TensionLM-117M-TS-Reasoner-v10

v10 broadens the interactive CPU TS reasoner from v9.

Supported deterministic families:

• graph/transitivity,
• arithmetic traces,
• code traces,
• boolean logic,
• set operations,
• string transforms.

The reasoning path remains CPU-only and uses the frozen TensionLM-117M-Reasoning-v2 substrate plus explicit TS graph/program operators.

New v10 Families

Boolean logic:

python inference.py --prompt "Logic board: A=true; B=false; C=true. Evaluate A AND NOT B:" --category boolean_logic --json

Set reasoning:

python inference.py --prompt "Set ledger: A={a,b}; B={b,c}. Compute A union B:" --category set_reasoning --json

String reasoning:

python inference.py --prompt "String ops: start='alpha'; reverse; append 'x'. Result:" --category string_reasoning --json

Eval Receipts

Fixed benchmark scores:

System	TAC v2	TAC v3	TAC v4
TS-Reasoner-v10	120/120	120/120	120/120

Broadened benchmark scores:

Receipt	Score
Public v10 examples	30/30
Server smoke	pass
New families standard seed 9501	3000/3000
New families paraphrase seed 9502	3000/3000
New families unknown seed 9503	3000/3000
New families mixed seed 9505	3000/3000
All six families mixed seed 9504	6000/6000

For unknown prompts, correctness means explainable abstention. These are system scores, not raw LLM scores.

CLI and Server

python ts_reasoner_v10.py solve --prompt "Logic board: A=true; B=false. Evaluate A XOR B:" --category boolean_logic --json
python ts_reasoner_v10.py examples
python ts_reasoner_v10.py serve --host 127.0.0.1 --port 7860

Endpoints:

• GET /
• GET /healthz
• GET /examples
• POST /solve

Limitations

This artifact handles generated formal prompt families covered by the included operators. It is not a chat assistant, not raw model improvement, and not a claim of open-ended natural language understanding. The confidence values are rule-calibrated system signals over these families, not probabilities over all natural language.