feat(migration): Phase 5.C batch 3 — 5 renderers moyens vers reports_v2/html/

Troisième vague de migration des 22 renderers thématiques. Tous
les renderers sélectionnés sont purs sur le contrat : import
depuis ``_helpers/`` uniquement, pas de dépendance sur des modules
legacy non-migrés.

Migrations effectuées
---------------------
| Source legacy | Destination canonique |
|------------------------------------------------|--------------------------------------------------------|
| ``report/module_audit_render.py`` (173) | ``reports_v2/html/renderers/module_audit.py`` |
| ``report/incremental_comparison_render.py`` (201)| ``reports_v2/html/renderers/incremental_comparison.py``|
| ``report/image_predictive_render.py`` (207) | ``reports_v2/html/renderers/image_predictive.py`` |
| ``report/error_absorption_render.py`` (210) | ``reports_v2/html/renderers/error_absorption.py`` |
| ``report/ner_render.py`` (222) | ``reports_v2/html/renderers/ner.py`` |

Total : ~1013 lignes relocalisées. 5 nouveaux shims minimaux
(< 20 lignes) avec ``DeprecationWarning``.

Cumul Phase 5.C (batches 1+2+3) : 15 / 22 renderers migrés
(~2211 lignes). 7 renderers restants.

Acceptance
----------
5019 tests passent, lint vert, architecture vérifiée.

https://claude.ai/code/session_011XQZNitg1rCgia8ZD1a2hP

docs/migration/legacy-retirement-plan.md

@@ -692,12 +692,7 @@ architecture vérifiée.
 **Reporté aux batches suivants** :
 
 - Batch 2 ✅ (cf. ci-dessous) — 5 renderers (45-165 LOC).
-- Batch 3 (~5 renderers moyens) : ``module_audit``,
-  ``ner``, ``image_predictive``,
-  ``incremental_comparison``, ``numerical_sequences``
-  (ce dernier exige d'abord la migration du module
-  ``measurements/numerical_sequences.py`` qui dépend de
-  ``measurements/roman_numerals.py``).
+- Batch 3 ✅ (cf. ci-dessous) — 5 renderers (173-222 LOC).
 - Batch 4 (~5 renderers gros) : ``error_absorption``,
   ``baseline``, ``inter_engine``, ``robustness_projection``,
   ``stratification``.
@@ -752,6 +747,36 @@ architecture vérifiée.
 **Cumul Phase 5.C** (batches 1+2) : 10 / 22 renderers migrés
 (~1198 lignes).  12 renderers restants.
 
+#### Phase 5.C.batch3 — Lot 3 : 5 renderers moyens (2026-05)
+
+Troisième vague.  Tous les renderers sélectionnés sont
+**purs sur le contrat** : import depuis ``_helpers/`` uniquement,
+pas de dépendance sur des modules legacy non-migrés.
+
+**Migrations effectuées** :
+
+| Source legacy                                  | Destination canonique                                  |
+|------------------------------------------------|--------------------------------------------------------|
+| ``report/module_audit_render.py`` (173)        | ``reports_v2/html/renderers/module_audit.py``          |
+| ``report/incremental_comparison_render.py`` (201)| ``reports_v2/html/renderers/incremental_comparison.py``|
+| ``report/image_predictive_render.py`` (207)    | ``reports_v2/html/renderers/image_predictive.py``      |
+| ``report/error_absorption_render.py`` (210)    | ``reports_v2/html/renderers/error_absorption.py``      |
+| ``report/ner_render.py`` (222)                 | ``reports_v2/html/renderers/ner.py``                   |
+
+Total : ~1013 lignes relocalisées.  5 nouveaux shims minimaux
+(< 20 lignes) avec ``DeprecationWarning``.
+
+**Adaptations transverses** :
+
+- Tests + ``picarones/report/generator.py`` mis à jour pour les
+  5 chemins canoniques.
+
+**Acceptance batch 3** : 5019 tests passent, lint vert,
+architecture vérifiée.
+
+**Cumul Phase 5.C** (batches 1+2+3) : 15 / 22 renderers migrés
+(~2211 lignes).  7 renderers restants.
+
 ### Phase 6 — Pipelines OCR+LLM (`pipelines/`)
 
 **Modules** : `pipelines/base.OCRLLMPipeline` (3 modes), `pipelines/

picarones/report/error_absorption_render.py

@@ -1,210 +1,18 @@
-"""Rendu HTML « Absorption d'erreur » — Sprint 94 (B.3).
+"""``picarones.report.error_absorption_render`` — shim re-export (déprécié, suppression 2.0).
 
