docs(visible): record the 0.12.1-vs-0.14 head-to-head + backend quality

Full-dataset validation of reverse-alpha (v0.12.1) vs the current localize->fill:
doubao/jimeng identical (100% coverage + clearance across all backends); gemini
strict coverage a few points below reverse-alpha (the FP tightening), every
missed mark recovered under assume_ai, clearance ~98% both, no outside-box
damage. Clearance is fill-independent (cv2/MI-GAN/LaMa all strip the mark shape);
the difference is visual fill quality on textured/structured backgrounds -- LaMa
best, MI-GAN can ghost/hallucinate, cv2 smears -- which motivates auto = LaMa >
MI-GAN > cv2. Added to module-internals, known-limitations, and the CLAUDE.md
compact list.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

docs/known-limitations.md

@@ -9,6 +9,30 @@ measurements, incident history, oracle runs, and the reasoning behind each
 decision. Read the relevant section here before changing the diffusion
 pipelines, strength defaults, or metadata coverage.
 
+## Visible-mark fill quality is background- and backend-dependent
+
+Visible-mark removal is localize -> fill (the reverse-alpha pixel recovery was
+dropped; see `docs/module-internals.md`). The fill only touches the mark's
+footprint, so there is never collateral damage outside it, and whether the mark
+is *removed* is fill-independent -- cv2, MI-GAN and LaMa all strip the mark's
+shape. What varies is the *quality of the recovered region*, and it depends on
+the background:
+
+- **Flat backgrounds:** every backend is clean; cv2 is often the crispest.
+- **Textured / regular-structured backgrounds** (fabric, foliage, a lattice or
+  grid): an inpaint can only guess the hidden pixels. `cv2` (the classical
+  no-deps floor) visibly smears; `migan` (light, learned) can leave a ghost or
+  hallucinate structure; `lama` (heavy, learned) is the most reliable and
+  recovers structure best.
+
+The old reverse-alpha recovered the *true* pixels under a well-captured, static
+mark, so on structured backgrounds it was sometimes cleaner than any inpaint.
+The trade for localize -> fill is robustness (it also handles moved / re-rendered
+marks and needs no per-mark alpha capture) and a simpler, swappable pipeline.
+`auto` resolves best-first (`LaMa > MI-GAN > cv2`) and warns once when it falls
+back to cv2 because no learned backend is installed; a memory-tight deployment
+that cannot afford LaMa's ~4.7 GB peak pins `--backend migan` explicitly.
+
 ## Invisible-pipeline resolution handling (native / 1024 floor / `--max-resolution`; MPS memory tiers)
 
 `invisible` pipeline processes at **native resolution for inputs whose long side is >= 1024px**, and **auto-upscales smaller inputs UP to a 1024px floor** (`min_resolution=1024`, the default; `--min-resolution 0` disables) before diffusion -- SDXL img2img distorts badly on a tiny latent (a 381x512 portrait wrecks at native, the #36 follow-up), and the output is restored to the original input size so the floor is a transparent quality boost (it adds time/memory on small inputs). The floor upscale uses Lanczos by default; **`--upscaler esrgan`** (opt-in, the `esrgan` extra) runs Real-ESRGAN first for better detail before the Lanczos resize to the exact target (`upscaler.py` / `InvisibleEngine._esrgan_upscale`, falls back to Lanczos if the extra is absent). `max_resolution=0` (default) means no downscale cap, matching the hosted raiw.cc backend (fal fast-sdxl, no pre-downscale). The old forced downscale-to-1024 -> upscale-back round-trip for LARGE images was the main quality loss (issue #10) and is gone; at strength ~0.05 SDXL img2img does not need a downscale.

docs/module-internals.md

@@ -63,6 +63,8 @@ module.
 
