{ "version": "2026-11-04.3", "reviewed_at": "2026-11-04", "recorded_at": "2026-11-02", "selected_case": "This is one selected native result, not a complete account of current research.", "learning": { "title": "A repaired model. A better next experiment.", "summary": "In a bounded macOS pipe investigation, Tend learned a model of notifications, found a contradiction, and repaired it. That repair helped it choose a useful later observation sooner.", "before": "The three-state model missed a distinction between an empty pipe with notifications enabled and an empty pipe whose dispatched notification remained disabled.", "observation": "An actual pipe observation contradicted the model. Refinement separated the missing fourth state.", "after": "In a later servicing task, the repaired model chose drain then enable directly. The preceding model first tried draining twice; a new observation disproved that choice.", "scope": "One bounded native interface on macOS, with supplied actions, reset semantics, observation projection and task predicates. This is not evidence of transfer across unfamiliar systems.", "initial": { "states": 3, "queries": 52, "actions": 148 }, "repaired": { "states": 4, "queries": 86, "actions": 362 }, "prediction_catalogue": { "correct": 64, "total": 64, "applies_to": "Both models on the later prediction catalogue" }, "comparison": [ { "method": "Repaired model", "acquisitions": 2, "actions": 12, "syscalls": 38, "verified_jobs": 4 }, { "method": "Preceding model", "acquisitions": 3, "actions": 19, "syscalls": 58, "verified_jobs": 4 }, { "method": "Direct procedure", "acquisitions": 2, "actions": 13, "syscalls": 39, "verified_jobs": 4 }, { "method": "Source-aware procedure", "acquisitions": 2, "actions": 12, "syscalls": 38, "verified_jobs": 4 }, { "method": "Shortlex search", "acquisitions": 3, "actions": 19, "syscalls": 58, "verified_jobs": 4 } ], "limits": [ "The source-aware ordinary procedure ties the repaired model. Superiority over the strongest ordinary method is not established.", "The comparison measures additional investigation work after common history. It does not show that learning and repair pay for themselves over the complete lifecycle.", "Both models achieve 64/64 predictions on the later catalogue. The demonstrated benefit is query selection at this endpoint, not higher prediction accuracy.", "The interface, actions, reset contract and task predicates were supplied. Tend did not discover these representations.", "The earlier two-versus-six comparison was invalid because its preceding-model control ignored an already paid prefix. The corrected, evidence-aware result is two versus three.", "The producer receipt binds the aggregate result, not every individual supporting file. Later audits do not retroactively authenticate the entire producer chain." ] }, "recovery": { "title": "Experience that survives a tested interruption", "summary": "In a tested macOS workflow, Tend retained its learned model and earlier verified result through a forced interruption, then continued the same job.", "scope": "macOS arm64; an owned nonblocking pipe and kqueue; a supplied fixed action plan. Continuation used the same job, executable, clock window and original limits. The interruption was a SIGKILL after a native action and before its journal commit.", "retained_entries": 4, "retained_states": 4, "verified_services_after_continuation": 2, "limit": "This demonstrates recovery at the exercised cut. It does not promise recovery at every instruction, after a reboot, under power loss, or after moving a job. It is engineering evidence, separate from the learning comparison." }, "research_updates": [ { "id": "model-descendants", "recorded_at": "2026-11-03", "kind": "Mechanism demonstrated", "title": "Generate alternatives from an earned model", "summary": "A Rust research prototype generated four alternatives from a model learned through observation. Each alternative changed one output in that two-state model and retained its parentage.", "finding": "In the authored example, a contrary observation refuted the parent and narrowed two possible target forecasts to one. A separate actual target observation was still required to establish the answer.", "limit": "The mutation rules, output alphabet and question were supplied. The family stayed fixed after generation. This demonstrates model-descendant generation, not autonomous evolution or a proven improvement in later investigations.", "cost": "Direct observation was cheaper: 2 queries and 6 attempted steps versus 3 queries and 8 steps with the diagnostic, for the same supported answer. Donor preparation cost another 10 queries and 22 steps.", "status_note": "Research-only prototype; no supported runtime or installed release changed. Fifteen focused checks passed, but the producer retained a failed closeout after a stale excluded-test count. A separate saved-evidence reader