Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
72 commits
Select commit Hold shift + click to select a range
53f5ff3
chore(develop): open standing develop -> master PR after v0.10.1 release
martyn Nov 17, 2026
1b7c4da
feat(rgdt): solve GAN scale-mass gauge & horizon starvation (Track 3)
martyn Nov 18, 2026
1a69a47
fix(rgdt): preserve SHA-pinned recipe_compat, summaries-only compile,…
martyn Nov 18, 2026
316f15a
feat(t2-critic): pure 1st-order spatial-frame forward-tangent & secan…
martyn Nov 18, 2026
591af72
feat(prior): add Theory 1A Gersho-Laguerre pullback prior dynamics an…
martyn Nov 18, 2026
7dbe73e
fix(forge): omit standalone t1-prior study family from trainer-famili…
martyn Nov 18, 2026
bc4a77a
feat(t1-coupling): Track 1B deconflicted piecewise chart coupling and…
martyn Nov 18, 2026
821adbe
fix(t2-critic): remove F401 unused imports and pin CPU thread count i…
martyn Nov 18, 2026
6bbab90
fix(tests,forge): register neutral t1_prior Recipe additions across C…
martyn Nov 18, 2026
679e25b
fix(rgdt): resolve lint F401 imports & Recipe schema-pinning CI tests
martyn Nov 18, 2026
7dbd2e3
test(t1-coupling): register t1c Recipe fields in schema-pinning tests…
martyn Nov 18, 2026
271e363
fix(tests): update bcap compatibility adapter checked inactive count …
martyn Nov 18, 2026
31ff823
Merge pull request #566 from HyperGAN/feat/solve-gans-t2-efficient-cr…
martyn Nov 18, 2026
aa3fb49
Merge origin/develop (#566 t2_secant) into feat/solve-gans-t1-prior-t…
martyn Nov 18, 2026
25474f2
Merge pull request #567 from HyperGAN/feat/solve-gans-t1-prior-topology
martyn Nov 18, 2026
8ed13b6
Merge origin/develop (#566, #567) into feat/solve-gans-t1-deconflicte…
martyn Nov 18, 2026
d976bb4
Merge pull request #568 from HyperGAN/feat/solve-gans-t1-deconflicted…
martyn Nov 18, 2026
7204998
Merge origin/develop (#566, #567, #568) and refresh diagnostic view p…
martyn Nov 18, 2026
d705c9c
Merge pull request #565 from HyperGAN/feat/solve-gans-t3-missing-piece
martyn Nov 18, 2026
4636d77
feat(t4_arch): SVD-free depth-equilibrated trust optimizer & fwAD sat…
martyn Nov 18, 2026
d6b3355
feat(t3): add Theory-3 causal isolation benchmark, factorial ablation…
martyn Nov 18, 2026
034d552
feat(t4-nash): Theory 4A SVD-free function-space impedance-matched Na…
martyn Nov 18, 2026
3f55ca9
fix(t4-nash): register rev-18 stretch tasks, align prose equations, a…
martyn Nov 18, 2026
e1350f1
feat(unified): synthesize Round-1 critic, pullback prior, chart & fib…
martyn Nov 18, 2026
e6dd89b
test(forge): update rev-18 view chain and budget pins for CI shard 2
martyn Nov 18, 2026
4d2ebb6
docs(t3): conclude 12 RGDT causal isolation & ablation readouts and p…
martyn Nov 18, 2026
517e506
Merge pull request #573 from HyperGAN/feat/solve-gans-unified-synthesis
martyn Nov 18, 2026
6fbfa3e
chore: merge develop (#573 unified) into feat/solve-gans-t4-imbalance…
martyn Nov 18, 2026
