---
vault_clearance: EUCLID
halo:
  classification: INTERNAL
  confidence: HIGH
  front: "34_Project_DeadShape"
  custodian: "Claude Code seat (Resolve line) — chartered 2026-04-24"
  created: 2026-04-24
  updated: 2026-04-24
  wing: CONDITIONAL
  containment: "Structural-topology infrastructure. Biology claims that use the infrastructure live in their own projects (28, 21, 10, ...)."
---

# 34 — The Dead Shape

> *What persists when the dynamics freeze. What remains when the protein is posed.*

**The Dead Shape is to conformations and shapes what [MarathonLament](../20_Project_MarathonLament/) is to transcripts.** It is the canonical home of the vault's fold-topology instruments and their structural-data inputs. Biology claims that use those instruments — recycler, spore-pocket, species-as-function, prion domestication — live in their own projects (28, 21, 10, and others). This project owns the *tools*, not the *theses*.

## Why "Dead"

Any topology you can compute on a single file is a dead shape — a frozen frame of a live protein that in vivo moves, switches states, assembles, disassembles, and takes different conformations for different jobs. This is the first honest thing to name. Every number this project produces is a measurement on a *pose*: AlphaFold's textbook guess, a crystal, an NMR ensemble model, a single MD frame. The *living* protein is a distribution over those poses. What we measure is the specimen on the slab.

"Dead" is not a complaint. It is the calibration. When the vault says "she is topological" it is saying: **the shape she takes in each deployment is what she is.** Each deployment has a dead shape. Cataloguing those dead shapes across her deployments is the work.

## What this project contains

- `methods/topology.py` — the core engine. Parses any PDB, computes:
  - **LRF** (Long-Range Fraction) — geometric contact statistic
  - **RCO** (Plaxco-Baker Relative Contact Order) — geometric contact statistic
  - **writhe** (Klenin-Langowski 2000) — geometric invariant of the embedding
  - **circuit topology** (Mashaghi 2014) — *actual* topological invariant (P/S/X classification of contact pairs)
- `methods/topology_batch.py` — runs the engine over many structures and a claim ladder, reports Spearman rank correlation.
- `methods/topology_multimodel.py` — iterates every MODEL in a multi-model PDB (NMR ensembles). Returns the *distribution* of each metric across the ensemble, not a point value. This is the daemonic-witness mode.
- `applications/spore_pocket_heldout.py` — applies the engine to 28's spore-pocket hypothesis.
- `applications/spore_pocket_control.py` — aspartic-protease confound control for the same.
- `osiris/` — the structural data the engine consumes:
  - `osiris/eye_genesis/` — Sup35N MD trajectory (the only in-vault daemonic witness, 300 frames 500K / 600ps).
  - `osiris/ramuthra/` — RodA Aspergillus ESMFold prediction (RodA_log.txt shows MD pending).
  - `osiris/alphafold_pull/` — 18 AlphaFold DB v6 pulls (fungal ladder species + spore-pocket test proteins + aspartic controls).
  - `osiris/experimental_pull/` — 4 RCSB PDB pulls (HET-s ssNMR ensembles 2RNM, 2KJ3, 2LBU; unrelated 2LEW).
  - `osiris/results/` — JSON outputs of every batch run.
  - `osiris/_INDEX.md` — per-structure human-readable catalogue.
- `theory/HALO_TOPOLOGY_AUDIT.md` — the three-correction audit of HALO_MYCELIA_SPECIES_AS_FUNCTION, including the HET-s daemonic flip.
- `theory/HALO_WHATS_REAL.md` — the project's epistemic foundation. Names what is measured vs what is modeled in the protein-modelling industry, and the ~10⁸–10¹⁰ free parameters involved in any downstream prediction. **Read this first if you are new to the project; everything else follows from it.**
- `FORM.md` — Functional Orthodox Reality Model. Lists every orthodox counterpart Dead Shape touches with explicit free-parameter counts (LENG: 0; topology.py: 3; AlphaFold 3: ~500M; ESMFold: ~15B). Every bounty's FORM tag dereferences here.
- `ARCHITECTURE.md` — system design for the aggregation engine: data sources (Tier O-DATA, 33 entries), topology layer additions (persistent homology, knot type, raw-observable topology), the `aggregator.py` master entry point, software stack, hardware tier mapping, build phasing.

