographic.pngCeuta records timelineThe infographic separates the evaluation setup, the reported boundary crossing and the independent review. Approximate figures are marked, and the final panel preserves the distinction between an evaluation environment and a production deployment.2026-09-12T11:14:18+00:00monthlyhttps://read0nly.net/2026/09/11/two-sum-in-python-trade-a-little-memory-for-one-pass/2026-09-11T09:51:29+00:00monthlyhttps://read0nly.net/2026/09/10/spains-ceuta-records-show-a-warning-before-the-july-crossings/https://read0nly.net/wp-content/uploads/2026/09/ceuta-infographic-v3.pngCeuta records timeline v3The timeline separates the documented 29 July messages from the reported outcome and the later release of records. The final sentence is an editorial evidence boundary, not a finding about whether the crisis could have been prevented.https://read0nly.net/wp-content/uploads/2026/09/ceuta-infographic-v2.pngCeuta records timeline v2The timeline separates the documented 29 July messages from the reported outcome and the later release of records. The final sentence is an editorial evidence boundary, not a finding about whether the crisis could have been prevented.https://read0nly.net/wp-content/uploads/2026/09/ceuta-infographic.pngCeuta records timelineThe timeline separates the documented 29 July messages from the reported outcome and the later release of records. The final sentence is an editorial evidence boundary, not a finding about whether the crisis could have been prevented.2026-09-10T11:30:17+00:00monthlyhttps://read0nly.net/2026/09/10/us-agencies-accuse-six-chinese-ai-firms-of-industrial-scale-model-distillation/https://read0nly.net/wp-content/uploads/2026/09/ai-distillation-infographic-v1.pngAI model distillation campaign overviewThe advisory separates reported tactics from recommended provider responses. This editorial graphic is a synthesis of the cited sources, not an independent reconstruction of any company’s operations.2026-09-10T10:27:39+00:00monthlyhttps://read0nly.net/2026/09/09/top-k-selection-in-python-stop-sorting-values-you-will-discard/https://read0nly.net/wp-content/uploads/2026/09/top_k_diagram-v2.pngTop-k selection with a bounded heapThe heap keeps the cut small, although every input value still gets a look.2026-09-09T08:02:49+00:00monthlyhttps://read0nly.net/2026/09/08/claude-formalises-fermats-last-theorem-but-it-did-not-discover-the-proof/https://read0nly.net/wp-content/uploads/2026/09/fermat-formalisation-r3.pngFermat proof formalisation and machine checking revision 3A simplified evidence map of the reported workflow. The machine checks concern the Lean artefact, not the autonomy of the AI system.https://read0nly.net/wp-content/uploads/2026/09/fermat-formalisation-r2.pngFermat proof formalisation and machine checking revision 2A simplified evidence map of the reported workflow. The machine checks concern the Lean artefact, not the autonomy of the AI system.https://read0nly.net/wp-content/uploads/2026/09/fermat-formalisation.pngFermat proof formalisation and machine checkingA simplified evidence map of the reported workflow. The machine checks concern the Lean artefact, not the autonomy of the AI system.2026-09-08T10:46:41+00:00monthlyhttps://read0nly.net/2026/09/07/euclids-algorithm-in-python-use-remainders-instead-of-repeated-subtraction/https://read0nly.net/wp-content/uploads/2026/09/euclidean-gcd-work-v2.pngEuclid's algorithm: loop-count comparison2026-09-07T09:31:31+00:00monthlyhttps://read0nly.net/2026/09/06/openspec-v1-12-0-what-the-scaffold-proves-and-what-it-cannot/https://read0nly.net/wp-content/uploads/2026/09/openspec-repository-layout-1200x675-1.pngOpenSpec repository layoutOpenSpec's two repository locations and how archiving connects themhttps://read0nly.net/wp-content/uploads/2026/09/openspec-review-boundary-1200x675-1.pngOpenSpec