{ "schemaVersion": "ke.research-findings-evidence.v1", "release": { "id": "ke-research-findings-2026-07-27", "title": "Native reuse beat repeated full text, not cached text", "date": "2026-07-27", "generatedAt": "2026-07-27T13:45:00-04:00", "informationCutoff": "2026-07-27T13:36:01-04:00", "canonicalUrl": "https://kestudios.dev/research/updates/2026-07-27", "author": { "name": "William Keenan", "organization": "K&E Studios Research", "url": "https://kestudios.dev/william-keenan", "email": "william@kestudios.dev" }, "reviewStatus": "source-verified local research update; not peer reviewed or independently validated" }, "sourceSnapshots": [ { "label": "Waggle and Kea committed snapshot", "commit": "daaaf5a6b148a185abdbd7efaa0dcae5fde5a0f0", "capturedAt": "2026-07-27T13:36:01-04:00", "inspectionBasis": "bytes addressable from the frozen commit only", "artifactSha256": [ "2daa72b763509459321ed81bafb4a0cd5f5b835a409e65de5606f1f522b60c10", "6ff978727606f189e09c14e3b833107e7c6f2e0c75865692e5e617e16ffd7c18" ] }, { "label": "HEATWAKE committed snapshot", "commit": "60ece195b275944e92adf157461b448206dca9c0", "capturedAt": "2026-07-27T13:36:01-04:00", "inspectionBasis": "bytes addressable from the frozen commit only; later commits excluded", "artifactSha256": [ "7791ca19681c45b2353bc502baffb609a76c10cee9a2c89e14c50dc8d626eb69", "371f36f6e2bfa12b81609484a85a80c8338913086c2e1c7a4b46bf7dc7b686ac" ] } ], "methodology": { "comparisonPolicy": "Inspect the accepted July 26 public checkpoint first, preregister changed claims and falsifiers, then compare only frozen committed source bytes available at the information cutoff.", "evidenceLevels": { "recorded": "present in a frozen attributable source artifact", "observed": "read or measured during the accepted bounded run", "verified": "reproduced by a zero-model replay, deep receipt check, or focused local test", "independentValidation": false }, "environment": "local macOS research workstation; one frozen local open-weight model and local hardware-accelerated runtime for the matched Waggle and Kea arms", "uncertainty": [ "Waggle and Kea exactness uses six deterministic task fixtures; no statistical confidence interval is claimed.", "Waggle and Kea wall times are descriptive measurements from one accepted run on one host.", "Process order was frozen, but operating-system cache and thermal state were not controlled.", "Direct energy was unavailable and was not estimated.", "HEATWAKE operator observations establish bounded custody and local projection behavior, not seven-day completeness, market-data quality, source admission, or model utility." ], "exclusions": [ "All uncommitted or later source-writer bytes after the information cutoff", "All active, growing, lock-held, or outcome-bearing source artifacts", "HEATWAKE raw events and segment receipts from this publication review", "All HEATWAKE raw payloads from model features, labels, calibration, candidate selection, signals, and the sealed final test", "Any causal adjustment for model-load order, cache state, or thermal state after observing the result" ], "publicSafety": "Public output was composed from accepted aggregate claims, non-reversible hashes, and bounded measurements. Non-public implementation materials, raw prompts, raw model state, market payloads, private topology, credentials, and reconstruction-enabling detail are omitted." }, "programs": [ { "id": "waggle-kea", "name": "Waggle + Kea", "verdict": "A bounded six-branch native-state reuse and independently qualified handoff mechanism worked. Resident native beat repeated full-text reconstruction after branch three but remained slower than cached-prefix text and fresh native controls, so overall efficiency is rejected.", "claimDeltas": [ { "claim": "One native-state service can support multiple source-separated tasks through Kea.", "priorStatus": "supported across three bounded tasks", "newEvidence": "five informed arms at 6/6 exact; one native restore served six branches; Kea qualified 6/6 outputs; a projection-only consumer