BoggersTheFish / TS Research Archive

Building the Thinking System.

Graph-based reasoning. Tension telemetry. Constraint propagation. Reproducible traces.

Exploring small, inspectable foundations for interpretable and reliable AI.

bounded verifier path

git clone https://github.com/BoggersTheFish/TS-Reasoner-v0
cd TS-Reasoner-v0
python3 inference.py \
  --question "If some artists are makers and all makers are creators, are all artists creators?" \
  --premise "Some artists are makers." \
  --premise "All makers are creators."
Start HereExplore ResearchRun TS-Reasoner
Live

TS-Reasoner v10.0.0

Runnable v10 verifier-first runtime with typed proof-boundary receipts.

verifier runtimepolicy contracts
Explore TS-Reasoner v10.0.0
Active

TS-Core

Minimal graph/tension runtime for inspectable constraint propagation and relaxation.

graph runtimetension telemetry
Explore TS-Core
Experimental

TensionLM

Sigmoid tension attention experiments with model artifacts and bounded benchmark claims.

language modelssigmoid tension
Explore TensionLM
In Progress

CIG

Persistent provenance-aware claim/evidence graph for bounded research claims.

provenanceknowledge graph
Explore CIG
Published

Proof Ranker

Verifier-backed proof scoring and repair ladder published as Hugging Face artifacts.

proof rankingverifiers
Explore Proof Ranker

Quick Links

Start HereBegin with the framework map.Proof BankExperiments, receipts, and limitations.GitHubSource repositories, experiments, and project code.Hugging FaceModel artifacts, checkpoints, and research receipts.Canonical HomeThe public entrypoint for the research archive.DocsGuides, references, and technical notes.

Boggers Credits (BGC)

Support the work without financial promises.

Donations are donations. Boggers Credits (BGC) are experimental off-chain community credits: not cryptocurrency, not an investment, not legal tender, not equity, not redeemable for cash, may remain off-chain forever, and not financial advice.

not crypto / not investment

Learn more about BGC

Current stack

Verifier first, models second

The public surface is organized around what can be inspected today, not broad claims about general reasoning.

TS-Reasoner

verified local repo

Role

Verifier/control loop

Evidence

v10 runtime OS receipt, policy contracts, checkpoint/restore, recovery drill, zero candidate graph contamination.

TensionLM

verified HF metadata

Role

Model-line experiments / tension-field learning

Evidence

HF-visible v10 reasoner artifact and v11 trace-distilled student.

CIG

bounded project claim

Role

Provenance-aware claim/evidence graph

Evidence

Claim discipline, contradiction framing, and revision-oriented proof notes.

Proof Ranker

receipt sync needed

Role

Proof scoring / repair ladder

Evidence

Published artifact ladder is linked; exact receipt sync remains a TODO.

Receipts before claims

Latest visible receipts

These cards only state what is present in local artifacts or public Hugging Face metadata/model cards. Broad AGI or general-reasoning claims are deliberately absent.

TensionLM-TS-Trace-Distilled-v11

external verified

HF API/card show a compact trace-distilled student, 1,920 v10 trace rows, 580 training steps, and raw held-out exact answer/rule hits of 0/48.

The v11 card says raw exact answer/rule generation is not solved; v10 remains the working reasoner.

Inspect receipt

TensionLM-117M-TS-Reasoner-v10

external verified

HF card lists deterministic graph, arithmetic, code, boolean, set, and string families with bounded system receipts.

The card scopes the scores to generated formal families and says they are not raw LLM scores.

Inspect receipt

TS-Reasoner v10 verifier-first runtime OS

local verified

Release receipt shows replay, policy contracts, tamper-evident ledger, checkpoint/restore, and recovery under typed verifier authority.

Runtime integrity is not claim truth; not a chatbot, theorem prover, or broad language-understanding claim.

Inspect receipt

Golden-path abstention trace

local verified

A local run rejects the overstrong all-artists conclusion and settles on 'Not enough information.'

Single bounded example, useful as a trace-contract demo rather than a capability claim.

Inspect receipt

Claim boundary

Toy and bounded receipts are real. Broad claims are not.

No claim that TS proves AGI.

No claim that TensionLM beats transformers generally.

No broad benchmark numbers without linked receipts.

Dates and versions are verified or labelled local/draft.

Choose your path

A cleaner way into the archive

Different visitors need different first moves. This section keeps the serious routes obvious.

I'm new

Read the plain-language map before opening project details.

Open route

I'm technical

Inspect the runtime, model, graph, and proof-control surfaces.

Open route

I want evidence

Start with claims, setups, results, limits, and replay notes.

Open route

I want to support

Support research, compute, writing, docs, and open-source work.

Open route