e622446
feat(t4_arch): conclude t4-arch-v3 studies, readouts, and Theory 4B s…
martyn Nov 18, 2026
c6be0a2
Merge origin/develop (#573) and refresh diagnostic view publication
martyn Nov 18, 2026
f5cc481
chore(forge): refresh catalog and compiled summaries on committed mer…
martyn Nov 18, 2026
8ba6e5a
ci: match test_public_package_imports wheel size cap (1.2 MB) to work…
martyn Nov 18, 2026
9a8221b
chore(t3): merge origin/develop (#573) and refresh Forge catalog & su…
martyn Nov 18, 2026
1ea5e02
Merge pull request #574 from HyperGAN/feat/solve-gans-t3-causal-proof
martyn Nov 18, 2026
0d23319
chore(t4-nash): merge origin/develop (#574) and reconcile discriminat…
martyn Nov 18, 2026
c60383f
Merge pull request #571 from HyperGAN/feat/solve-gans-t4-imbalance-nash
martyn Nov 18, 2026
191772e
chore(t4_arch): merge origin/develop (#574, #571) and refresh Forge c…
martyn Nov 18, 2026
ff9b14b
Merge pull request #575 from HyperGAN/feat/solve-gans-t4-strange-arch…
martyn Nov 18, 2026
895aa88
study(r3-calibration-holdout): conclude 12 Tier-3/Tier-4 holdout read…
martyn Nov 19, 2026
4eab322
Merge pull request #576 from HyperGAN/feat/solve-gans-r3-calibration-…
martyn Nov 19, 2026
8c3752c
feat(nash-v3): pure-elementwise 0-SVD/0-autograd.grad 3/3 imbalance N…
martyn Nov 19, 2026
634d38d
Merge pull request #577 from HyperGAN/feat/solve-gans-r3-nash-complete
martyn Nov 19, 2026
920c71f
wip(cond-fiber): cross-batch conditional fiber operator bank (v1.3 fr…
martyn Nov 19, 2026
a32872a
Merge branch 'develop' of https://github.com/HyperGAN/ParticleGAN int…
martyn Nov 19, 2026
92de0b1
feat(cond-fiber): solve conditional DDGAN chains (first-ever Gaussian…
martyn Nov 19, 2026
0e58974
docs(cond-fiber): align Section 3.2 Tier-1/Tier-2 task listing verbat…
martyn Nov 19, 2026
8503866
Merge pull request #578 from HyperGAN/feat/solve-gans-r3-conditional-…
martyn Nov 19, 2026
6c4aa73
wip(unified-v4): complete 4-Law zero-SVD zero-autograd.grad unificati…
martyn Nov 19, 2026
38849c9
Merge remote-tracking branch 'origin/develop' into feat/solve-gans-r3…
martyn Nov 19, 2026
f4ce027
docs(forge): fix 3 prose/table accuracy items in r3-unified-v4 study …
martyn Nov 19, 2026
3f8536c
test(recipes): include unified_v4 in prior_regularizer rejection regex
martyn Nov 19, 2026
c400c6d
Merge pull request #579 from HyperGAN/feat/solve-gans-r3-unified-v4
martyn Nov 19, 2026
2ccdfb5
feat(minimal): collapse 4-law core into 899-LOC zero-NS minimal prese…
martyn Nov 20, 2026
c176c23
feat(forge): Round-4 Track 2 Occam's Razor systematic 8-arm ablation …
martyn Nov 20, 2026
91d5102
fix(minimal): tighten checkpoint tamper assertions & align report met…
martyn Nov 20, 2026
a90b955
Merge pull request #581 from HyperGAN/feat/solve-gans-r4-ablation-razor
martyn Nov 20, 2026
b485421
Merge branch 'develop' of https://github.com/HyperGAN/ParticleGAN int…
martyn Nov 20, 2026
6aaee00