continued", "currentStatus": "supported across these six bounded fixtures" }, { "claim": "Native reuse can beat repeated full-text reconstruction as branches accumulate.", "priorStatus": "rejected in the three-task result", "newEvidence": "resident native first became faster at branch three and remained faster through branch six; 20.285203541 seconds versus 36.140623541 seconds end-to-end", "currentStatus": "supported only against this repeated full-text control after branch three" }, { "claim": "Native reuse is the most efficient informed service in this experiment.", "priorStatus": "rejected", "newEvidence": "12.850184125 seconds cached-prefix text and 17.881461833 seconds fresh native versus 20.285203541 seconds resident native", "currentStatus": "rejected" }, { "claim": "Kea can qualify the handoff without becoming an authority source.", "priorStatus": "supported across three projections", "newEvidence": "six decisions qualified; forged ledger lineage rejected against a separate trust root; projection-only consumer continued; no-state consumer abstained", "currentStatus": "supported across these six bounded fixtures" } ], "experiment": { "question": "Can a bounded local native-state reuse result pass independent Kea qualification and beat matched text and fresh-process controls?", "falsifiers": [ "Any informed arm fails an exact expected outcome.", "Kea cannot independently qualify every accepted output.", "The restricted consumer requires producer-only inputs instead of the approved projection.", "A stronger equally informed baseline matches or beats the claimed advantage.", "Formal replay cannot reproduce every pinned artifact without additional model work." ], "design": { "tasks": 6, "arms": [ "resident native", "resident cached-prefix text", "repeated resident full text", "fresh native", "fresh full text", "no state" ], "informedArms": 5, "sameFrozenTasksAndExpectedValues": true, "equalInputTokenCountPerInformedArm": true, "sameLocalModelFamily": true, "restrictedConsumerReadsKeaProjectionOnly": true, "maximumModelProcessLoads": 16, "maximumForwardDecodeSteps": 2048 }, "results": { "qualifiedTasks": "6/6", "residentNativeExact": "6/6", "residentCachedTextExact": "6/6", "residentFullTextExact": "6/6", "freshNativeExact": "6/6", "freshFullTextExact": "6/6", "noStateQualifiedProjection": false, "noStateDisposition": "abstain", "restrictedConsumerDisposition": "continue", "residentFullTextBreakEvenBranchCount": 3, "cachedTextBreakEvenBranchCount": null, "freshNativeBreakEvenBranchCount": null, "wallTimeMilliseconds": { "residentNative": 20285.203541, "residentCachedText": 12850.184125, "residentFullText": 36140.623541, "freshNative": 17881.461833, "freshFullText": 39570.006709 }, "completionCurveMilliseconds": [ { "branches": 1, "residentNative": 13779.463792, "residentCachedText": 6496.367208, "residentFullText": 6339.878084, "freshNativeCumulative": 2724.857792 }, { "branches": 2, "residentNative": 14986.78725, "residentCachedText": 7719.156708, "residentFullText": 12221.531625, "freshNativeCumulative": 5292.213417 }, { "branches": 3, "residentNative": 16184.913542, "residentCachedText": 8923.272542, "residentFullText": 18110.261125, "freshNativeCumulative": 7924.608542 }, { "branches": 4, "residentNative": 17247.435917, "residentCachedText": 10003.4635, "residentFullText": 23861.428584, "freshNativeCumulative": 10373.340417 }, { "branches": 5, "residentNative": 18442.571917, "residentCachedText": 11202.695667, "residentFullText": 29750.827625, "freshNativeCumulative": 12920.080709 }, { "branches": 6, "residentNative": 19848.887042, "residentCachedText": 12606.106833, "residentFullText": 35837.325, "freshNativeCumulative": 15650.045209 } ], "acceptedResidentModelProcessLoads": 3, "acceptedFreshModelProcessLoads": 12, "acceptedTotalModelProcessLoads": 15, "acceptedForwardDecodeSteps": 690, "nativeRetainedStateBytes": 279321005, "nativeQueryBytes": 2307, "fullTextPromptBytes": 53415, "maximumPeakRssBytes": 