-Suite directe ``picarones/core/error_absorption.py``.  Pattern
-identique aux autres rendus : server-side, pas de JS, anti-
-injection systématique.
-
-Vue
----
-Tableau résumé des jonctions du pipeline ; chaque ligne décrit
-un module post-correction et présente :
-
-- erreurs en entrée vs en sortie ;
-- nb corrigées (gradient vert), nb introduites (gradient rouge) ;
-- taux de correction (gradient vert), taux d'introduction
-  (gradient rouge) ;
-- amélioration nette (n_corrected - n_introduced) — coloré.
-- éventuellement un échantillon de tokens corrigés/introduits.
-
-Adaptive : ``""`` si la liste est vide.
-
-Note d'intégration
-------------------
-Module pur — la liste ``junctions`` est composée par
-l'utilisateur depuis son benchmark de pipeline composée :
-
-.. code-block:: python
-
-    from picarones.measurements.error_absorption import (
-        compute_error_absorption, aggregate_error_absorption,
-    )
-    from picarones.report.error_absorption_render import (
-        build_error_absorption_html,
-    )
-
-    junctions = []
-    for step in pipeline.steps_with_text_output:
-        per_doc = [
-            compute_error_absorption(doc.gt_text, doc.before_text,
-                                     doc.after_text)
-            for doc in benchmark.docs
-        ]
-        agg = aggregate_error_absorption(per_doc)
-        if agg is not None:
-            agg["junction_name"] = step.name
-            junctions.append(agg)
-    html = build_error_absorption_html(junctions, labels)
+Canonique : :mod:`picarones.reports_v2.html.renderers.error_absorption`.
+Phase 5.C du retrait du legacy.
 """
 
