Skip to content
TVIJO / RESEARCH ATLAS 2.0.0-candidate

Executor × model × conditions

Choose a testable configuration for a task. Connect projects, acceptance criteria, benchmarks, and a plan for your own evaluation.

3 projects8 evaluation lenses0 own snapshotsCatalog: 2026-11-09

Compare the complete configuration

Model, executor, provider, runtime, effort, and tools are separate parameters. Matching names do not establish matching conditions.

Readiness first, results next

Exact ID → adapter and tool roundtrip → streaming and cancellation → usage and a complete trace → frozen taskset. Ready to test does not yet mean better.

Configurations: 20

Codex × GPT-6.1 Sol · max

Candidate · planned
Executor / harness
Codex
Model — full label
GPT-6.1 Sol
API model ID
gpt-6.1-sol
Provider / route
Unpinned / не закреплён
Reasoning effort
max
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

Proposed engineering baseline. Pin the installed executor and provider revision before running.

Codex × GPT-6.1 Terra · xhigh

Candidate · planned
Executor / harness
Codex
Model — full label
GPT-6.1 Terra
API model ID
gpt-6.1-terra
Provider / route
Unpinned / не закреплён
Reasoning effort
xhigh
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

Proposed independent reviewer. Role is a hypothesis, not a measured superiority claim.

DSH 0.2.2? × DS41F?

Candidate · planned
Executor / harness
DSH Harness
Model — full label
DS41F — alias unresolved
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
0.2.2 desired
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

Requested candidate. Distinct from DeepSeek Harness. Resolve exact executable and provider/model ID.

DSH 0.2.2? × MiniMax M3.1 Flash?

Candidate · planned
Executor / harness
DSH Harness
Model — full label
MiniMax M3.1 Flash — requested alias
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
0.2.2 desired
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

Compatibility and provider route are unverified; do not infer them from the alias.

Aider × DS41F?

Candidate · planned
Executor / harness
Aider
Model — full label
DS41F — unresolved
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

A requested comparison arm. First verify exact model ID, adapter, tool protocol and installed version.

Aider × MiniMax M3.1 Flash?

Candidate · planned
Executor / harness
Aider
Model — full label
MiniMax M3.1 Flash — unresolved
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

A requested comparison arm. First verify exact model ID, adapter, tool protocol and installed version.

OpenCode × DS41F?

Candidate · planned
Executor / harness
OpenCode
Model — full label
DS41F — unresolved
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

A requested comparison arm. First verify exact model ID, adapter, tool protocol and installed version.

OpenCode × MiniMax M3.1 Flash?

Candidate · planned
Executor / harness
OpenCode
Model — full label
MiniMax M3.1 Flash — unresolved
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

A requested comparison arm. First verify exact model ID, adapter, tool protocol and installed version.

Oh My Pi × DS41F?

Candidate · planned
Executor / harness
Oh My Pi
Model — full label
DS41F — unresolved
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

A requested comparison arm. First verify exact model ID, adapter, tool protocol and installed version.

Oh My Pi × MiniMax M3.1 Flash?

Candidate · planned
Executor / harness
Oh My Pi
Model — full label
MiniMax M3.1 Flash — unresolved
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

A requested comparison arm. First verify exact model ID, adapter, tool protocol and installed version.

Pi × DS41F?

Candidate · planned
Executor / harness
Pi (Mario Zechner)
Model — full label
DS41F — unresolved
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

A requested comparison arm. First verify exact model ID, adapter, tool protocol and installed version.

Pi × MiniMax M3.1 Flash?

Candidate · planned
Executor / harness
Pi (Mario Zechner)
Model — full label
MiniMax M3.1 Flash — unresolved
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

A requested comparison arm. First verify exact model ID, adapter, tool protocol and installed version.

Hermes Agent × DS41F?

Candidate · planned
Executor / harness
Hermes Agent
Model — full label
DS41F — unresolved
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

A requested comparison arm. First verify exact model ID, adapter, tool protocol and installed version.

Hermes Agent × MiniMax M3.1 Flash?

Candidate · planned
Executor / harness
Hermes Agent
Model — full label
MiniMax M3.1 Flash — unresolved
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

A requested comparison arm. First verify exact model ID, adapter, tool protocol and installed version.

Claude Code × pinned native model

Candidate · planned
Executor / harness
Claude Code
Model — full label
Model to pin / модель не закреплена
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

No model or compatibility is inferred. Pin the actually accessible native route before comparison.

GLiDE v1.2 × typed decisions

Candidate · planned
Executor / harness
Typed decision adapter · proposed
Model — full label
GLiDE v1.2
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
read-only state · structured choices · probability vector

Research candidate from the decision-model family; exact API, revision and availability are unverified.

Drex Max v1.1 × typed decisions

Candidate · planned
Executor / harness
Typed decision adapter · proposed
Model — full label
Drex Max v1.1
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
read-only state · structured choices · probability vector

Research candidate from the decision-model family; exact API, revision and availability are unverified.

Mammoth 27B × typed decisions

Candidate · planned
Executor / harness
Typed decision adapter · proposed
Model — full label
Mammoth 27B
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
unverified
Harness version
unverified
Runtime version — separate axis
Unverified
Tools / skills
read-only state · structured choices · probability vector

Research candidate from the decision-model family; exact API, revision and availability are unverified.

Rules / solver baseline

Candidate · planned
Executor / harness
Deterministic baseline
Model — full label
No LLM / без LLM
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
not applicable
Harness version
to pin
Runtime version — separate axis
Unverified
Tools / skills
deterministic rules · numerical oracle

A necessary comparator for narrow decisions and numeric tasks. No performance claim until tested.

Router → worker → independent review

Candidate · planned
Executor / harness
Tvijo multi-model protocol · proposed
Model — full label
Typed router → DSH × DS41F → Codex × GPT-6.1 Terra
API model ID
Exact ID unresolved
Provider / route
Unpinned / не закреплён
Reasoning effort
role-specific; unverified
Harness version
protocol proposal
Runtime version — separate axis
Unverified
Tools / skills
isolated files · restricted shell · independent tests

Evaluate the complete pipeline against single-model and deterministic baselines. Include every cost and failure.

Documented executor capabilities

Aider

Coding assistant with configurable models, edit formats and an architect/editor split.

  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.
  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.

OpenCode

Coding executor with explicit agent/model configuration and structured session output.

  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.
  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.

DSH Harness

Requested local DSH candidate. Build, installed versions and exact model routes need attestation.

Version: 0.2.2 · desired

  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.
  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.
  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.

Hermes Agent

Agent with persistent sessions, configurable terminal backends and supporting models.

  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.
  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.
  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.

Codex

Coding executor with structured run events and separate model/provider/effort settings.

  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.
  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.
  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.

Claude Code

Coding executor with structured headless output, subagents and usage instrumentation.

  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.
  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.
  • Capture full events, task acceptance, exact settings and reconciled usage. Installed-version behavior must be checked.

Tvijo multi-model protocol · proposed

A proposed router–worker–verifier configuration with independent acceptance.

  • Count all models, shared tools and retries. Correlated errors invalidate independent-voting assumptions.
Evidence status matters.

A documented feature, a test candidate, and a measured result are different states. Session cost does not measure success, and coding-agent performance does not establish trading profitability.