checked the named passes and trace without rerunning them; the earlier compile failure remains recorded.", "comparison": [ { "method": "Prefix, diagnostic, actual target", "queries": 3, "steps": 8 }, { "method": "Prefix, direct actual target", "queries": 2, "steps": 6 } ], "preparation": { "queries": 10, "steps": 22, "states": 2, "descendants": 4 }, "source": { "path": "docs/bounded-model-variation-result-v1.md", "revision": "a80bcbe3f1a29369983b85970c1f76ab86362857", "blob": "f9c9abfd8fc8c76b57a746f1391c61d0852a047a" }, "card_limit": "Supplied mutation rules; fixed family. Direct observation was cheaper." }, { "id": "shared-action-limit", "recorded_at": "2026-11-04", "kind": "A useful limit", "title": "Different guidance can share the same blind spot", "summary": "In a study of how to allocate investigation time, guidance from past experience and current progress pointed to the same next action. A read-only reconstruction confirmed the agreement at both decisions under review.", "finding": "The nominations agreed before and after the shared affordability filter. A gate that only chooses between those nominees cannot change either action. Agreement also did not certify a good choice: one prior recorded outcome was worse than a competent ordinary procedure.", "limit": "This closes that particular two-ranking intervention, not the broader model-population or transfer idea. Earlier disagreements and unsupported current scores remain in the record; they did not establish an actionable difference at this selected boundary.", "cost": "The check reconstructed 27 saved decision views. It made no native calls, fitted no gate, and acquired no new observations. Retained preparation still carried 2,680.504419 recorded seconds of prior cost.", "status_note": "This is a reconstruction of exposed saved evidence by one coordinator, not an independently executed applicability test, prospective learning result or installed acceptance. The broader question remains open: when advice agrees, what permitted observation could test a shared assumption before a useful later action?", "audit": { "views": 27, "selected_decisions": 2, "current_nominees": [ 6, 6 ], "retained_nominees": [ 6, 6 ], "native_calls": 0, "learner_updates": 0, "new_observations": 0 }, "source": { "path": "docs/learning-curve-applicability-premise-result-v1.md", "revision": "06246f160e8ec8d4966cf0ba27a5fc7ec57c8b59", "blob": "b7c064ab3444324b767bd5a0d7129f5d014bd4ca" }, "card_limit": "Choosing which guidance to trust could not change those decisions. Agreement did not establish that the shared choice was good." } ], "vision": { "goal": "We’re exploring whether learned models of systems—little worlds of how things work—can become a useful collection: alternatives Tend can test, connect and draw on during later investigations.", "boundary": "Connected world models, learned language and general self-improvement are research ambitions, not established capabilities.", "language": "A longer-term ambition is for Tend to learn to understand and communicate about those worlds through experience." }, "today": { "title": "Where we are today", "copy": "Tend is built in Rust and runs locally. In bounded experiments, it has learned and repaired models, and used a repair to choose a better later experiment. General self-improvement and learning across unfamiliar systems remain open research." }, "positioning": { "title": "Is Tend an LLM?", "copy": "No. Tend’s current predictive core learns explicit models from structured actions and observed outcomes, without pretrained language-model weights. Predictions guide tests; actual observations support answers. Models and evidence remain inspectable and reusable.", "boundary": "Today, people still supply the interfaces and task boundaries.", "source": { "path": "README.md", "blob": "aaebe20fa3a26586c0e9b20a88cb4b4864964287", "reviewed_at": "2026-11-04" } }, "provenance": { "repository": "jkordish/tend", "revision": "12255a772a54595200a97969aea63acdb59215f5", "records": [ { "path": "docs/kqueue-prefix-service-result-v2.md", "blob": "ac3825541f43c88eadcef97c1541ad5c5898eba1", "purpose": "Corrected learning comparison" }, { "path": "docs/native-service-session-result-v1.md", "blob": "ad682955ae8435519e159e52e61a5947d0d8f1c8", "purpose": "Installed recovery result" }, { "path": "research/native-service-release-v1/installed-v2-result.json", "blob": "8448b699aba9fa78312b1c555f8839b4694994de", "purpose": "Installed acceptance record" }, { "path": "research/native-service-release-v1/completion-audit-v3-result.json", "blob": "4a6da561505144ee05fd21a51688b0b1cceef7b7", "purpose": "Qualification acceptance record" } ], "public_material_boundary": "Public materials are curated summaries of committed result records. They are not original native traces, snapshots, execution archives, or an independent rerun. Private local paths and authorization records are omitted." } }