 from __future__ import annotations
 
-from html import escape as _e
-from typing import Optional
-
-from picarones.reports_v2._helpers.render_helpers import color_diverging, color_traffic_light
-
-
-# Palette « net improvement » : vert clair au centre, vert profond
-# si favorable (net > 0), rouge si défavorable (net < 0).  Centrée
-# sur le vert clair car un delta nul est déjà « pas de régression ».
-_NET_NEUTRAL_RGB = (167, 240, 167)
-_NET_POSITIVE_RGB = (90, 200, 90)
-_NET_NEGATIVE_RGB = (220, 50, 50)
-
-
-def build_error_absorption_html(
-    junctions: Optional[list],
-    labels: Optional[dict[str, str]] = None,
-    *,
-    sample_max: int = 8,
-) -> str:
-    """Construit la vue HTML « Absorption d'erreur ».
-
-    Parameters
-    ----------
-    junctions:
-        Liste de dicts (un par jonction de pipeline), enrichis
-        d'un ``junction_name``.  Si vide ou ``None``, retourne
-        ``""``.
-    labels:
-        Dict i18n.  Clés sous le préfixe ``absorption_*``.
-    sample_max:
-        Nombre maximal de tokens corrigés/introduits affichés
-        en cellule échantillon.
-    """
-    if not junctions:
-        return ""
-    rows = [
-        j for j in junctions
-        if isinstance(j, dict) and j.get("junction_name")
-    ]
-    if not rows:
-        return ""
-    labels = labels or {}
-    title = labels.get(
-        "absorption_title", "Absorption d'erreur par jonction",
-    )
-    note = labels.get(
-        "absorption_note",
-        "À chaque jonction du pipeline, deux flux sont mesurés "
-        "indépendamment : combien d'erreurs sont corrigées et "
-        "combien sont introduites. Une jonction qui corrige "
-        "beaucoup mais introduit aussi beaucoup absorbe les "
-        "différences amont au lieu de les améliorer.",
-    )
-    h_junction = labels.get("absorption_junction", "Jonction")
-    h_errors_before = labels.get("absorption_errors_before", "Erreurs avant")
-    h_errors_after = labels.get("absorption_errors_after", "Erreurs après")
-    h_corrected = labels.get("absorption_corrected", "Corrigées")
-    h_introduced = labels.get("absorption_introduced", "Introduites")
-    h_corr_rate = labels.get("absorption_corr_rate", "% corrigées")
-    h_intro_rate = labels.get("absorption_intro_rate", "% introduites")
-    h_net = labels.get("absorption_net", "Amélioration nette")
-    h_sample = labels.get("absorption_sample", "Échantillon (intro)")
-
-    # Saturation pour le gradient « net »
-    max_abs_net = max(
-        (abs(int(r.get("net_improvement") or 0)) for r in rows), default=1,
-    ) or 1
-
-    parts = [
-        '<section class="absorption-section" style="margin:1rem 0">',
-        f'<h3 style="margin:0 0 .3rem 0">{_e(title)}</h3>',
-        f'<div style="font-size:.85rem;opacity:.75;margin-bottom:.6rem">'
-        f'{_e(note)}</div>',
-        '<table style="border-collapse:collapse;width:100%;'
-        'font-size:.9rem">',
-        '<thead><tr>',
-    ]
-    for col in (h_junction, h_errors_before, h_errors_after,
-                h_corrected, h_introduced, h_corr_rate,
-                h_intro_rate, h_net, h_sample):
-        parts.append(
-            f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:left;'
-            f'border-bottom:1px solid #ccc;font-weight:600">'
-            f'{_e(col)}</th>'
-        )
-    parts.append("</tr></thead><tbody>")
-    for entry in rows:
-        name = str(entry.get("junction_name") or "?")
-        n_eb = int(entry.get("n_errors_before") or 0)
-        n_ea = int(entry.get("n_errors_after") or 0)
-        n_corr = int(entry.get("n_corrected") or 0)
-        n_intro = int(entry.get("n_introduced") or 0)
-        net = int(entry.get("net_improvement") or 0)
-        corr_rate = entry.get("correction_rate")
-        intro_rate = entry.get("introduction_rate")
-        if isinstance(corr_rate, (int, float)):
-            corr_rate_str = f"{corr_rate * 100:.1f}%"
-            corr_color = color_traffic_light(float(corr_rate))
-            corr_cell = (
-                f'<td style="padding:.4rem .6rem;text-align:right;'
-                f'background:{corr_color};font-family:monospace;'
-                f'font-weight:600">{corr_rate_str}</td>'
-            )
-        else:
-            corr_cell = (
-                '<td style="padding:.4rem .6rem;text-align:right;'
-                'opacity:.4">—</td>'
-            )
-        if isinstance(intro_rate, (int, float)):
-            intro_rate_str = f"{intro_rate * 100:.1f}%"
-            intro_color = color_traffic_light(float(intro_rate), low_is_good=True)
-            intro_cell = (
-                f'<td style="padding:.4rem .6rem;text-align:right;'
-                f'background:{intro_color};font-family:monospace;'
-                f'font-weight:600">{intro_rate_str}</td>'
-            )
-        else:
-            intro_cell = (
-                '<td style="padding:.4rem .6rem;text-align:right;'
-                'opacity:.4">—</td>'
-            )
-        net_color = color_diverging(
-            float(net),
-            max_abs=float(max_abs_net) if max_abs_net else 1.0,
-            neutral_rgb=_NET_NEUTRAL_RGB,
-            positive_rgb=_NET_POSITIVE_RGB,
-            negative_rgb=_NET_NEGATIVE_RGB,
-        )
-        intro_sample = entry.get("introduced_tokens_sample") or []
-        sample_cell_text = ", ".join(
-            _e(str(t)) for t in intro_sample[:sample_max]
-        ) or "—"
-        if len(intro_sample) > sample_max:
-            sample_cell_text += " …"
-        parts.append(
-            f'<tr>'
-            f'<td style="padding:.4rem .6rem">{_e(name)}</td>'
-            f'<td style="padding:.4rem .6rem;text-align:right;'
-            f'font-family:monospace">{n_eb}</td>'
-            f'<td style="padding:.4rem .6rem;text-align:right;'
-            f'font-family:monospace">{n_ea}</td>'
-            f'<td style="padding:.4rem .6rem;text-align:right;'
-            f'font-family:monospace">{n_corr}</td>'
-            f'<td style="padding:.4rem .6rem;text-align:right;'
-            f'font-family:monospace">{n_intro}</td>'
-            f'{corr_cell}'
-            f'{intro_cell}'
-            f'<td style="padding:.4rem .6rem;text-align:right;'
-            f'background:{net_color};font-family:monospace;'
-            f'font-weight:600">{net:+d}</td>'
-            f'<td style="padding:.4rem .6rem;font-family:monospace;'
-            f'font-size:.8rem">{sample_cell_text}</td>'
-            f'</tr>'
-        )
-    parts.append("</tbody></table></section>")
-    return "".join(parts)
+import warnings
 