planning and review boundaryOpenSpec planning records and review boundary2026-09-06T18:34:57+00:00monthlyhttps://read0nly.net/2026/09/06/github-spec-kit-v1-0-4-what-the-scaffold-proves-and-what-it-cannot/https://read0nly.net/wp-content/uploads/2026/09/spec-kit-documentation-boundary-v6-1-0-4.pngSpec Kit documentation-only scaffold boundary, connector-correctedDocumentation-based Spec Kit artefact sequence with clean connector terminations and an explicitly untested agent stage.https://read0nly.net/wp-content/uploads/2026/09/spec-kit-documentation-boundary-v3-1-0-4.pngSpec Kit documentation-only scaffold boundary, correctedDocumentation-based Spec Kit artefact sequence with the untested agent stage labelled clearly.https://read0nly.net/wp-content/uploads/2026/09/spec-kit-documentation-boundary-v1-0-4.pngSpec Kit documentation-only scaffold boundaryA documentation-based sequence of Spec Kit artefacts with an explicitly untested Codex step.https://read0nly.net/wp-content/uploads/2026/09/spec-kit-control-loop-v1-0-4.pngSpec Kit controlled hand-offA diagram of intent, design, agent work and the return path for evidence or changed intent.2026-09-06T12:23:27+00:00monthlyhttps://read0nly.net/2026/09/06/spec-driven-development-with-ai-turning-intent-into-reviewable-work/2026-09-06T01:20:18+00:00monthlyhttps://read0nly.net/2026/09/05/amd-instinct-mi400-what-is-specified-and-what-remains-unmeasured/https://read0nly.net/wp-content/uploads/2026/09/amd-mi400-decision-map-20260905-r3.pngAMD Instinct MI400 documented product rolesEditorial synthesis of AMD’s stated product roles, specifications and availability. MI430X is not shown as part of Helios. It is not an application benchmark. Sources: AMD MI400, MI455X, MI430X, Helios and ROCm documentation.https://read0nly.net/wp-content/uploads/2026/09/amd-mi400-decision-map-20260905-r2.pngAMD Instinct MI400 documented product rolesEditorial synthesis of AMD’s stated product roles, specifications and availability. MI430X is not shown as part of Helios. It is not an application benchmark. Sources: AMD MI400, MI455X, MI430X, Helios and ROCm documentation.https://read0nly.net/wp-content/uploads/2026/09/amd-mi400-decision-map-20260905-r1.pngAMD Instinct MI400 documented product rolesEditorial synthesis of AMD’s stated product roles, specifications and availability. MI430X is not shown as part of Helios. It is not an application benchmark. Sources: AMD MI400, MI455X, MI430X, Helios and ROCm documentation.https://read0nly.net/wp-content/uploads/2026/09/amd-mi400-decision-map-20260905.pngamd-mi400-decision-map-20260905.png2026-09-05T10:10:05+00:00monthlyhttps://read0nly.net/2026/09/04/binary-exponentiation-in-python-use-the-exponents-bits-instead-of-counting-to-it/https://read0nly.net/wp-content/uploads/2026/09/binary-exponentiation-workflow-v3.pngBinary exponentiation workflowA visual comparison of repeated multiplication and square-and-multiply.2026-09-04T17:51:57+00:00monthlyhttps://read0nly.net/2026/09/04/gpt-6-astra-a-new-model-a-higher-price-and-a-safety-deployment/https://read0nly.net/wp-content/uploads/2026/09/gpt-6-astra-evidence-boundaries-v3.pngGPT-6 Astra: evidence classes and external testsPublished API terms, the OpenAI-reported Preparedness Framework classification, and external but test setup-specific results are separate evidence classes. Local workload testing remains necessary.https://read0nly.net/wp-content/uploads/2026/09/gpt-6-astra-decision-aid.pngGPT-6 Astra public record and customer checksA decision aid built from OpenAI’s published model and safety records. The right-hand panel is Read0nly’s evidence boundary, not a claim that the missing measurements do not exist.2026-09-04T08:05:20+00:00monthlyhttps://read0nly.net/2026/09/03/what-the-g20-carolina-principles-and-eu-ai-act-records-actually-establish/https://read0nly.net/wp-content/uploads/2026/09/g20-documents-ai-act-scope-20260903-v3.pngG20 