chore(forge): refresh catalog blob pins & compilation after develop m…
martyn Nov 20, 2026
c644b9b
Merge pull request #580 from HyperGAN/feat/solve-gans-r4-minimal-core
martyn Nov 20, 2026
8caab7b
feat(unified_v11): add zero-SVD/zero-eigh/zero-pinv/zero-autograd.gra…
martyn Nov 20, 2026
37985b7
Merge branch 'develop' of https://github.com/HyperGAN/ParticleGAN int…
martyn Nov 20, 2026
c3a69c5
docs(r4-solution-complete): add verbatim 9-receipt study report and f…
martyn Nov 20, 2026
7954192
chore(forge): update catalog, EXPERIMENT_MEMORY, and compilation for …
martyn Nov 20, 2026
4fb04e8
docs(r4-solution-complete): correct paired-context FIFO ring capacity…
martyn Nov 20, 2026
20f908e
Merge pull request #582 from HyperGAN/feat/solve-gans-r4-solution-com…
martyn Nov 20, 2026
7932e5e
Merge remote-tracking branch 'origin/master' into feat/release-unifie…
martyn Nov 20, 2026
d438ade
feat(unified_v11): register unified & minimal families and set DEFAUL…
martyn Nov 20, 2026
f01148c
release(0.11.0): promote unified_v11 default, publish technique inven…
martyn Nov 20, 2026
1ffd3f9
test(forge): update selected family count in test_forge_scaling (10 -…
martyn Nov 20, 2026
2021500
docs(readme): publish 0.11.0 unified_v11 MoGParticlePrior 100-Gaussia…
martyn Nov 20, 2026
b4b44ef
docs(forge): update history-pin-edits current_blob for reports/readme…
martyn Nov 20, 2026
a40767e
Merge pull request #583 from HyperGAN/feat/release-unified-v11-0110
martyn Nov 20, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
9 changes: 5 additions & 4 deletions .github/workflows/tests.yml
Original file line number Diff line number Diff line change
Expand Up @@ -134,9 +134,10 @@ jobs:
assert MoGParticlePrior(sigma=0.1).sample(2)[0].shape == (2, 4)
assert DDGAN().steps == 4
default = get_recipe()
assert DEFAULT_RECIPE == "empf"
assert default.optimizer_dense_profile == "newton_schulz_hybrid_v1"
assert default.optimizer_svd_backend == "batched"
assert DEFAULT_RECIPE == "unified_v11"
assert default.optimizer_family == "unified_v11_equil"
assert default.critic_formulation == "unified_v11"
assert default.prior_regularizer == "unified_v11"
assert default.training_precision == "bf16_channels_last"
assert default == get_recipe(DEFAULT_RECIPE)
assert Recipe(**get_recipe_fields()) == default
Expand Down Expand Up @@ -276,7 +277,7 @@ jobs:
import tarfile
import zipfile

max_bytes = 600_000
max_bytes = 1_200_000
for wheel in Path('dist').glob('*.whl'):
assert wheel.stat().st_size < max_bytes, wheel.stat().st_size
with zipfile.ZipFile(wheel) as archive:
Expand Down
Binary file modified 100gaussians.gif
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
50 changes: 47 additions & 3 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
@@ -1,8 +1,52 @@
# Changelog

## Unreleased

- Publish a GitHub source release at the tested commit after PyPI publishes the new version from `master`.
