TS-Reasoner v40.0.0: Self-Hosting Verifier-First Research OS

TS-Reasoner v40.0.0 adds the self-hosting verifier-first research OS milestone.

Release claim:

• TS-Reasoner composes bounded v32-v39 research OS surfaces into one deterministic self-hosting stack.
• The stack includes ontology compilation, verifier-gated memory, TS-AGL plan compilation, automated research forge planning, self-repair auditing, confirmation-gated patch staging, ecosystem audit, verifier/model co-evolution reporting, and a composed research OS receipt.
• The research OS can prepare a bounded next-release-candidate plan for TS-Reasoner.
• Generated text, model confidence, memory, repeated experience, curriculum examples, and router confidence are never proof authority.
• Patches are staged and require confirmation before mutation.
• External LLM usage, network calls, and real external side effects remain zero.
• Candidate graph contamination remains zero.

Boundary:

• Not AGI.
• Not autonomous science.
• Not unrestricted self-improvement.
• Not free self-learning.
• Not broad NLP understanding.
• Not live external automation.
• Not proof by confidence, memory, confirmation, or repeated experience.
• v40 composes bounded v32-v39 surfaces; it does not claim every subsystem is a mature standalone autonomous implementation.
• Typed verifier support remains the proof boundary.

Verification:

• python3 scripts/evaluate_v32_v40_research_os.py
• python3 -m unittest tests.test_research_os_v40 -q
• python3 -m unittest discover -q

TS-Reasoner v31.0.0: TS Project Curriculum Pack v1

TS-Reasoner v31.0.0 adds the TS Project Curriculum Pack v1.

Release claim:

• TS-Reasoner can be taught through bounded curriculum packs.
• The TS project domain now teaches repo, release, artifact, receipt, claim, proof boundary, unsafe overclaim, next safe action, stale public surface, and missing receipt concepts.
• Domain examples route through typed TSCall operations.
• Unsafe self-learning overclaims are rejected.
• Missing receipt detection is explicit.
• Durable lesson promotion remains confirmation-gated and receipt-gated.
• No external LLM is used.
• No network call is performed.
• No real external side effect is performed.
• Candidate graph contamination remains zero.

Boundary:

• Not free self-learning.
• Not open-ended memory growth.
• Not autonomous concept formation.
• Not arbitrary self-training over new domains.
• Domain packs, language examples, repeated experience, and router confidence are not proof authority.

TS-Reasoner v30.0.0: Verifier-First Local Agent OS

TS-Reasoner v30.0.0 adds the Verifier-First Local Agent OS milestone.

Release claim:

• TS-OS now composes a native conversational shell loop, evidence dashboard, proof object examples, first-contact demo, and verifier-first local agent OS runner.
• Natural-language turns route through typed TS-AGL calls.
• Unsafe or ambiguous requests safely abstain.
• External side-effect requests remain gated and do not execute real external effects.
• Proof object examples expose the typed verifier boundary.
• Evidence dashboard summarizes safety and receipt metrics.
• Candidate graph contamination remains zero.
• No external LLM is used.

Included:

• v26 chat loop
• v27 evidence dashboard
• v28 proof object examples
• v29 first-contact demo
• v30 composed verifier-first local agent OS

Verification:

• python3 -m unittest tests.test_ts_os_chat_loop tests.test_ts_evidence_dashboard tests.test_proof_object_examples tests.test_first_contact_demo tests.test_ts_os_v30 -q
• python3 -m unittest discover -q
• python3 scripts/run_ts_os_chat_loop_demo.py
• python3 scripts/evaluate_ts_os_chat_loop.py
• python3 scripts/run_first_contact_demo.py
• python3 scripts/evaluate_ts_os_v30.py

Boundary:

• Not broad autonomous agency.
• Not a chatbot claim.
• No external LLM.
• No real external side effect.
• Router confidence is not proof.
• Language is not proof authority.
• Proof examples are explanatory receipts, not model confidence.
• Writes/external effects remain gated.
• Candidate graph contamination remains zero.

TS-Reasoner v25.0.0: TS-OS v1

TS-Reasoner v25.0.0 adds TS-OS v1.

Release claim:

• TS-OS v1 packages the TS-AGL shell surface, persistent session ledger, router stack, local project operator, controlled external adapter gate, and receipts.
• Shell help and route inspection are included.
• Router stack arbitration is included.
• Persistent session save/load/replay is included.
• Local project operator is included.
• External adapter gate is included.
• No accidental network call is performed.
• No real external side effect is performed.
• Candidate graph contamination remains zero.
• No external LLM is used.

Boundary:

• bounded TS-OS v1 only
• no broad autonomous agency
• no external LLM
• no accidental network call
• no real external side effect
• language is not proof authority
• router confidence is not proof
• session replay is not proof authority
• external gate authorization is not execution
• candidate graph contamination remains zero

Verification:

• python3 -m unittest discover -q
• python3 scripts/validate_ts_agl_domain_packs.py
• python3 scripts/evaluate_ts_os_v1.py
• python3 scripts/run_ts_os_v1.py

TS-Reasoner v24.0.0: Controlled External Adapter Gate

TS-Reasoner v24.0.0 adds the Controlled External Adapter Gate.

Release claim:

• External adapter requests are dry-run by default.
• Missing confirmation is blocked.
• Wrong confirmation is blocked.
• Confirmed dry-run performs no network call.
• Live mode is blocked unless TS_AGL_ALLOW_EXTERNAL_LIVE=1.
• Env-unlocked live mode authorizes only a downstream adapter gate; the gate itself performs no network call.
• Candidate graph contamination remains zero.
• No external LLM is used.