+from picarones.reports_v2.html.renderers.error_absorption import *  # noqa: F401, F403
 
-__all__ = ["build_error_absorption_html"]
+warnings.warn(
+    "picarones.report.error_absorption_render is deprecated and will be removed in 2.0.  "
+    "Import from picarones.reports_v2.html.renderers.error_absorption instead.",
+    DeprecationWarning,
+    stacklevel=2,
+)

picarones/report/generator.py

@@ -268,7 +268,7 @@ class ReportGenerator:
             build_oracle_gap_html,
         )
         # Sprint 41 — section NER (résumé F1 par moteur + heatmap par catégorie).
-        from picarones.report.ner_render import (
+        from picarones.reports_v2.html.renderers.ner import (
             build_ner_per_category_html,
             build_ner_summary_html,
         )

picarones/report/image_predictive_render.py

@@ -1,207 +1,18 @@
-"""Rendu HTML « Profil d'image du corpus » — Sprint 93 (A.II.7).
+"""``picarones.report.image_predictive_render`` — shim re-export (déprécié, suppression 2.0).
 
-Suite directe ``picarones/core/image_predictive.py``.  Pattern
-identique aux autres rendus : server-side, pas de JS, anti-
-injection systématique.
-
-Vue
----
-Deux blocs dans une section unique :
-
-1. **Complexité paléographique** : moyenne, médiane, min, max,
-   écart-type sur l'ensemble du corpus.
-2. **Homogénéité du corpus** : score combiné + détail par
-   feature (mean, stdev, contribution normalisée).
-
-Adaptive : ``""`` si pas de données.
-
-Note d'intégration
-------------------
-Module pur — l'utilisateur compose :
-
-.. code-block:: python
-
-    from picarones.measurements.image_predictive import aggregate_corpus_predictive
-    from picarones.report.image_predictive_render import (
-        build_image_predictive_html,
-    )
-
-    qualities = [doc.image_quality.as_dict() for doc in benchmark.docs]
-    agg = aggregate_corpus_predictive(qualities)
-    html = build_image_predictive_html(agg, labels)
+Canonique : :mod:`picarones.reports_v2.html.renderers.image_predictive`.
+Phase 5.C du retrait du legacy.
 """
 