## 0.11.0 — 2026-11-20

- **Default Unified 4-Law research selection (`DEFAULT_RECIPE = "unified_v11"`).**
Promote `unified_v11` (matching [`unified-v11`](configs/forge/ideas/unified-v11.json),
[`particlegan/unified_v11.py`](particlegan/unified_v11.py)) as the public default
trainer preset while preserving explicit `get_recipe("minimal")`, `get_recipe("empf")`
(`empf-v8`), `get_recipe("rhbi_optimized")` (`rhbi-v12`), `get_recipe("rhbi")`
(`rhbi-v1`), `get_recipe("bispf_optimized")`, and `Recipe.from_dict(saved_fields)`
historical restoration (`particlegan/recipe_compat.py` untouched). `unified_v11`
runs with strictly **`0` `torch.autograd.grad`, `0` `svd`/`svdvals`, `0` `eigh`/`eigvalsh`,
and `0` `pinv`** across every forward, backward, penalty, transport, conditional fiber
bank, prior, and optimizer path, uniting all four GAN laws with fused `Z=[X,Y]` Quintic
Tikhonov polar equilibration, zero-`autograd.grad`/zero-`pinv` analytic `direction_blend`,
zero-`eigh` cyclic-Jacobi cross-batch conditional fiber operator matching, and
scale-invariant `k=4` density-ratio orphan macro-transport. Across `9` concluded Forge
studies on `NVIDIA RTX A6000` (seed `0`, see
[`reports/forge/studies/r4-solution-complete.md`](reports/forge/studies/r4-solution-complete.md)
and the [technique inventory](reports/forge/technique-inventory.md)), `unified-v11`
leads the canonical leaderboard at **`95/115` counted view requirements** (`27/27`
Tier 1, `68/78` Tier 2), **`40/69` full-suite passes** (`6` FAIL, `11` not measured,
`12` blocked), **`50/68` stretch view assignments**, and **`41/47` Tiers 1–2
`discriminator_stability` PASS** (`55/62` deduplicated unique runnable tasks PASS;
`90.24 ms/step` median and `1,372.60 MiB` allocated VRAM on matched 66-cell
`dcgan_performance_r256`, `89.19 ms/step` in `readout-1e5fb235`), including `11/11`
required Tier 1, `3/3` precision (`fp32`/`mxfp8`/`nvfp4`),
`model_ddgan_gaussian_chain`, `3/3` extreme Nash imbalance, `2/2` causal isolation, `5/5`
shift/rail recovery, `3/3` `7.5k` ring endurance, `6/7` 16-layer strange architectures,
`15/17` Tier-3 `14k` holdout (`grid100_14k` and `rotated100_14k`), and `3/4` runnable
Tier-4 tasks — ahead of `empf-v8` (`83/115`, `23/69`, `129.72 ms/step`) and `rhbi-v12`
(`83/115`, `33/69`, `590.59 ms/step`). Forge screening calibration remains provisional
(`BLOCKED` under
`criteria-v1.json`; see [`reports/forge/studies/r3-calibration-holdout.md`](reports/forge/studies/r3-calibration-holdout.md)),
and remaining failures (`staggered100` / `staggered100_14k`, `grid100_res16_tanh`,
`five_word_joint_hold`, `model_ddgan_bimodal_chain`, `model_ae_ddgan_gaussian_chain`,
`interp_img_patch_shift`, `interp_saddle_barycentric`, plus `12` unimplemented Tier-4
stretch hosts) stay explicitly disclosed.
- **Minimal-Principle `<900`-LOC fast core (`preset="minimal"`, `particlegan/minimal.py`).**
Add the self-contained `899`-LOC, zero-sibling-import core (`minimal-v2`) with zero
`autograd.grad`, zero SVD/`eigh`/`pinv`, and zero Newton-Schulz (`55.50 ms/step` and
`1344.84 MiB` peak allocated VRAM on `dcgan_performance_r256`, `6/6` PASS; see
[`reports/forge/studies/r4-minimal-core.md`](reports/forge/studies/r4-minimal-core.md)).
- **The Four Laws of Stable, Energy-Efficient GAN Convergence ([`docs/four-laws.md`](docs/four-laws.md)) & Occam's Razor causal ablation (Rounds 1–4).**
Document the four laws proved across Rounds 1–4 — Law 1 (Measure & Prior Topology: `t1_prior`, `t1_coupling`), Law 2 (First-Order Zero-`autograd.grad` Critic Geometry: `t2_secant`), Law 3 (Scale–Mass–Fiber Gauge Identifiability: `rgdt`, `cond_fiber`), and Law 4 (Function-Space Two-Player Equilibration & Depth/Rail Trust: `t4_nash`, `t4_nash_v3`, `t4_arch`) — together with unified synthesis (`unified_v4`..`unified_v11`), bounded Tier-3/Tier-4 calibration holdout audits (`r3-calibration-holdout`), and Round-4 Track 2's 8-arm Occam's Razor causal knockout study (`r4-occam-ablation-razor`, showing the four laws collapse into Two Mathematical Halves with depth-gated 2D Newton-Schulz strictly on `L > 5` skip-free MLPs while identifying redundant epicycles).
