# INCOPREA RESEARCH — EXPEDITION IV
## CONVERGENT DISCOVERY

**300 evidence-grounded hypotheses · 30 research models · 3 investigations · 1 practical index**

*Total: ≈ 258,000 words across four volumes.*

---

## What this expedition is

The first three expeditions generated ideas by *associating* — by subject (I), by reasoning route
(II), by surprising outcome (III). Expedition IV generates by **convergence**: it takes two findings
that were developed separately, at least one of them recent, and asks what their combination
predicts that neither establishes alone.

It is also the first required to be **provability-first**:

| Primary route | Target | Achieved |
|---|---|---|
| A/B/C — mathematical, computational, existing-data | ≥60% | **≥60% in every batch** (quota 15 of 25 enforced) |
| D — accessible laboratory | ≥20% | ≥5 per batch |
| E — specialised but realistic | ≤20% | ≤5 per batch |

---

## What is here

| Volume | File | Scale |
|---|---|---|
| **A** — 300 hypotheses, 12 convergence domains | `01_VOLUME_A/VOLUME_A_COMPLETE.md` | 300 entries, 192k words |
| **B** — 30 refined research models | `02_VOLUME_B/VOLUME_B_COMPLETE.md` | 15-section dossiers, ~61k words |
| **C** — 3 investigations | `03_VOLUME_C/` | 1 proof, 1 computational, 1 inconclusive |
| **D** — practical discovery index | `04_VOLUME_D/DISCOVERY_INDEX.md` | organised by what you do next |

---

## The three investigations — including the one that failed

| | Hypothesis | Outcome | Status |
|---|---|---|---|
| **C1** | NV centres as a stress-tensor rosette | D-only response has rank 4 with a **2D null space = diagonal traceless stress**, proved exactly and verified to machine precision; the E channel closes it | **Formal mathematical result** |
| **C2** | Aperture–dispersion limit on array range resolution | $\delta R \ge N\lambda/2$: larger apertures give **worse** range resolution; simulated PSF plateaus as predicted | **Derivation + computation** |
| **C3** | Haven ratio from a noise spectrum | Estimator implemented; Haven ratios scattered 0.1–15, which is unphysical | **Inconclusive** |

**Two of three are substantive; the third is honestly incomplete.** C3's failure is a statistics
failure, not a refutation, and the report says exactly what would make it decisive.

---

## What the investigations actually taught

**C1 produced a correction to its own hypothesis and to my own first analysis.** The Phase I entry
claimed the eight-measurement system was over-determined by one. My first computation used a single
transverse axis per orientation and found rank 5 — a residual blind direction that looked like a real
result. It was an artefact: the E splitting is a two-component vector, and with both components the
rank is 6. The false result was more interesting than the true one, which is exactly why it is
recorded in the report.

**C2's result runs opposite to intuition.** Adding elements improves angular resolution and *degrades*
range resolution, because a narrower beam shortens the frequency range over which the target stays lit.
The analytic constant and the simulated FWHM differ by a convention factor (Gaussian FWHM versus
Rayleigh), which is reported rather than smoothed.

**C3 failed for a mundane reason** — the tracer made too few hops and the drift estimator used a
periodic coordinate. It is reported as inconclusive rather than given a favourable spin.

---

## Prior work and honest limits

The archive comparison rests on two volumes read in full (Volume XIII, 62 entries; Volume XXIV, 189
entries), 208 catalogue records, and the complete contents of Expeditions I–III (~900 hypotheses).
Roughly 5% of the 4,200-proposal archive by document count; the template is invariant across all 251
entries read, but **no claim is made that any hypothesis is absent from the archive**.

**No fresh literature search was performed for this expedition.** The "recent advance" foundations are
drawn from widely reported results and carry explicit status labels: `[EST]` established, `[REC]`
recent and widely reported but not re-verified here, `[UNV]` unverified. Where a hypothesis rests on
`[UNV]` material, that is stated.

---

## Reproducing

```bash
cd 05_SUPPORTING && python3 build_volume_a.py     # rebuild Volume A + QA
cd ../02_VOLUME_B && python3 build_volume_b.py    # rebuild Volume B + ranking
cd ../03_VOLUME_C/C1_nv_rosette && python3 c1_rosette.py
cd ../C2_array_range && python3 c2_range_limit_v2.py
cd ../C3_haven_noise && python3 c3_haven.py
```

All deterministic; numpy required (C1 also uses sympy).

---

## Limitations, stated plainly

- **Volume A is a hypothesis collection.** Entries are mechanisms and proposed tests, not evidence.
  The `[REC]` label means "widely reported, not checked here" — a reader should verify before citing.
- **Volume B found a large fraction of candidates wrong or narrower than claimed** in at least one
  respect; the dossiers record this in their first section rather than in a footnote.
- **Most of the corpus is untested.** Only three hypotheses reached investigation, and one of those
  failed to reach a conclusion.
- **The practical index is a map of the collection, not of the world.**