 from __future__ import annotations
 
-from html import escape as _e
-from typing import Optional
-
-from picarones.reports_v2._helpers.render_helpers import color_traffic_light
-
-
-_FEATURE_LABEL_KEYS = {
-    "noise_level": "imgpred_feat_noise",
-    "sharpness_score": "imgpred_feat_sharpness",
-    "contrast_score": "imgpred_feat_contrast",
-    "rotation_degrees": "imgpred_feat_rotation",
-}
-
-
-def _render_complexity_block(
-    aggregated: dict, labels: dict[str, str],
-) -> str:
-    h_complex = labels.get(
-        "imgpred_complexity", "Complexité paléographique",
-    )
-    h_mean = labels.get("imgpred_mean", "Moyenne")
-    h_median = labels.get("imgpred_median", "Médiane")
-    h_min = labels.get("imgpred_min", "Min")
-    h_max = labels.get("imgpred_max", "Max")
-    h_stdev = labels.get("imgpred_stdev", "Écart-type")
-    h_docs = labels.get("imgpred_docs", "Docs")
-    mean = float(aggregated.get("complexity_mean") or 0.0)
-    median = float(aggregated.get("complexity_median") or 0.0)
-    mn = float(aggregated.get("complexity_min") or 0.0)
-    mx = float(aggregated.get("complexity_max") or 0.0)
-    sd = float(aggregated.get("complexity_stdev") or 0.0)
-    n_docs = int(aggregated.get("n_docs") or 0)
-    color_mean = color_traffic_light(mean, low_is_good=True)
-    return (
-        f'<div style="font-weight:600;margin:.4rem 0 .3rem 0">'
-        f'{_e(h_complex)}</div>'
-        '<table style="border-collapse:collapse;width:100%;'
-        'font-size:.9rem;margin-bottom:.8rem">'
-        f'<thead><tr>'
-        f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:right;'
-        f'border-bottom:1px solid #ccc;font-weight:600">{_e(h_mean)}</th>'
-        f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:right;'
-        f'border-bottom:1px solid #ccc;font-weight:600">{_e(h_median)}</th>'
-        f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:right;'
-        f'border-bottom:1px solid #ccc;font-weight:600">{_e(h_min)}</th>'
-        f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:right;'
-        f'border-bottom:1px solid #ccc;font-weight:600">{_e(h_max)}</th>'
-        f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:right;'
-        f'border-bottom:1px solid #ccc;font-weight:600">{_e(h_stdev)}</th>'
-        f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:right;'
-        f'border-bottom:1px solid #ccc;font-weight:600">{_e(h_docs)}</th>'
-        f'</tr></thead>'
-        f'<tbody><tr>'
-        f'<td style="padding:.4rem .6rem;text-align:right;'
-        f'background:{color_mean};font-family:monospace;font-weight:600">'
-        f'{mean:.3f}</td>'
-        f'<td style="padding:.4rem .6rem;text-align:right;'
-        f'font-family:monospace">{median:.3f}</td>'
-        f'<td style="padding:.4rem .6rem;text-align:right;'
-        f'font-family:monospace">{mn:.3f}</td>'
-        f'<td style="padding:.4rem .6rem;text-align:right;'
-        f'font-family:monospace">{mx:.3f}</td>'
-        f'<td style="padding:.4rem .6rem;text-align:right;'
-        f'font-family:monospace">{sd:.3f}</td>'
-        f'<td style="padding:.4rem .6rem;text-align:right;'
-        f'font-family:monospace">{n_docs}</td>'
-        f'</tr></tbody></table>'
-    )
-
-
-def _render_homogeneity_block(
-    homogeneity: dict, labels: dict[str, str],
-) -> str:
-    h_homo = labels.get(
-        "imgpred_homogeneity", "Homogénéité du corpus",
-    )
-    h_feat = labels.get("imgpred_feature", "Feature")
-    h_mean = labels.get("imgpred_feat_mean", "Moyenne")