 **Localize -> fill by policy (replaced reverse-alpha):** each mark is localized to a binary full-frame footprint mask (a `Localization`), and one shared, swappable fill inpaints that mask via `fill(image, mask, backend=...)` (delegates to `region_eraser.erase`). This replaced the old reverse-alpha removal (invert a captured alpha map, `original = (wm - a*logo)/(1-a)`, plus a thin residual inpaint) for ALL marks — gemini, doubao, jimeng, samsung, and jimeng_pill. **Why it changed:** reverse-alpha depended on a fixed captured alpha map at a fixed position, so it broke whenever a vendor moved or re-rendered its mark; and it was not color-lossless even with the right map (it amplifies 8-bit quantization and JPEG-chroma error by `1/(1-a)`), which showed up as "the color just changed, not removed" reports. Localize -> fill has a benign failure mode: a slightly-off localization just inpaints a small region near-losslessly instead of leaving a color-shifted smear. The captured alpha maps are still used to DETECT the marks and to shape the mask (gemini's footprint), but NOT for pixel recovery. Fill backends: `cv2` (classical inpaint, no deps, the floor), `migan` (MI-GAN ONNX, light, the memory-tight pick where LaMa will not fit), `lama` (big-LaMa ONNX, best quality, heavier, auto-preferred when a learned backend is available); `auto` = LaMa > MI-GAN > cv2, best available. Each `KnownMark` ties a key to {usual `location`, `in_auto` flag, a `_detect` callable → uniform `MarkDetection`, a `_mask` callable → full-frame footprint mask}; `KnownMark.remove(image, *, backend="auto", provenance=False, force=False)`. Entries today: `gemini` (bottom-right sparkle), `doubao` (bottom-right "豆包AI生成"), `jimeng` (bottom-right "★ 即梦AI"), `samsung` (bottom-**LEFT** "✦ Contenuti generati dall'AI", Samsung Galaxy AI, Italian locale), and the capture-less `jimeng_pill` (top-left "AI生成"). `detect_marks(image, *, provenance=frozenset())` scans all (strict, for the identify verdict); `remove_auto_marks(image, *, sensitivity="auto", provenance=frozenset(), backend="auto")` removes every detected mark in one pass. **Sensitivity (`auto`/`strict`/`assume_ai`)** decides how hard a borderline mark is trusted: the visual detectors are pixel-based (no metadata needed) and the recall gain comes from relaxing the false-positive gate, not from metadata. `resolve_relax` turns the policy + evidence into the per-mark relaxation boolean — `strict` never relaxes, `assume_ai` always relaxes (the caller asserts AI, e.g. a metadata-stripped screenshot; ~46% -> ~92% Gemini recall, at the cost of a small near-lossless fill on some clean corners), `auto` relaxes only on same-product evidence (metadata provenance for that vendor, or a confidently strict-detected sibling of the same `_PRODUCT_OF` — Doubao and Jimeng are both bottom-right ByteDance but distinct products, so they do NOT cross-relax). **Perception / decision / action are separated three ways** (the removal path only; `identify` keeps calling `KnownMark.detect` directly, so its verdict is untouched): `_build_candidates(image)` is PERCEPTION — it runs each detector at both trust levels and packages raw verdicts + the pill's flatness feature into `Candidate`s, no policy; `decide(candidates, Context(sensitivity, provenance)) -> [Decision]` is the pure DECISION arbiter — all keep/drop policy (`resolve_relax` cross-mark corroboration + the pill gate) in one image-free, unit-testable function (`tests/test_watermark_registry.py::TestArbiter`); then `remove_auto_marks` does the ACTION, localizing -> filling each winner. The Gemini FP gate deliberately stays inside `gemini_engine` (not the arbiter) because `identify` reads that same gated confidence — pulling it out would drift the provenance verdict. Behavior is byte-identical to the pre-arbiter two-pass (corpus-verified: strict/auto 46%, assume_ai 92% Gemini recall unchanged).
 
+**Head-to-head validation (v0.12.1 reverse-alpha vs the current localize -> fill):** run over the full labelled visible-mark set, with the cv2 / MI-GAN / LaMa fills each compared against the old reverse-alpha. **doubao and jimeng are identical** across every backend -- 100% coverage and 100% clearance either way. **gemini** strict coverage is a few points below reverse-alpha's (the deliberate false-positive tightening), but every strict-missed mark is recovered under `assume_ai`, clearance is equal (~98% both), and neither version touches pixels outside the mark box (outside-box PSNR ~99). **Clearance is fill-independent** -- cv2, MI-GAN and LaMa all strip the mark's shape equally, so the re-detect metric does not separate them; the difference is purely the *visual fill quality* on the recovered region, and it is background-dependent. reverse-alpha recovered textured and especially regular/structured backgrounds (a lattice, a grid) more cleanly than any inpaint; **LaMa closes most of that gap** (the best learned backend), **MI-GAN can ghost or hallucinate structure**, and **cv2 smears** (the last-resort floor). This is why `auto` resolves `LaMa > MI-GAN > cv2` (`preferred_inpaint_backend`) and warns once on the cv2 fallback; on flat backgrounds every backend is clean.
+
 **Provenance prior:** when local metadata already confirms the vendor, the mark's detection trust gate is relaxed (a confirmed vendor means the mark is present with high prior, so a mark the conservative detector would demote as a content false positive is trusted). `detect_marks` / `remove_auto_marks` take a `provenance` frozenset and `KnownMark.remove` a `provenance` flag. Mapping: a Google/Gemini C2PA issuer relaxes gemini (skips its false-positive gate and lowers the trust threshold from 0.5 to 0.35); a China-AIGC (TC260) label relaxes doubao/jimeng; `samsung_genai` relaxes samsung. Corpus finding: on Google-C2PA images, Gemini sparkle recall rose from ~46% (plain detector) to ~90% with the provenance prior (recovering marks the vendor moved or re-rendered). The localizer is cheap CPU (cv2/numpy), so a memory-tight caller runs it anywhere; the heavy MI-GAN/LaMa fill is opt-in and chosen by the caller.
 
 **Cross-engine confidences aren't directly comparable**, so the gemini adapter applies the corpus-validated 0.5 sparkle threshold (`_GEMINI_AUTO_MIN_CONF`) for its `detected` flag (lowered to 0.35 under the Google/Gemini provenance prior) — otherwise the gemini engine's loose internal threshold weakly fires (~0.36) on the Doubao text and hijacks `auto`. The shape-keyed Doubao/Jimeng/Samsung NCC detectors don't cross-fire (jimeng scores ~0.22 on the Doubao strip, well under its 0.45 threshold; Samsung is bottom-left so it shares no corner with the others, and scored 0.0 on Doubao/Jimeng captures and they 0.0 on a real Samsung photo), so `auto` picks the right one. `cli.cmd_visible` is registry-driven: `--mark auto` → `remove_auto_marks` (removes every detected mark), `--mark <key>` → that mark; `--mark` choices come from `mark_keys()`.