Add Fugu orchestration evaluation and design rationale#8
Open
jhave wants to merge 2 commits into
Open
Conversation
Evaluate Sakana Fugu's orchestration-as-a-model against Narracode's harness. Maps Fugu's Thinker/Worker/Verifier triad and swappable multi-provider routing onto Narracode's existing AUTO_MODE model assignment, and proposes a "local Fugu" dispatcher that routes each pass to Claude/Gemini/GPT via API. Critically refuses Fugu's output synthesis and learned-reward coordination as median-collapse hazards, following the house "what transfers / what doesn't" pattern. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Mxqax7CCW9RZ59pHEshZvq
…rracode Standalone Glia-styled page (Narracode logo + Glia footer/bio/funding/ license) arguing against bolting Fugu-style orchestration onto the harness: routing tables rot as models churn; model choice is an irreducibly subjective aesthetic judgment, not a reward function; orchestration adds complexity and worsens context rot; and platforms already enhance prompts toward the median the harness exists to refuse. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Mxqax7CCW9RZ59pHEshZvq
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This PR adds two new documents to the Narracode project: a detailed technical evaluation of Sakana AI's Fugu orchestration model and an accompanying essay explaining why Fugu-style enhancements would not improve Narracode's literary harness.
Changes
plans/2026-06-28_fugu-orchestration-evaluation-and-routing-proposal.md: A comprehensive technical proposal evaluating Fugu's applicability to Narracode. The document:narracode.mdto formalize routing as a first-class, auditable decision2026-06-28_fugu-orchestration-consideration.html: A public-facing essay articulating the design philosophy behind the rejection. The essay:Notable Details
The evaluation is grounded in the principle that Fugu optimizes against verifiable objectives (SWE-Bench, GPQA) while Narracode exists precisely because prose that scores highest on automatic metrics is the failure mode. The proposal preserves Narracode's core commitment: keeping the human prompter's aesthetic judgment as the irreducible decision point, while making routing transparent and auditable rather than hidden or learned.
The HTML essay is styled for publication on glia.ca and includes proper attribution, funding acknowledgments, and Creative Commons licensing.
https://claude.ai/code/session_01Mxqax7CCW9RZ59pHEshZvq