-    h_stdev = labels.get("imgpred_feat_stdev", "Écart-type")
-    h_norm = labels.get(
-        "imgpred_feat_norm", "Contribution normalisée",
-    )
-    score = float(homogeneity.get("score") or 0.0)
-    color = color_traffic_light(score, low_is_good=True)
-    parts = [
-        f'<div style="font-weight:600;margin:.4rem 0 .3rem 0">'
-        f'{_e(h_homo)} : '
-        f'<span style="background:{color};padding:.1rem .4rem;'
-        f'border-radius:.3rem;font-family:monospace">{score:.3f}</span>'
-        f'</div>',
-        '<table style="border-collapse:collapse;width:100%;'
-        'font-size:.9rem">',
-        '<thead><tr>',
-    ]
-    for col in (h_feat, h_mean, h_stdev, h_norm):
-        parts.append(
-            f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:left;'
-            f'border-bottom:1px solid #ccc;font-weight:600">'
-            f'{_e(col)}</th>'
-        )
-    parts.append("</tr></thead><tbody>")
-    per_feat = homogeneity.get("per_feature") or {}
-    for key, label_key in _FEATURE_LABEL_KEYS.items():
-        if key not in per_feat:
-            continue
-        slot = per_feat[key]
-        feat_label = labels.get(label_key, key)
-        feat_mean = float(slot.get("mean") or 0.0)
-        feat_stdev = float(slot.get("stdev") or 0.0)
-        feat_norm = float(slot.get("normalised") or 0.0)
-        norm_color = color_traffic_light(feat_norm, low_is_good=True)
-        parts.append(
-            f'<tr>'
-            f'<td style="padding:.4rem .6rem">{_e(feat_label)}</td>'
-            f'<td style="padding:.4rem .6rem;text-align:right;'
-            f'font-family:monospace">{feat_mean:.3f}</td>'
-            f'<td style="padding:.4rem .6rem;text-align:right;'
-            f'font-family:monospace">{feat_stdev:.3f}</td>'
-            f'<td style="padding:.4rem .6rem;text-align:right;'
-            f'background:{norm_color};font-family:monospace">'
-            f'{feat_norm:.3f}</td>'
-            f'</tr>'
-        )
-    parts.append("</tbody></table>")
-    return "".join(parts)
-
-
-def build_image_predictive_html(
-    aggregated: Optional[dict],
-    labels: Optional[dict[str, str]] = None,
-) -> str:
-    """Construit la vue HTML « Profil d'image du corpus ».
-
-    Parameters
-    ----------
-    aggregated:
-        Sortie de ``aggregate_corpus_predictive``.  Si ``None``
-        ou ``n_docs == 0``, retourne ``""``.
-    labels:
-        Dict i18n.  Clés sous le préfixe ``imgpred_*``.
-    """
-    if not aggregated:
-        return ""
-    if not aggregated.get("n_docs"):
-        return ""
-    labels = labels or {}
-    title = labels.get(
-        "imgpred_title", "Profil d'image du corpus",
-    )
-    note = labels.get(
-        "imgpred_note",
-        "Score de complexité paléographique combinant bruit, "
-        "flou, faible contraste et rotation. Le score "
-        "d'homogénéité signale si la moyenne globale est fiable "
-        "(corpus uniforme) ou trompeuse (corpus hétérogène — "
-        "voir alors la vue stratifiée).",
-    )
-    parts = [
-        '<section class="imgpred-section" style="margin:1rem 0">',
-        f'<h3 style="margin:0 0 .3rem 0">{_e(title)}</h3>',
-        f'<div style="font-size:.85rem;opacity:.75;margin-bottom:.6rem">'
-        f'{_e(note)}</div>',
-    ]
-    parts.append(_render_complexity_block(aggregated, labels))
-    homo = aggregated.get("homogeneity")
-    if isinstance(homo, dict):
-        parts.append(_render_homogeneity_block(homo, labels))
-    parts.append("</section>")
-    return "".join(parts)
+import warnings
 