documents and EU AI Act scope, Read0nly illustrationRead0nly illustration: the G20 texts state policy aims under members’ own legal frameworks. The EU material identifies a regulation, scope conditions, authorities and a timetable. The 2024 Article 2 text includes certain third-country providers or deployers where output is used in the Union. Check the applicable version for a specific case.https://read0nly.net/wp-content/uploads/2026/09/g20-documents-ai-act-scope-20260903-v2.pngG20 documents and EU AI Act scope, Read0nly illustrationRead0nly illustration: the G20 texts state policy aims under members’ own legal frameworks, while the EU material identifies a regulation, scope conditions, authorities and a timetable. It is a document comparison, not a legal conclusion.https://read0nly.net/wp-content/uploads/2026/09/g20-consensus-ai-act-enforcement-20260903.pngG20 consensus and EU AI Act enforcement, Read0nly illustrationRead0nly illustration: the accessible G20 record is a consensus statement, while the Commission’s EU material describes enforcement roles and powers. It is an editorial comparison, not a causal sequence.2026-09-03T11:19:06+00:00monthlyhttps://read0nly.net/2026/09/02/sliding-window-maximums-in-python-a-monotonic-deque-avoids-repeated-scans/https://read0nly.net/wp-content/uploads/2026/09/sliding-window-monotonic-deque-v2.pngSliding-window maxima with a monotonic dequeA diagram of repeated scanning and candidate retention for fixed-width maximums.2026-09-02T07:41:35+00:00monthlyhttps://read0nly.net/2026/08/31/prefix-sums-in-python-pay-once-answer-static-range-totals-quickly/https://read0nly.net/wp-content/uploads/2026/08/prefix-sums-range-queries-v2.pngPrefix sums for static range queriesThe lower route stores a little extra bookkeeping so later totals do not keep rereading the same values.2026-08-31T06:42:13+00:00monthlyhttps://read0nly.net/archive/2026-08-30T21:36:56+00:00weekly0.6https://read0nly.net/2026/08/30/gpt-5-6-sol-terra-and-luna-what-changes-between-openais-three-tiers/https://read0nly.net/wp-content/uploads/2026/08/gpt56-sol-terra-luna-tiers-20260830-r2.pnggpt56-sol-terra-luna-tiers-20260830-r2.pnghttps://read0nly.net/wp-content/uploads/2026/08/gpt56-sol-terra-luna-tiers-20260830.pnggpt56-sol-terra-luna-tiers-20260830.png2026-08-30T21:06:01+00:00monthlyhttps://read0nly.net/2026/08/30/omarchy-what-arch-linux-looks-like-with-a-desktop-policy/https://read0nly.net/wp-content/uploads/2026/08/omarchy-policy-stack.pngomarchy-policy-stack.png2026-08-30T21:04:53+00:00monthlyhttps://read0nly.net/2026/08/12/binary-search-in-python-stop-checking-every-item/https://read0nly.net/wp-content/uploads/2026/08/binary-search-halving-corrected.pngBinary search narrowing a sorted listHalf the candidates leave after every comparison. Ruthless, but efficient.2026-08-30T12:54:45+00:00monthlyhttps://read0nly.net/2026/08/12/b-trees-the-filing-system-behind-fast-databases/https://read0nly.net/wp-content/uploads/2026/08/btree-search-42-balanced-20260813.pngSearch for 42 in a balanced B-treeThe search visits one node per level and leaves separator equality in the current node.https://read0nly.net/wp-content/uploads/2026/08/btree-balanced-model-20260813.pngBalanced B-tree teaching modelEvery leaf is at depth 2; equality with a separator is resolved in the node containing it.https://read0nly.net/wp-content/uploads/2026/08/btree-split-visual-revision.pngSplitting a full B-tree page, revisedWhen a node is full, it divides the paperwork and informs management.https://read0nly.net/wp-content/uploads/2026/08/btree-search-42-visual-revision.pngSearching a B-tree for 42, revisedSearching for 42: choose the interval, ignore everything outside it, then