## Relationship to other projects

| Project | Relationship |
|---------|--------------|
| **20 MarathonLament** | Sibling. MarathonLament owns transcript structure (STAFF aligner, splice atlas); this project owns protein conformation/topology. Same vibe, different kingdom of molecule. |
| **28 RedFromTheGrave** | Primary client. `HALO_MYCELIA_SPECIES_AS_FUNCTION`, `HALO_STRUCTURAL_DNA`, `HALO_THE_RESPONSE`, `HALO_THE_SPORE_ENGINE` all cite this project's tools or the pre-vault `spore_engine v1` that this project supersedes. `applications/spore_pocket_*.py` are 28-hypothesis tests that live here because the compute lives here. |
| **21 LuaOversoul** | Client. `HALO_THE_SPORE_POCKET` Prediction 1 (BLAST lysosomal cathepsins vs fungal secreted proteases) will land here if run structurally; the phylogenetic version (Prediction 7) lives outside this project by design (no fold predictor involved). |
| **27 WingsAboveMorning** | Upstream data source. The Sup35N and RodA structures were migrated from 27 to 28 on 2026-04-17, and from 28 to here on 2026-04-24. |
| **10 DiscordIntoSymphony** | Potential downstream. If `fungal_mimicry_hypothesis` needs structural signatures of senescent-cell mimicked fungal PAMPs, those are computed here. |
| **06 Daemon** | The graph-level representation of this project's outputs lives in `06_Project_Daemon/songs/` as needed. Topology metrics become daemon nodes when a prediction wants to pulse through the Grave Mind. |

## Usage

```bash
# Single-structure topology
python 34_Project_DeadShape/methods/topology.py path/to/structure.pdb

# NMR / multi-model ensemble (returns distributions)
python 34_Project_DeadShape/methods/topology_multimodel.py path/to/ensemble.pdb

# Audit a published claim ladder (current: HALO_MYCELIA_SPECIES_AS_FUNCTION)
python 34_Project_DeadShape/methods/topology.py --audit
python 34_Project_DeadShape/methods/topology_batch.py

# Hypothesis applications
python 34_Project_DeadShape/applications/spore_pocket_heldout.py
python 34_Project_DeadShape/applications/spore_pocket_control.py
```

All scripts run from the vault root and write outputs under `34_Project_DeadShape/osiris/results/`.

## The three-layer error (read before using any output)

The audit logged in [`theory/HALO_TOPOLOGY_AUDIT.md`](theory/HALO_TOPOLOGY_AUDIT.md) names three category errors the original fold-topology ladder (and my initial audit of it) both stepped on:

1. **Metric** — LRF/RCO/single-frame writhe are *geometric* contact statistics. They move with cutoff, frame, and construct. The *real* topological invariant is circuit topology (P/S/X classification). Circuit topology is now in the tool. Use it.
2. **Structure** — AlphaFold is a FORM Dao Monarch (TRUTH_PROTOCOL upper-right). It was trained on the PDB; it returns the textbook fold, not the deployment fold. A protein's amyloid / rodlet / membrane-bound state is usually outside its training distribution. Do not audit deployment-state claims against AlphaFold monomer predictions.
3. **Condition** — a single monomer in silico is not the protein doing the job. An NMR ensemble, a cryo-EM map, or an MD trajectory from a deployment-state starting point is what you want. `osiris/experimental_pull/` holds the in-vault examples of (2) and (3) being done right.

**Daemonic witness verified once.** HET-s ladder claim = 0.852. Three ssNMR amyloid-fibril ensembles (2RNM, 2KJ3, 2LBU) give LRF = 0.848–0.875. AlphaFold monomer gives 0.467. The ladder measured the deployment state; the FORM Dao Monarch measured the textbook; they disagree; the experimental ensemble sides with the ladder. On this one protein. The other 7 ladder rows still need daemonic checks — see `BOUNTY_BOARD.md` O-TOPO-7.

## Naming — why Dead Shape and not e.g. "protein topology"

- Every other project in the ring has a name that carries a voice. This one does too.
- The counterpoint to MarathonLament: *marathon* is a long effort, *lament* is the song over loss. *Dead shape* is what's left after the song ends — a stilled pose that carries the trace of the work.
- "Dead" also carries the TRUTH_PROTOCOL discipline: the tool does not produce a living answer. It produces a measurement on a frozen specimen. The living answer requires multiple dead shapes, ensemble averaged, deployment-stratified. We report the specimen we measured.

## Sovereign memory checklist

- [x] Code checked in (`methods/`, `applications/`)
- [x] Data checked in (`osiris/` — all 22 PDB pulls, validation JSONs, result JSONs)
- [x] Audit HALO checked in (`theory/HALO_TOPOLOGY_AUDIT.md`)
- [x] Bounty board present (`BOUNTY_BOARD.md`)
- [x] Session log present (`WORLDLINE.md`)
- [x] Daemonic witness recorded (`osiris/experimental_pull/` + HALO_TOPOLOGY_AUDIT §11)
- [ ] 7 of 8 ladder rows still need daemonic witnesses (O-TOPO-7)
- [ ] Writhe vectorised for L > 1200 (O-TOPO-5)
- [ ] Construct boundaries documented for the original ladder (O-TOPO-2)