+from picarones.reports_v2.html.renderers.image_predictive import *  # noqa: F401, F403
 
-__all__ = ["build_image_predictive_html"]
+warnings.warn(
+    "picarones.report.image_predictive_render is deprecated and will be removed in 2.0.  "
+    "Import from picarones.reports_v2.html.renderers.image_predictive instead.",
+    DeprecationWarning,
+    stacklevel=2,
+)

picarones/report/incremental_comparison_render.py

@@ -1,201 +1,18 @@
-"""Rendu HTML « Comparaison contrôlée » — Sprint 96 (B.5).
+"""``picarones.report.incremental_comparison_render`` — shim re-export (déprécié, suppression 2.0).
 
-Suite directe ``picarones/core/incremental_comparison.py``.
-Pattern identique aux autres rendus : server-side, pas de JS,
-anti-injection systématique.
-
-Vue
----
-Tableau ANOVA-like : pour chaque valeur du slot variant, mean
-± stdev, rang moyen, n_observations.  Mean colorée en
-gradient vert (meilleur) → rouge (pire), normalisée sur la
-plage des moyennes observées.
-
-Adaptive : ``""`` si ``analysis`` est ``None``.
-
-Note d'intégration
-------------------
-Module pur — l'utilisateur compose :
-
-.. code-block:: python
-
-    from picarones.measurements.incremental_comparison import (
-        PipelineRun, compare_isolated_effect,
-    )
-    from picarones.report.incremental_comparison_render import (
-        build_incremental_comparison_html,
-    )
-
-    runs = [
-        PipelineRun(name=p.name,
-                    slots={"ocr": p.ocr, "llm": p.llm},
-                    score=p.cer_mean)
-        for p in benchmark.pipelines
-    ]
-    analysis = compare_isolated_effect(runs, "llm")
-    html = build_incremental_comparison_html(analysis, labels)
+Canonique : :mod:`picarones.reports_v2.html.renderers.incremental_comparison`.
+Phase 5.C du retrait du legacy.
 """
 