repeat.https://read0nly.net/wp-content/uploads/2026/08/btree-routes-visual-revision.pngB-tree interval routes, revisedA tree that favours wide shelves over endless branches.https://read0nly.net/wp-content/uploads/2026/08/btree-search-42-colour-key-en.pngSearching a B-tree for 42: colour keySearching for 42: yellow marks the values and comparisons, while coral marks the route.https://read0nly.net/wp-content/uploads/2026/08/btree-split-routes-en.pngSplitting a full B-tree pageThe separator moves upstairs, and its pointers leave in different directions.https://read0nly.net/wp-content/uploads/2026/08/btree-search-42-en.pngSearching a B-tree for 42Searching for 42: choose the interval, ignore everything outside it, then repeat.https://read0nly.net/wp-content/uploads/2026/08/btree-routes-en.pngB-tree routes created by separator keysEach arrow leaves from the interval selected by the comparison.https://read0nly.net/wp-content/uploads/2026/08/btree-split-en.pngB-tree: splitting a full nodeWhen a node is full, it divides the paperwork and informs management.2026-08-30T12:54:39+00:00monthlyhttps://read0nly.net/2026/08/12/designing-distributed-software-systems-a-practical-introduction/https://read0nly.net/wp-content/uploads/2026/08/distributed-architecture-primary-replication-20260813.pngDistributed application with primary database replicationApplication traffic and database replication are separate flows; replication runs from the primary database to the read replica.https://read0nly.net/wp-content/uploads/2026/08/distributed-saga-visual-revision.pngDistributed order saga, revisedA saga: less magic wand, more carefully organised bookkeeping.https://read0nly.net/wp-content/uploads/2026/08/distributed-architecture-visual-revision.pngAnatomy of a distributed system, revisedEverything looks beautifully orderly until one arrow takes five seconds to reply.https://read0nly.net/wp-content/uploads/2026/08/distributed-saga-en.pngA distributed order sagaA saga: less magic wand, more carefully organised bookkeeping.https://read0nly.net/wp-content/uploads/2026/08/distributed-architecture-en.pngAnatomy of a distributed systemEverything looks beautifully orderly until one arrow takes five seconds to reply.https://read0nly.net/wp-content/uploads/2026/08/distributed-saga.pngSaga di un ordine distribuitoUna saga: meno bacchetta magica, più contabilità ben organizzata.https://read0nly.net/wp-content/uploads/2026/08/distributed-architecture.pngAnatomia di un sistema distribuitoTutto molto ordinato, finché una freccia non impiega cinque secondi a rispondere.2026-08-30T12:54:33+00:00monthlyhttps://read0nly.net/2026/08/13/memoisation-in-python-stop-solving-the-same-problem-twice/https://read0nly.net/wp-content/uploads/2026/08/memoisation-partial-call-tree-objective-20260813.pngMemoisation and a partial Fibonacci call treePartial call tree: repeated subproblems are shown selectively; memoisation stores one result per input.https://read0nly.net/wp-content/uploads/2026/08/memoisation-recursion-comparison-visual-revision.pngPlain recursion and memoised recursion, revisedPlain recursion keeps reopening old cases. Memoisation gives it a notebook.https://read0nly.net/wp-content/uploads/2026/08/memoisation-recursion-comparison-v2.pngPlain recursion and memoised recursionPlain recursion keeps reopening old cases. Memoisation gives it a notebook.https://read0nly.net/wp-content/uploads/2026/08/memoisation.pngRepeated Fibonacci work and memoisationPlain recursion keeps reopening old cases. Memoisation gives it a notebook.2026-08-30T12:54:28+00:00monthlyhttps://read0nly.net/2026/08/13/claudes-invisible-watermark-is-a-clue-not-a-lie-detector/https://read0nly.net/wp-content/uploads/2026/08/claude-invisible-watermark-provenance.pngClaude invisible watermark and provenanceA useful clue, provided nobody promotes