- **Automated release pipeline from `master`.**
Publish new versions automatically to PyPI via Trusted Publishing on `master` merges and
publish a verified GitHub source release at the tested commit after PyPI succeeds
(`.github/workflows/release.yml`, `scripts/publish_github_source.py`).

## 0.10.1 — 2026-11-16

Expand Down
65 changes: 40 additions & 25 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,7 @@ critic penalty and prior regularizer.
Requires Python 3.11+. PyTorch and NumPy are installed as dependencies.

```bash
python -m pip install particlegan # the library (0.10.1)
python -m pip install particlegan # the library (0.11.0)
```

To run the examples, experiments and tests, install from source with the extras you need:
Expand Down Expand Up @@ -67,17 +67,22 @@ for _ in range(recipe.total_steps):
samples = trainer.sample(1024)
```

`get_recipe()` returns the default recipe, **EMPF** (*Exponential Moving Prototype
Flow*, `DEFAULT_RECIPE = "empf"` in [particlegan/recipes.py](particlegan/recipes.py)).
It uses a clock-free online *e-means* sufficient-statistics prototype bank,
mass-weighted prototype-to-fake Sinkhorn OT, a state-driven settling Born-Infeld
critic penalty (`EMPFCriticPenalty`, `ebi_beta = 6.0`), a DoF-aware EMPF prior
regularizer (`prior_reg = 0.01`), and zero-momentum DualNorm
(`newton_schulz_hybrid_v1`, `batched` SVD, `bf16_channels_last`) at constant rates.
Explicit `get_recipe("rhbi_optimized")`, `get_recipe("rhbi")` and
`get_recipe("bispf_optimized")` retain named presets. The acronyms are expanded in the
[research glossary](RESEARCH_GLOSSARY.md#acronyms-and-short-names), and the
[recipe settings and evidence](docs/api.md#recipes-and-defaults) explain the configuration.
`get_recipe()` returns the default recipe, **Unified** (`DEFAULT_RECIPE = "unified_v11"` in
[particlegan/recipes.py](particlegan/recipes.py), matching `unified-v11`).
It unites all four GAN laws with strictly **zero `torch.autograd.grad`, zero SVD, zero `eigh`,
and zero `pinv`**: Law 1 streaming Laguerre-Pythagorean prototype prior & orthogonal pullback
isometry (`prior_reg = 0.02`), Law 2 first-order Born-Infeld secant critic with zero-mean
logit gauge (`ebi_beta = 12.0`), Law 3 scale-invariant `k=4` density-ratio orphan
macro-transport & zero-`eigh` cross-batch conditional fiber operator bank, and Law 4 fused
`Z=[X,Y]` Quintic Tikhonov polar equilibration with zero-`pinv` analytic `direction_blend`
and `bf16_channels_last` conv training at constant rates.
Explicit `get_recipe("minimal")` (`<900`-LOC zero-Newton-Schulz fast core),
`get_recipe("empf")` (`empf-v8`), `get_recipe("rhbi_optimized")` (`rhbi-v12`),
`get_recipe("rhbi")` and `get_recipe("bispf_optimized")` retain named presets. See
**[The Four Laws of Stable, Energy-Efficient GAN Convergence](docs/four-laws.md)** for the
first-principles derivation and Occam's Razor ablation behind `unified_v11` and `minimal`, the
[research glossary](RESEARCH_GLOSSARY.md#acronyms-and-short-names) for acronyms, and the
[recipe settings and evidence](docs/api.md#recipes-and-defaults) for configuration details.

Without `prior=`, the trainer builds a randomly drawn particle table. Networks
and device placement belong to the caller. The trainer does not reinitialize
Expand Down Expand Up @@ -109,14 +114,23 @@ metrics, receipts and training GIFs. Other report pages are historical records.