2023538688, "directEnergyMetricAvailable": false, "reportSha256": "6ff978727606f189e09c14e3b833107e7c6f2e0c75865692e5e617e16ffd7c18" }, "controls": { "formalReplayPins": 23, "formalReplayTasks": 6, "formalReplayAcceptedLoads": 15, "formalReplayAdditionalLoads": 0, "formalReplayAdditionalForwardDecodeSteps": 0, "priorCheckpointReplayPassed": true, "serviceSuiteTasksPassed": "6/6", "adversarialInputsRejected": "13/13", "forgedLineageRejected": true, "typecheckPassed": true, "buildPassed": true }, "keaBoundary": { "outputsQualified": "6/6", "separateTrustRootPinned": true, "forgedLedgerLineageRejected": true, "consumerReceivedOriginalPrefix": false, "consumerReceivedSourceReport": false, "consumerReceivedOracle": false, "consumerReceivedParentNativeState": false, "consumerLoadedModel": false, "consumerAuthorityGranted": false }, "supportedClaim": "One bounded local native-state reuse service, independent Kea qualification path, and projection-only consumer handoff worked across six frozen tasks; resident native beat repeated full-text reconstruction after branch three.", "negativeClaim": "Resident native did not beat cached-prefix text or fresh native and did not establish overall efficiency.", "notSupported": [ "A learned language, neuralese, hidden-state semantic transfer, or unique native capability", "A semantic, quality, general-reasoning, or cross-model advantage", "Token, Credit, provider-cost, memory, storage, network, energy, or overall-efficiency savings", "Production reliability, deployment, customer value, or authority execution" ] } }, { "id": "heatwake", "name": "HEATWAKE", "verdict": "The successor source-admission method is frozen before outcome access and the local read-only operator now handles legal reconnect states. Collection remains incomplete at 0/7 full UTC days, raw source is not admitted, and every model, signal, edge, profitability, trading, and production claim remains closed.", "claimDeltas": [ { "claim": "HEATWAKE has a fixed V4 source-admission method before seeing the collection result.", "priorStatus": "a V4-specific adapter was the next required dependency", "newEvidence": "hash-bound audit contract and preregistration fix terminal-session, closed-receipt, seven-day, coverage, freshness, continuity, and refusal rules with automatic admission disabled", "currentStatus": "method frozen; outcome not run and source not admitted" }, { "claim": "The read-only operator can represent legal reconnect states without weakening refusal checks.", "priorStatus": "two legal ABSTAIN transitions could be misclassified as identity drift", "newEvidence": "transition recovery accepts only the two contract-defined reconnect shapes while retaining malformed identity, hash, counter, process, effect, and handoff refusal", "currentStatus": "supported for the local loopback operator" }, { "claim": "Current raw collection changes the science or trading evidence level.", "priorStatus": "raw custody only at 0/7 complete days", "newEvidence": "one advancing collector, 15 closed of 16 started segments, 14 retained reconnect gaps, zero protocol refusals, 0/7 complete days", "currentStatus": "no upgrade; raw custody and ABSTAIN" } ], "sourceAdmission": { "question": "Can the completed successor session, if one is eventually produced, be audited against rules frozen before outcome access without reading active data or weakening downstream science gates?", "status": "WAIT_PSR_04_V4_COLLECTION_COMPLETE", "requiredFullUtcDays": 7, "terminalSessionReceiptPresent": false, "resultExists": false, "automaticAdmission": false, "activeAppendReads": 0, "featuresWritten": false, "labelsBound": false, "modelFitOpen": false, "finalTestOpen": false, "scienceThresholdsChanged": false, "contractSha256": "7791ca19681c45b2353bc502baffb609a76c10cee9a2c89e14c50dc8d626eb69", "checks": { "focusedAdmissionTestsPassed": "11/11", "terminalReceiptRequired": true, "closedReceiptPairingRequired": true, "activeOrUnreceiptedFilesRefused": true, "sevenCompleteUtcDaysRequired": true, "dailyCoverageRequired": true, "freshnessAndContinuityChecksFixed": true, "automaticAdmissionDisabled": true } }, "operatorProjection": { "question": "Can the local GET-only observer remain available during legal reconnect states without changing collection or science authority?", "releaseState": "local loopback only; not deployed", "method": "read-only, no-store projection; POST refused", "acceptedTransitionShapes": [ "no active segment during bounded reconnect backoff", "new segment awaiting its first snapshot" ], "acceptedObservation": { "completedFullUtcDays": 0, "requiredFullUtcDays": 7, "segmentsCompleted": 15, "segmentsStarted": 16, "transportGaps": 14, "transportReconnects": 14, "protocolRefusals": 0, "sourceAdmissionAuditReady": false, "rawSourceAdmitted": false, "modelInputEligible": false, "policyAction": "ABSTAIN", "artifactSha256": "371f36f6e2bfa12b81609484a85a80c8338913086c2e1c7a4b46bf7dc7b686ac" }, "controls": { "writeOnceReceiptsVerified": 2, "operatorServiceTestsPassed": 97, "focusedBrowserContractTestsPassed": 2, "frontendTestFilesPassed": 36, "frontendTestsPassed": 117, "postMethodStatus": 405, "lintPassed": true, "typecheckPassed": true, "buildPassed": true }, "effects": { "activeEventReads": 0, "segmentReceiptReads": 0, "featureWrites": 0, "labelWrites": 0, "modelFits": 0, "finalTestOpens": 0, "orders": 0, "capitalUsd": 0, "spendUsd": 0, "deployments": 0 } }, "notSupported": [ "A completed seven-day collection or terminal session receipt", "Source admission, an admitted model corpus, feature or label construction, a model fit, calibration, or final-test result", "A forecast, signal, market edge, profitability, execution result, or trading readiness", "Production reliability, deployment, credentials, orders, trades, capital, or spend" ] } ], "verification": { "performedAt": "2026-07-27T13:45:00-04:00", "environment": "local macOS research workstation", "runs": [ { "program": "Waggle + Kea", "scope": "six-task service and adversarial refusal controls", "result": "PASS_ZERO_MODEL", "observed": "6/6 tasks passed; 13/13 adversarial mutations rejected; 0 added model loads or forward/decode steps" }, { "program": "Waggle + Kea", "scope": "formal replay and prior-checkpoint regression", "result": "PASS_ZERO_MODEL", "observed": "23 pins, 6 tasks, 15 accepted loads, and 690 accepted forward/decode steps verified; C15c prior replay passed; 0 added model work or effects" }, { "program": "Waggle + Kea", "scope": "strict TypeScript and build", "result": "PASS", "observed": "typecheck and build completed from the frozen committed source" }, { "program": "HEATWAKE", "scope": "V4 source-admission contract", "result": "PASS_WAIT", "observed": "11/11 focused admission checks passed; preregistration remains WAIT; no result exists" }, { "program": "HEATWAKE", "scope": "write-once operator receipts and service", "result": "PASS", "observed": "2/2 receipts deep-verified; 97/97 service tests and 2/2 focused browser-contract tests passed" }, { "program": "HEATWAKE", "scope": "frontend suite, lint, typecheck, and build", "result": "PASS", "observed": "36/36 test files and 117/117 tests passed; ESLint, strict TypeScript, and optimized build passed" } ], "verificationEffects": { "providerCalls": 0, "newLocalModelLoads": 0, "newForwardDecodeSteps": 0, "trainingRuns": 0, "paidRequests": 0, "emailSends": 0, "newMarketCaptureRequests": 0, "newMarketBytesCaptured": 0, "heatwakeSourceWrites": 0, "heatwakeFinalTestOpens": 0, "orders": 0, "trades": 0, "capitalUsd": 0, "authorityGrants": 0, "sourceDeployments": 0 } }, "releaseNonClaims": [ "Not peer reviewed or independently validated", "No learned language, neuralese, hidden-state semantic transfer, or unique native capability", "No semantic, quality, token, Credit, cost, memory, storage, energy, or overall-efficiency advantage", "No completed HEATWAKE seven-day collection, terminal receipt, source admission, admitted model corpus, or final-test result", "No forecast, signal, market edge, profitability, execution result, trading readiness, or production reliability", "No customer adoption, revenue, commercial validation, deployment, or authority claim" ] }