the magnifying glass to judge and jury.2026-08-30T12:52:43+00:00monthlyhttps://read0nly.net/2026/08/14/bloom-filters-in-python-a-tiny-index-that-sometimes-says-maybe/https://read0nly.net/wp-content/uploads/2026/08/bloom-filter-membership-v2.pngHow a Bloom filter answers a membership queryA few shared bits can answer definitely absent. Everything else gets a polite maybe.2026-08-30T12:52:31+00:00monthlyhttps://read0nly.net/2026/08/15/why-anthropic-moved-its-misalignment-risk-estimate-from-very-low-to-low/https://read0nly.net/wp-content/uploads/2026/08/anthropic-risk-label-uncertainty-2026-v2.pngAnthropic misalignment risk label updateOne rung higher on a qualitative scale, with the small print doing most of the heavy lifting.2026-08-30T12:52:26+00:00monthlyhttps://read0nly.net/2026/08/17/the-flyweight-pattern-in-python-one-style-many-objects/https://read0nly.net/wp-content/uploads/2026/08/flyweight-sharing-v3.pngFlyweight pattern: shared style stateThe coordinates keep their own seats. The style objects agree to share a table.2026-08-30T12:52:19+00:00monthlyhttps://read0nly.net/2026/08/21/a-rust-dependency-update-became-a-build-time-malware-delivery-system/https://read0nly.net/wp-content/uploads/2026/08/arrayref-supply-chain-v4.pngCargo build-time supply-chain attack diagram, revisedThe dangerous step happened before the application ran: Cargo compiled a dependency whose build script downloaded and launched the second stage.https://read0nly.net/wp-content/uploads/2026/08/arrayref-supply-chain-v3.pngCargo build-time supply-chain attack diagram, revisedThe dangerous step happened before the application ran: Cargo compiled a dependency whose build script downloaded and launched the second stage.https://read0nly.net/wp-content/uploads/2026/08/arrayref-supply-chain-v2.pngCargo build-time supply-chain attack diagram, revisedThe dangerous step happened before the application ran: Cargo compiled a dependency whose build script downloaded and launched the second stage.https://read0nly.net/wp-content/uploads/2026/08/arrayref-supply-chain.pngCargo build-time supply-chain attack diagramThe dangerous step happened before the application ran: Cargo compiled a dependency whose build script downloaded and launched the second stage.2026-08-30T12:52:13+00:00monthlyhttps://read0nly.net/2026/08/21/the-birthday-problem-in-python-estimate-collision-risk-before-ids-do/https://read0nly.net/wp-content/uploads/2026/08/birthday-collision-v3.pngThe birthday problem: collision risk in pairsCollision risk comes from the number of pairs, not only from how full the identifier space looks.https://read0nly.net/wp-content/uploads/2026/08/birthday-collision-v2.pngThe birthday problem: collision risk in pairsCollision risk comes from the number of pairs, not only from how full the identifier space looks.https://read0nly.net/wp-content/uploads/2026/08/birthday-collision.pngThe birthday problem: collision risk in pairsCollision risk comes from the number of pairs, not only from how full the identifier space looks.2026-08-30T12:52:07+00:00monthlyhttps://read0nly.net/2026/08/22/openai-pauses-some-frontier-rl-training-as-cyber-safeguards-expand/https://read0nly.net/wp-content/uploads/2026/08/openai-pacing-revised-v2.pngOpenAI frontier safeguards, corrected synthesis diagramCase-specific Read0nly synthesis of OpenAI's disclosed changes. The arrows connect the reported capability signal, controls and selected pauses; they are not a published OpenAI process diagram. The 20% estimate applies to monitored inference, not to all company compute.https://read0nly.net/wp-content/uploads/2026/08/openai-pacing-revised.pngOpenAI frontier safeguards, corrected synthesis diagramCase-specific Read0nly synthesis of OpenAI's disclosed changes. The arrows connect the reported capability signal, controls and selected