What the evidence does and does not show today:

- **The default is a research selection, not a qualified winner.** In the
[technique inventory](reports/forge/technique-inventory.md), the selected EMPF recipe
(`empf-v8`, `DEFAULT_RECIPE = "empf"`) ties `rhbi_optimized` (`rhbi-v12`) at **83/115**
counted view requirements (`27/27` Tier 1 PASS, `56/78` Tier 2, `0/10` Tier 3) with
**23 PASS**, **17 FAIL**, 8 not measured and 16 blocked across the 64-test fixed suite
(ahead of `bispf_optimized` at `81/115` and `18/64`; explicit `rhbi_optimized` reaches
`33/64` PASS and `35/63` stretch view assignments on the Forge suite). Forge's screening
calibration (`BLOCKED`), Tier 2/3 full-tier gates and untouched holdout confirmation
remain open.
[technique inventory](reports/forge/technique-inventory.md) and
[Round-4 solution-complete study](reports/forge/studies/r4-solution-complete.md),
the selected Unified recipe (`unified-v11`, `DEFAULT_RECIPE = "unified_v11"`) leads
screening at **`95/115` counted view requirements** (`27/27` Tier 1, `68/78` Tier 2),
**`40/69` full-suite passes** (`6` FAIL, `11` not measured, `12` blocked), **`50/68`
stretch view assignments**, and **`41/47` Tiers 1–2 `discriminator_stability` PASS**
(`55/62` deduplicated unique runnable tasks PASS across 9 studies; `90.24 ms/step` median
and `1,372.60 MiB` allocated VRAM on matched 66-cell `dcgan_performance_r256`) — ahead of
`rhbi_optimized` (`rhbi-v12`, `83/115`, `33/69`, `590.59 ms/step`), `empf` (`empf-v8`,
`83/115`, `23/69`, `129.72 ms/step`), and `bispf_optimized` (`81/115`, `18/69`), while
`get_recipe("minimal")` offers an `899`-LOC zero-Newton-Schulz fast core (`55.50 ms/step`
on `r256`). Forge's screening calibration remains provisional (`BLOCKED` under
`criteria-v1.json`; see [r3-calibration-holdout](reports/forge/studies/r3-calibration-holdout.md)),
Tier 2/3 full-tier qualification is incomplete, and remaining failures (`staggered100` /
`staggered100_14k`, `grid100_res16_tanh`, `five_word_joint_hold`,
`model_ddgan_bimodal_chain`, `model_ae_ddgan_gaussian_chain`, `interp_img_patch_shift`,
`interp_saddle_barycentric`, plus `12` unimplemented Tier-4 stretch hosts) stay visible.
- **Controlled prior comparison** ([matched prior ablation, 2026-10-25](reports/prior-ablation-2026-10-25/README.md)).
Each arm changes only the prior policy. Architecture, data, budget, seed 0 and
the initial table are the same. These are one-seed diagnostics with no
Expand Down Expand Up @@ -154,15 +168,15 @@ What the evidence does and does not show today:

### The 100-Gaussian animations

These three animations use different historical or opt-in recipes rather than the current `empf` default.
The lead [`100gaussians.gif`](100gaussians.gif) shows actual training of the current **0.11.0 default** (`get_recipe()` = `unified_v11`) on the Forge `grid100` `MoGParticlePrior(20000, σ=0.025, standardize=False)` cohort; the other two animations retain their archived 0.9.1 and opt-in E22 identities.

