Gemini's project-wide key-restriction 403 is cleared (key restricted to the
Generative Language API), so Gemini rejoins the grid. Full 2×4 ranking
(pipeline/debate × Claude/GLM-4.7/Gemini/heterogeneous), judge-ranked 1-8:
D2 > P2 > P3 > P1 > D3 > PG > DG > D1.
Adds a third finding: model strength is task-dependent — Gemini-2.5-flash
placed 2nd on the §7.3 concision brief but 6th-7th here on reasoning
("generic list", "idea repeated"): concise-but-shallow. Reinforces that the
optimal config is a joint choice over {topology × per-role model × task}.
Gemini cells ran via the durable async run path (enqueue → worker → poll).
Co-Authored-By: Claude Opus 4.8 <[email protected]>
The GEMINI_API_KEY was restricted (fixing the prior 403 enforcement), so
Gemini-2.5-flash rejoins the per-role leaderboard. Full ranking:
GLM-4.7 > Gemini-2.5-flash > Claude > Kimi > GLM-5.2. Flagship GLM-5.2 still
last on the concision-weighted brief.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Crosses pipeline x debate with 3 model casts (Claude, GLM-4.7, heterogeneous)
on a fixed decision task, judge-ranked 1-6. Central finding: the topology x
model interaction is real and non-monotone — debate amplifies the strongest
model (GLM-4.7: 2nd->1st) but degrades others (Claude: 3rd->last). The optimal
configuration is a joint choice over {topology x per-role model}, which neither
a topology-only nor model-only study can surface. 21 turns, sequential, within
subscription quota caps. Safety invariant holds across all cells.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Single-role drafter on a fixed concision-weighted brief, one turn per backend,
judge-ranked: GLM-4.7 > Claude > Kimi > GLM-5.2. Notable inversion — the
flagship GLM-5.2 ranked last by over-writing the 25-word brief, underscoring
that the best model is role/task/rubric-dependent. Adds the worker_glm5
(glm-5.2 Opus-class) tier to the example config. Gemini excluded this run
(Google June-19 API-key-restriction 403; infra, not model).
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Adds the Phase C result: five backends (Claude/Kimi/GLM/Gemini/Groq) as
interchangeable role-agents, per-node model selection via node.attrs["agent"],
and a 3-config drafter->refiner sweep judged for quality. Heterogeneous
GLM->Gemini won over the all-Claude baseline — an existence proof of
"which model wins which role in which topology." Notes quota-bounded sweep
scope as future work.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Ran the comparison harness end-to-end against a production model with an
LLM-judge, deployed live: debate wins quality (0.840, ~3x tokens), pipeline is
best value (0.790, cheapest), swarm is strictly dominated. Efficient frontier
{debate, pipeline}; the ranking changes vs the offline proxy — confirming
topology is a measurable quality/cost lever with §5 authority-invariance intact.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Working draft consolidating the platform: thesis, related work (ADAS/Darwin-
Gödel/SwarmAgentic + Autonomous Organizational Evolution), the topology model +
classifier, the safe-execution result (authority is topology-invariant — the §15
contract holds across any topology because the orchestrator only sequences safe
turns), the comparison harness + Pareto + workflow-of-topologies, preliminary
offline results, limitations/future (real-model runs, evolution), and a
reproduction section. Empirical tables are offline/illustrative placeholders.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