pauses; they are not a published OpenAI process diagram. The 20% estimate applies to monitored inference, not to all company compute.https://read0nly.net/wp-content/uploads/2026/08/openai-pacing.pngOpenAI frontier research safeguards diagramOpenAI describes a chain from a preliminary cyber-capability signal to tighter controls and slower frontier work. The 20% estimate applies to monitored inference, not to all company compute.2026-08-30T12:52:01+00:00monthlyhttps://read0nly.net/2026/08/23/a-humanoid-robot-ran-100-metres-in-9-39-seconds-here-is-what-the-result-does-and-does-not-show/https://read0nly.net/wp-content/uploads/2026/08/humanoid-100m-v3.pngReported humanoid robot test attributionThe stopwatch is simple. The source trail is not.https://read0nly.net/wp-content/uploads/2026/08/humanoid-100m-v2.pngA reported humanoid robot race result, three comparisonsThe stopwatch is simple. The comparison is not.https://read0nly.net/wp-content/uploads/2026/08/humanoid-100m.pngA fast humanoid robot, three different comparisonsThe stopwatch is simple. The comparison is not.2026-08-30T12:51:56+00:00monthlyhttps://read0nly.net/2026/08/24/dijkstra-in-python-replace-the-full-scan-with-a-heap/https://read0nly.net/wp-content/uploads/2026/08/dijkstra-next-move-v2-1200.pngDijkstra next move diagramDijkstra picks the smallest known estimate, then gives its neighbours a chance to improve.2026-08-30T12:51:50+00:00monthlyhttps://read0nly.net/2026/08/25/gpt-5-6-in-kiro-two-milestones-one-provider-reported-cost-claim/https://read0nly.net/wp-content/uploads/2026/08/gpt56-kiro-timeline.pngGPT-5.6 in Kiro timelineKiro's public records show a July availability entry; the 82% figure is a provider-reported result for successful Terminal-Bench 2.1 tasks.2026-08-30T12:51:44+00:00monthlyhttps://read0nly.net/2026/08/26/topological-sorting-in-python-turn-dependencies-into-a-build-order/https://read0nly.net/wp-content/uploads/2026/08/topological-sort-scan-or-queue-v2.pngTopological sorting scan and queue diagramThe scan keeps circling back. The queue waits for the next task to become ready.2026-08-30T12:51:38+00:00monthlyhttps://read0nly.net/2026/08/28/horners-method-in-python-evaluate-polynomials-with-less-repeated-work/https://read0nly.net/wp-content/uploads/2026/08/horner-method-comparison-v2.pngHorner's method comparison, corrected Read0nly illustrationRead0nly illustration: the direct power loop and Horner's recurrence shown here use 2n and n multiplications respectively for degree n. The count is exact for these implementations; timing gains remain workload-dependent.https://read0nly.net/wp-content/uploads/2026/08/horner-method-comparison.pngHorner's method comparison, Read0nly illustrationRead0nly illustration: direct power accumulation extends and applies each power, while Horner's recurrence folds coefficients into one accumulator. The count is exact for the implementations compared in the article; timing gains remain workload-dependent.2026-08-30T12:51:33+00:00monthlyhttps://read0nly.net/2026/08/29/openai-is-winding-down-cursor-access-what-the-contract-notice-actually-says/https://read0nly.net/wp-content/uploads/2026/08/cursor-openai-contract-timeline-20260829-v2.pngOpenAI and Cursor contract timeline, corrected Read0nly illustrationRead0nly illustration: OpenAI announced a proposed contractual wind-down on 28 August, with 12 November 2026 as the proposed end date. The timeline is an editorial synthesis, not a claim that Cursor will close.https://read0nly.net/wp-content/uploads/2026/08/cursor-openai-contract-timeline-20260829.pngOpenAI and Cursor contract timeline, Read0nly illustrationRead0nly illustration: OpenAI announced a proposed contractual wind-down on 28 August, with 12 November 2026 as the proposed shutoff date. The timeline is an editorial synthesis, not a claim that