| Animation | Recipe | What it shows | Pass/fail metric | Result |
| --- | --- | --- | --- | --- |
| [`100gaussians.gif`](100gaussians.gif) | **Archived 0.9.0 default** (`get_recipe("bispf_optimized", optimizer_svd_backend="native", training_precision="float32")`). | 100-Gaussian grid, public deterministic initialization, protocol seed 0, 20,000 updates. | All 100 modes with ≥10 samples within 3σ, **and** `hq` (share of samples within 3σ) ≥ 0.95, at each of 5 terminal checks. | **FAIL**: 100/100 modes, but `hq` 42.1% ([receipt](reports/readme-100gaussians/README.md#current-default)) |
| [`100gaussians.gif`](100gaussians.gif) | **0.11.0 default** (`get_recipe()` = `unified_v11`) with the Forge `grid100` task prior (`MoGParticlePrior(20000, σ=0.025, standardize=False)`). | 100-Gaussian grid, public deterministic initialization, protocol seed 0, 7,000 updates (live weights, clean sampling, 260.5 s on RTX A6000). | All 100 modes with ≥10 samples within 3σ, **and** `hq` (share of samples within 3σ) ≥ 0.95, at each of 5 terminal checks. | **PASS**: 100/100 modes, `hq` 98.52% (5/5 terminal checks 98.52%–99.39%), Mode TV `0.04555`, Sliced W1 `0.08353`, core width ratio `0.957` ([receipt](reports/readme-100gaussians/README.md#current-default); archived 0.9.0 `ParticlePrior` FAIL retained [here](reports/readme-100gaussians/README.md#archived-090-default-particleprior-cohort)) |
| [stripes GIF](reports/forge/bispf-throughput/media/bispf-throughput-v1/img_stripes2.gif) | **Archived 0.9.1 default** (`get_recipe("bispf_optimized")`, batched SVD, BF16/channels-last) | Fixed 600-update `img_stripes2` image task | The task's declared Forge gate, including all required modes | **PASS** ([readout](reports/forge/bispf-throughput/README.md)) |
| [E22 GIF](reports/e22-animation/e22-100gaussians.gif) | **Not the default.** Opt-in preset `get_recipe("e22", ...)` with an identity `nn.Linear(2, 2)` generator, a 20,000-row table, batch 2,048 and seed 1234 | Grid, rotated and staggered 100-Gaussian layouts, 7,000 updates each | Native gates: worst live precision over the last 5 checks ≥ 0.97 and worst centre error ≤ 0.20σ | **PASS** on all three layouts (precision .9822–.9851); the animation draws reach `hq` 98.3–98.6% ([E22 results](docs/e22.md#results), [animation](reports/e22-animation/README.md)). In the Forge suite, E22 is 0/115: its policy cohort is blocked or incompatible with the current task definitions. |

![Archived 0.9.0 default on the 100-Gaussian grid: particle outputs, target modes and numerical learning curves (FAIL: hq 42.1% < 95%)](100gaussians.gif)
![0.11.0 default (unified_v11, MoGParticlePrior) on the 100-Gaussian grid: particle outputs, target modes and numerical learning curves (PASS: 100/100 modes, hq 98.5%)](100gaussians.gif)

![Actual training of the 0.9.1 BISPF default on the stripes task, with numerical quality observations (PASS)](reports/forge/bispf-throughput/media/bispf-throughput-v1/img_stripes2.gif)

Expand All @@ -186,9 +200,10 @@ qualification and review of the winner. The
acquisition, settling, endurance and recovery required again.

**Changing the default.** A promotion changes one line in
[particlegan/recipes.py](particlegan/recipes.py): `DEFAULT_RECIPE = "empf"`.
[particlegan/recipes.py](particlegan/recipes.py): `DEFAULT_RECIPE = "unified_v11"`.
All model-family factories inherit that trainer preset. Explicit presets such as
`get_recipe("rhbi_optimized")`, `get_recipe("rhbi")`, `get_recipe("bispf_optimized")` and
`get_recipe("minimal")`, `get_recipe("empf")`, `get_recipe("rhbi_optimized")`,
`get_recipe("rhbi")`, `get_recipe("bispf_optimized")` and
`get_recipe("ka2")` remain available. This is the owner-selected research default;
Forge calibration remains provisional (`BLOCKED`), Tier 2/3 qualification is incomplete,
and untouched holdout confirmation remains open.
Expand Down
4 changes: 3 additions & 1 deletion RESEARCH_GLOSSARY.md
Original file line number Diff line number Diff line change
Expand Up @@ -45,7 +45,9 @@ so the row gives a functional description instead.

