SHARED-STATE DECISION MODEL

One state.
Many decisions.

Give Blitz one view of the situation. Ask a panel of focused questions. Get explicit distributions that software can compose into its next action—without generating a paragraph.

See the workflow → Try one decision

shared state→question panel→ distributions→code action

01 / SHARED-STATE FANOUT

One trace. Four focused decisions.

Blitz acts like a semantic sensor panel. Each question reads the same evidence independently; ordinary code combines the resulting distributions.

SHARED STATE

Agent recovery trace

Three attempts changed the same files and reproduced identical failures. Valid observations contradicted a necessary prediction of the working hypothesis.

Attempts3
Repeated failureYes
BudgetMedium

01Same root cause?

YES81.9%

02Would another retry help?

NO96.8%

03Hypothesis invalidated?

YES80.0%

04Needs specialist review?

NO62.3%

CODE COMPOSES

if same_root_cause > .80
  and retry_helps < .20
  and hypothesis_invalid > .70:
    → change_strategy

Q4 · 62.3%below .70 → unknown

02 / INSPECT THE PRIMITIVE

Try one atomic decision.

Checking endpoint…

A workflow is built from small, focused judgments. Start with a real checkpoint example, then change the evidence, question, or options.

OUTPUT AWAITING RESULT

Your decision appears here

Ready when you are.

03 / MEASURED, NOT PROMISED

A compact model with visible receipts.

Fresh confirmation results, a control baseline, and measured warm serving receipts from the current research checkpoint.

69 mswarm batch-1 model · live snapshot

235.8/swarm A100 throughput · batch 32

+4.4 ppselected 75.9% vs control 71.5% agreement

96.8%accuracy · 15.4% coverage at confidence ≥ .90

0.160Brier score · 2,000 confirmation cases

Synthetic held-out evaluation. The confidence threshold is post-hoc and not deployment-validated; these are not claims of real-world accuracy or calibration.

0.6B parameters→one forward pass→choice probabilities

WHERE BLITZ FITS

Repeated decisions inside software workflows.

Agent recovery

Detect repeated failures and choose whether to retry, review, or change strategy.

Incident operations

Judge severity, route ownership, and decide when ambiguous evidence needs escalation.

Evaluation pipelines

Score retrieval relevance, context coverage, semantic equivalence, and experiment outcomes at scale.

One state.Many decisions.