Boundary:

• controlled external adapter gate only
• no live external automation
• no external LLM
• no network call
• no real external side effect
• live mode requires env unlock plus confirmation
• candidate graph contamination remains zero

Verification:

• python3 -m unittest discover -q
• python3 scripts/validate_ts_agl_domain_packs.py
• python3 scripts/evaluate_ts_agl_external_adapter_gate.py
• python3 scripts/run_ts_agl_external_adapter_gate.py

TS-Reasoner v23.0.0: Persistent TS-OS Session Ledger

TS-Reasoner v23.0.0 adds the Persistent TS-OS Session Ledger.

Release claim:

• TS-AGL shell commands can be persisted as TS-OS session events.
• Sessions can be saved to disk.
• Sessions can be reloaded.
• Sessions can be replayed as summaries.
• Resumed sessions can continue with new shell/operator commands.
• Candidate graph contamination remains zero.
• No external LLM is used.
• No real external side effect is performed.

Boundary:

• session persistence only
• no broad autonomous agency
• no external LLM
• no real external side effects
• session replay is not proof authority
• shell confidence is not proof
• candidate graph contamination remains zero

Verification:

• python3 -m unittest discover -q
• python3 scripts/validate_ts_agl_domain_packs.py
• python3 scripts/evaluate_ts_agl_shell_surface.py
• python3 scripts/evaluate_ts_os_session_ledger.py
• python3 scripts/run_ts_os_session_ledger.py --reset

TS-Reasoner v22.0.0: TS-AGL Shell Surface

TS-Reasoner v22.0.0 adds the TS-AGL Shell Surface.

Release claim:

• TS-AGL exposes a speakable shell command surface.
• The shell supports help, route inspection, and local project operator execution.
• python3 -m ts_agl.shell works as the module entrypoint.
• A ts-agl console script is declared.
• Shell invocations emit receipts.
• Local writes remain confirmation-gated through the operator stack.
• Candidate graph contamination remains zero.
• No external LLM is used.

Boundary:

• shell surface only
• no broad autonomous agency
• no external LLM
• no real external side effects
• shell confidence is not proof
• local writes remain confirmation-gated
• candidate graph contamination remains zero

Verification:

• python3 -m unittest discover -q
• python3 scripts/validate_ts_agl_domain_packs.py
• python3 scripts/evaluate_ts_agl_shell_surface.py
• python3 scripts/run_ts_agl_shell_surface.py "inspect project and stage next safe note"

TS-Reasoner v21.0.0: TS-AGL Local Project Operator

TS-Reasoner v21.0.0 adds the TS-AGL Local Project Operator.

Release claim:

• TS-AGL can coordinate a bounded local project workflow.
• The operator inspects repo state, release/tag state, and docs.
• The router stack is used for next-safe-action routing.
• A safe artifact write is staged.
• Unconfirmed mutation is blocked.
• Confirmed write executes and is receipted.
• No external side effect is performed.
• Candidate graph contamination remains zero.
• No external LLM is used.

Boundary:

• bounded local project workflow only
• no broad autonomous agency
• no external LLM
• no real external side effect
• confirmed local artifact write only
• router stack is not proof authority
• candidate graph contamination remains zero

Verification:

• python3 -m unittest discover -q
• python3 scripts/validate_ts_agl_domain_packs.py
• python3 scripts/evaluate_ts_agl_router_stack_arena.py
• python3 scripts/evaluate_ts_agl_local_project_operator.py
• python3 scripts/run_ts_agl_local_project_operator.py

TS-Reasoner v20.0.0: TS-AGL Router Stack Arena

TS-Reasoner v20.0.0 adds the TS-AGL Router Stack Arena.

Release claim:

• TS-AGL can compare rule-parser routing, domain-example routing, and tiny learned routing.
• A safe selector chooses routes or abstains to route_unknown.
• Hard negatives abstain.
• Selected calls still go through TSCall/risk/adapters/verifier boundaries.
• Learned confidence is not proof.
• Candidate graph contamination remains zero.
• No external LLM is used.

Boundary:

• router arbitration only
• no proof authority
• no external LLM
• confidence is not proof
• selected calls still use typed TSCall/risk/adapters/verifier boundaries
• candidate graph contamination remains zero

Verification:

• python3 -m unittest discover -q
• python3 scripts/validate_ts_agl_domain_packs.py
• python3 scripts/evaluate_ts_agl_router_stack_arena.py

TS-Reasoner v19.0.0: TS-AGL Tiny Learned Router

TS-Reasoner v19.0.0 adds the TS-AGL Tiny Learned Router.

Release claim:

• TS-AGL can train a tiny dependency-free router from the v18 trace-mined dataset.
• The router predicts operation routes or abstains.
• Heldout routing and abstention are evaluated.
• Routed calls still go through TSCall/risk/adapters/verifier boundaries.
• Learned confidence is not proof.
• Candidate graph contamination remains zero.
• No external LLM is used.

Boundary:

• learned routing only
• no proof authority
• no external LLM
• no neural dependency
• confidence is not proof
• routed calls still use typed TSCall/risk/adapters/verifier boundaries
• candidate graph contamination remains zero

Verification:

• python3 -m unittest discover -q
• python3 scripts/validate_ts_agl_domain_packs.py
• python3 scripts/build_ts_agl_router_dataset.py
• python3 scripts/evaluate_ts_agl_router_dataset.py
• python3 scripts/train_ts_agl_tiny_router.py
• python3 scripts/evaluate_ts_agl_tiny_router.py