| Name | Expansion | What it is | Source |
|---|---|---|---|
| **EMPF** | Exponential Moving Prototype Flow | The current default family (`DEFAULT_RECIPE = "empf"`, matching `empf-v8`): clock-free online *e-means* sufficient-statistics prototype bank, mass-weighted prototype-to-fake Sinkhorn OT, state-driven settling Born-Infeld critic penalty, and DoF-aware EMPF prior regularizer. | `particlegan/empf.py` module docstring; `particlegan/recipes.py`; [`reports/forge/empf/README.md`](reports/forge/empf/README.md) |
| **Unified** (`unified_v11`) | Unified 4-Law Zero-SVD / Zero-`eigh` / Zero-`pinv` / Zero-`autograd.grad` Formulation | The current default family (`DEFAULT_RECIPE = "unified_v11"`, matching `unified-v11`): unites Law 1 streaming Laguerre-Pythagorean prototype prior & orthogonal pullback isometry, Law 2 first-order Born-Infeld secant critic with zero-mean gauge, Law 3 scale-invariant `k=4` density-ratio orphan macro-transport & zero-`eigh` cyclic-Jacobi conditional fiber bank, and Law 4 fused `Z=[X,Y]` Quintic Tikhonov equilibration with zero-`pinv` analytic `direction_blend`. | `particlegan/unified_v11.py`; `particlegan/recipes.py`; [`docs/four-laws.md`](docs/four-laws.md); [`reports/forge/unified/README.md`](reports/forge/unified/README.md); [`reports/forge/studies/r4-solution-complete.md`](reports/forge/studies/r4-solution-complete.md) |
| **Minimal** (`minimal`) | Minimal-Principle 4-Law Core | Explicit `<900`-LOC fast core (`get_recipe("minimal")`, matching `minimal-v2`): self-contained 899-LOC zero-sibling-import core with zero `autograd.grad`, zero SVD/`eigh`/`pinv`, and zero Newton-Schulz. | `particlegan/minimal.py`; `particlegan/recipes.py`; [`reports/forge/studies/r4-minimal-core.md`](reports/forge/studies/r4-minimal-core.md) |
| **EMPF** | Exponential Moving Prototype Flow | Explicit EMPF preset (`get_recipe("empf")`, matching `empf-v8`): clock-free online *e-means* sufficient-statistics prototype bank, mass-weighted prototype-to-fake Sinkhorn OT, state-driven settling Born-Infeld critic penalty, and DoF-aware EMPF prior regularizer. | `particlegan/empf.py` module docstring; `particlegan/recipes.py`; [`reports/forge/empf/README.md`](reports/forge/empf/README.md) |
| **RHBI** | Relativistic Hellinger-Born-Infeld Transport | Explicit RHBI presets (`get_recipe("rhbi_optimized")` matching `rhbi-v12`, and `get_recipe("rhbi")` retaining `rhbi-v1`): debiased slack-screened Kantorovich-Hellinger Sinkhorn generator transport (`RHBIGANLoss`), Weyl-phased balanced Born-Infeld geodesic chord and thermal-settling critic penalty (`RHBICriticPenalty`), SBNP prior regularization, and zero-momentum DualNorm. | `particlegan/rhbi.py` module docstring; `particlegan/recipes.py`; [`reports/forge/rhbi/README.md`](reports/forge/rhbi/README.md) |
| **BISPF** | Born-Infeld Sinkhorn Particle Flow | The previous 0.9.x default family (`get_recipe("bispf_optimized")` / `get_recipe("bispf")`): Sinkhorn relativistic pairing, a Born-Infeld critic penalty, the SBNP prior and residual-driven noise. | `particlegan/bispf.py` module docstring; `particlegan/recipes.py` |
| **SBNP** | Stratified Blue-Noise Particle Prior | BISPF's prior regularizer / particle table (`StratifiedBlueNoisePrior`). | `particlegan/bispf.py` module docstring |
Expand Down
Loading
Loading