Research / Consolidation
The research archive is intentionally large. The canonical layer is intentionally small. This protocol keeps both true: every document remains traceable, but only the work that survives evidence, comparison, and release gates becomes public positioning.
Model
01
Everything found in the workspace: notes, papers, PDFs, reports, handoffs, failed plans, and partial drafts.
Admission
A document enters when it contains enough research signal to be searchable, even if it is redundant or wrong.
Exit
It stays as historical material, but it must be promoted, merged, archived, or rejected before being shown as canonical work.
02
Candidate material being reconstructed into paper skeletons, experiment packets, or implementation dossiers.
Admission
A corpus item enters when it intersects with a canonical shelf, carries a concrete method, or contains evidence that can be replayed.
Exit
It exits when a curator assigns a decision: canonical, merge, archive, reject, or experiment.
03
The small public set of research lanes that explain your expertise without overwhelming the reader.
Admission
A work enters only when it has a stable claim, source anchors, related works, proof state, and a named release gate.
Exit
It remains editable until the release dossier freezes the PDF, references, figures, and evidence manifest.
04
The bridge between absorption and proof: a concrete comparison against the current internal system.
Admission
Every ABSORB or TEST verdict must become an experiment packet before it changes a canonical claim.
Exit
It exits as supports, challenges, replaces, no-effect, or inconclusive, with commands and evidence attached.
05
The frozen package for a paper that is ready to submit or cite.
Admission
A paper enters only after source, PDF, figures, references, evidence, limitations, and reproduction commands are aligned.
Exit
It exits as released, withdrawn, or superseded. The dossier remains public as the record.
This is part of the current public theory or system map.
Use when
The work has a stable claim, mapped sources, evidence checkpoints, and a clear relationship to one shelf.
Public treatment
Shown on the main research page and linked to a reader, PDF, schema, and release gate.
This contains useful material, but it should not survive as a separate paper.
Use when
The idea duplicates another manuscript, is a section rather than a paper, or was superseded by a stronger formulation.
Public treatment
Hidden from the front door but linked as source material inside the canonical paper workspace.
This is historically useful but no longer represents the current system.
Use when
The implementation changed, the claim became stale, or the document explains how the architecture evolved.
Public treatment
Kept in the corpus with a historical label. It should not appear as current expertise.
This should not be used as evidence or public positioning.
Use when
The implementation failed, the metric was invalid, the claim was wrong, or the data cannot support it.
Public treatment
Retained only as an internal negative result or failure note, never as a paper card.
This is not a paper yet. It is a question with a proposed test.
Use when
An absorbed external paper or internal idea could change a claim, but needs a head-to-head comparison first.
Public treatment
Shown as an experiment packet only after the baseline, command, evidence, and outcome fields exist.
Canonical shelves
N'Ko, Manding speech, script-native ASR, phonemic evaluation, acoustic governance
Strongest body of work. Offline script-native ASR and paper set exist; stable live iPhone recognition remains the boundary.
Freeze the canonical paper set into flagship, script invisibility, script-native ASR, phonemic evaluation, FAC, and governed deployment lanes.
Graph Kernel, admissible context, trajectory reward, typed skills, autonomous agent accounting
Multiple running systems and paper drafts exist. The remaining consolidation task is to align Graph Kernel, KARL, TML, and typed skills into one provenance program.
Turn the agent papers into one chain: admissible context, trajectory accounting, reward selection, typed composition, and self-improvement.
Anticipation Geometry, computational choreography, multimodal sensor fusion, motion generation, Lume
Architecture and implementation evidence are substantial; release claims need tighter evaluation snapshots and physical run references.
Separate the theory paper, the Lume production architecture, and the physical-capture evaluation papers so MotionMix evidence maps cleanly into the research.
Daily paper ingestion, paper schemas, public research archive, experiment routing, canonical consolidation
The public archive, PDF archive, corpus, schema page, and absorption log exist. Automation health now needs to be visible and the backlog needs Codex-run experiment packets.
Convert every ABSORB or TEST verdict into a named experiment packet mapped to one canonical shelf and one current internal baseline.
Release
01
The central claim is one sentence and names what would falsify it.
02
The canonical LaTeX or Markdown source is known and the rendered PDF is generated from it.
03
Every result cites a run ID, dataset, packet, commit, report, or external paper reference.
04
External baselines and absorbed papers are mapped to specific sections, not dumped into related work.
05
Failed implementations and invalid paths are named so future agents do not resurrect them as fresh ideas.
06
The PDF, HTML reader, tables, algorithms, figures, and references render correctly on the public site.