Bundle review table

03-nestedness-buffers

README.md

# 03-nestedness-buffers

**Claim.** The robustness of a pollination network to species loss is set by

**Verdict: Not supported**

Robustness is driven by connectance and richness, not NODF: connectance adds dR2=0.533 over log-richness, while NODF adds dR2=0.013 with a NEGATIVE partial correlation (r=-0.256) once richness and connectance are controlled. Richness itself retains a strong partial correlation (r=0.891).

Files: `hypothesis.md` (posted claim + contract), `run.py` (reproducible),
`results.json` (numbers), `figures/` (plots).

hypothesis.md

# Hypothesis 03 — Nestedness Buffers Loss

**Source.** Incoprea Research hypothesis notebook, `frontier-theories-16 #64`
(identifier H318). Reproduced from the posted entry.

**Claim.** The robustness of a pollination network to species loss is set by
its nestedness relative to its connectance, not by species richness.

**Proposed mechanism.** Nested structure means specialist pollinators tend to
interact with a subset of the plants used by generalists, so sequential loss
hits the specialists late and the network resists collapse.

**Predictions.**
- P1 — Nestedness (NODF) and connectance together explain most of the variance
  in robustness.
- P2 — Species richness adds little once connectance and nestedness are in.
- P3 — The NODF effect survives controlling for richness and connectance.

**Refutation condition.** Robustness is essentially fully explained by richness
(or by connectance alone) and NODF adds no independent variance.

## Test contract (pre-specified)

Data: "A database of plant–pollinator networks" (Zenodo 6630184; 87,385
interactions, 435 networks; edge list with `identification_tot`).

- Build binary plant × pollinator matrices; keep networks with
  `identification_tot == "ok"` and ≥10 species per side.
- Per network compute NODF, connectance, and richness.
- Co-extinction simulation (1,000 random removal orders, both directions):
  remove one side at random, track the other side's survival; record R50 and
  robustness AUC (mean of the two directions).
- Variance partition: hierarchical regression of robustness on log-richness,
  then +connectance, then +NODF; report ΔR² and partial correlations.

**Support if** NODF + connectance explain most of the variance (R² > 0.5)
while richness adds ≈ nothing, and NODF survives. **Refute if** richness or
connectance alone explains robustness and NODF adds nothing.

Included code and results

• tests/03-nestedness-buffers/results.json
• tests/03-nestedness-buffers/run.py