Cursor will close.2026-08-30T12:51:28+00:00monthlyhttps://read0nly.net/2026/08/30/galapagos-coral-records-put-recent-el-ninos-in-a-millennium-scale-context/https://read0nly.net/wp-content/uploads/2026/08/galapagos-coral-el-nino-record-20260830-v2.pngGalápagos coral record and El Niño variability, corrected Read0nly illustrationRead0nly illustration: the coral archive records unusually large recent eastern-Pacific El Niño variability in the study comparison. The three panels are adjacent evidence states, not a causal chain or prediction.https://read0nly.net/wp-content/uploads/2026/08/galapagos-coral-el-nino-record-20260830.pngGalápagos coral record and El Niño variability, Read0nly illustrationRead0nly illustration: the coral archive records unusually large recent eastern-Pacific El Niño variability in the study comparison. The arrows are an editorial comparison, not an official causal diagram, and the future panel is not a forecast.2026-08-30T08:44:27+00:00monthlyhttps://read0nly.net/2026/08/27/openai-says-russian-linked-operators-used-chatgpt-to-promote-a-purported-expert-institute/https://read0nly.net/wp-content/uploads/2026/08/influence-operation-v2.pngAI influence operation evidence mapEditorial synthesis of OpenAI’s account. The dashed links show reported associations, not a process sequence; ChatGPT appears as one content layer in a wider operation.https://read0nly.net/wp-content/uploads/2026/08/influence-operation.pngAI influence operation evidence mapEditorial synthesis of OpenAI’s account. ChatGPT appears as one content layer in a wider operation; the report does not establish state control or large-scale persuasion.2026-08-27T10:54:05+00:00monthlyhttps://read0nly.net/2026/08/23/qwen3-8-27b-on-local-hardware-what-32-gb-actually-tells-us/2026-08-23T21:43:11+00:00monthlyhttps://read0nly.net/2026/08/22/epigenetic-clocks-are-responding-to-ageing-interventions-that-is-not-yet-proof-of-longer-life/https://read0nly.net/wp-content/uploads/2026/08/epigenetic-ageing-study-v2.pngCorrected epigenetic ageing study response mapRead0nly editorial synthesis of the Nature Medicine study. It compares the evidence path from intervention studies to biomarker response, and marks the unresolved step between a changed score and a proven clinical outcome.https://read0nly.net/wp-content/uploads/2026/08/epigenetic-ageing-study.pngEpigenetic ageing study response mapRead0nly editorial synthesis of the Nature Medicine study. It compares the evidence path from intervention studies to biomarker response, and marks the unresolved step between a changed score and a proven clinical outcome.2026-08-23T05:30:58+00:00monthlyhttps://read0nly.net/2026/08/23/microsoft-patches-cosnitch-a-one-click-copilot-attack-chain/https://read0nly.net/wp-content/uploads/2026/08/cosnitch-v2.pngCoSnitch Copilot test chain, Read0nly illustrationRead0nly editorial synthesis of the CoSnitch disclosure in Copilot Personal. It shows the tested sequence described by Varonis and corroborating reports, not a Microsoft architecture diagram or evidence of exploitation in the wild.https://read0nly.net/wp-content/uploads/2026/08/cosnitch.pngCoSnitch Copilot test chain, Read0nly illustrationRead0nly editorial synthesis of the CoSnitch disclosure. It shows the tested sequence described by Varonis and corroborating reports, not a Microsoft architecture diagram or evidence of exploitation in the wild.2026-08-22T17:46:34+00:00monthlyhttps://read0nly.net/2026/08/21/amdahls-law-in-python-why-one-fast-function-rarely-fixes-the-whole-programme/2026-08-20T14:17:52+00:00monthlyhttps://read0nly.net/2026/08/20/union-find-in-python-keep-connected-components-without-rescanning-everything/2026-08-20T14:10:00+00:00monthlyhttps://read0nly.net/info/2026-08-19T23:28:31+00:00weekly0.6https://read0nly.netdaily1.02026-10-01T16:40:20+00:00