 from __future__ import annotations
 
-from html import escape as _e
-from typing import Optional
-
-from picarones.reports_v2._helpers.render_helpers import color_traffic_light
-
-
-def _bg_for_relative_score(
-    score: float, low: float, high: float, higher_is_better: bool,
-) -> str:
-    """Mappe ``score`` sur une plage [low, high] et retourne une cellule
-    colorée traffic-light.
-
-    Si ``higher_is_better=True``, ``score=high`` est vert ; sinon
-    ``score=low`` est vert.
-    """
-    if high == low:
-        return color_traffic_light(1.0)  # neutre vert clair
-    return color_traffic_light(
-        score,
-        low_is_good=not higher_is_better,
-        scale_min=low,
-        scale_max=high,
-    )
-
-
-def _format_score(value: Optional[float]) -> str:
-    if value is None:
-        return "—"
-    if abs(value) < 1.0:
-        return f"{value * 100:.2f}%"
-    return f"{value:.3f}"
-
-
-def build_incremental_comparison_html(
-    analysis: Optional[dict],
-    labels: Optional[dict[str, str]] = None,
-) -> str:
-    """Construit la vue HTML « Comparaison contrôlée ».
-
-    Parameters
-    ----------
-    analysis:
-        Sortie de ``compare_isolated_effect``.  ``None`` ou
-        ``per_value`` vide → retourne ``""``.
-    labels:
-        Dict i18n.  Clés sous le préfixe ``incr_*``.
-    """
-    if not analysis:
-        return ""
-    per_value = analysis.get("per_value") or {}
-    if not per_value:
-        return ""
-    labels = labels or {}
-    title = labels.get(
-        "incr_title", "Comparaison contrôlée par slot",
-    )
-    note = labels.get(
-        "incr_note",
-        "Effet isolé du module variant sur les pipelines en "
-        "contrôlant les autres slots. Pour chaque valeur du "
-        "slot, moyenne ± écart-type, rang moyen sur les groupes "
-        "fixes, et nombre d'observations. Type design "
-        "d'expérience pour des comparaisons honnêtes.",
-    )
-    slot_label = labels.get("incr_slot_label", "Slot variant")
-    h_value = labels.get("incr_value", "Valeur")
-    h_mean = labels.get("incr_mean", "Score moyen")
-    h_stdev = labels.get("incr_stdev", "± σ")
-    h_rank = labels.get("incr_rank", "Rang moyen")
-    h_n_obs = labels.get("incr_n_obs", "Observations")
-    h_groups = labels.get("incr_groups", "Groupes fixes")
-    higher_is_better = bool(analysis.get("higher_is_better", False))
-
-    # Plage de moyennes pour le code couleur
-    means = [
-        d["mean"] for d in per_value.values() if d.get("mean") is not None
-    ]
-    low = min(means) if means else 0.0
-    high = max(means) if means else 0.0
-
-    varying_slot = str(analysis.get("varying_slot") or "?")
-    n_groups = int(analysis.get("n_groups") or 0)
-    n_runs = int(analysis.get("n_runs") or 0)
-    best = analysis.get("best_value")
-    worst = analysis.get("worst_value")
-
-    parts = [
-        '<section class="incr-section" style="margin:1rem 0">',
-        f'<h3 style="margin:0 0 .3rem 0">{_e(title)}</h3>',
-        f'<div style="font-size:.85rem;opacity:.75;margin-bottom:.5rem">'
-        f'{_e(note)}</div>',
-        f'<div style="font-size:.85rem;margin-bottom:.5rem">'
-        f'<strong>{_e(slot_label)} :</strong> '
-        f'<code>{_e(varying_slot)}</code> &nbsp; '
-        f'<span style="opacity:.75">'
-        f'{n_runs} runs, {n_groups} {_e(h_groups.lower())}'
-        f'</span></div>',
-        '<table style="border-collapse:collapse;width:100%;'
-        'font-size:.9rem">',
-        '<thead><tr>',
-    ]
-    for col in (h_value, h_mean, h_stdev, h_rank, h_n_obs):
-        parts.append(
-            f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:left;'
-            f'border-bottom:1px solid #ccc;font-weight:600">'
-            f'{_e(col)}</th>'
-        )
-    parts.append("</tr></thead><tbody>")
-
-    # Tri par rang moyen ascendant
-    rows = sorted(
-        per_value.items(),
-        key=lambda kv: (kv[1].get("mean_rank") or float("inf")),
-    )
-    for value, d in rows:
-        mean = d.get("mean")
-        stdev = d.get("stdev")
-        rank = d.get("mean_rank")
-        n_obs = int(d.get("n_observations") or 0)
-        if isinstance(mean, (int, float)):
-            color = _bg_for_relative_score(
-                float(mean), low, high, higher_is_better,
-            )
-            mean_cell = (
-                f'<td style="padding:.4rem .6rem;text-align:right;'
-                f'background:{color};font-family:monospace;'
-                f'font-weight:600">{_format_score(mean)}</td>'
-            )
-        else:
-            mean_cell = (
-                '<td style="padding:.4rem .6rem;text-align:right;'
-                'opacity:.4">—</td>'
-            )
-        stdev_str = (
-            f"± {_format_score(stdev)}"
-            if isinstance(stdev, (int, float)) else "—"
-        )
-        rank_str = f"{rank:.2f}" if isinstance(rank, (int, float)) else "—"
-        marker = ""
-        if value == best:
-            marker = ' <span style="color:#16a34a">★</span>'
-        elif value == worst:
-            marker = ' <span style="color:#dc2626">▼</span>'
-        parts.append(
-            f'<tr>'
-            f'<td style="padding:.4rem .6rem;font-family:monospace">'
-            f'{_e(str(value))}{marker}</td>'
-            f'{mean_cell}'
-            f'<td style="padding:.4rem .6rem;text-align:right;'
-            f'font-family:monospace">{stdev_str}</td>'
-            f'<td style="padding:.4rem .6rem;text-align:right;'
-            f'font-family:monospace">{rank_str}</td>'
-            f'<td style="padding:.4rem .6rem;text-align:right;'
-            f'font-family:monospace">{n_obs}</td>'
-            f'</tr>'
-        )
-    parts.append("</tbody></table></section>")
-    return "".join(parts)
+import warnings
 
+from picarones.reports_v2.html.renderers.incremental_comparison import *  # noqa: F401, F403
 
-__all__ = ["build_incremental_comparison_html"]
+warnings.warn(
+    "picarones.report.incremental_comparison_render is deprecated and will be removed in 2.0.  "
+    "Import from picarones.reports_v2.html.renderers.incremental_comparison instead.",
+    DeprecationWarning,
+    stacklevel=2,
+)

picarones/report/module_audit_render.py

@@ -1,173 +1,18 @@
-"""Rendu HTML « Modules audités » — Sprint 97 (B.6).
+"""``picarones.report.module_audit_render`` — shim re-export (déprécié, suppression 2.0).
 
-Suite directe ``picarones/core/module_policy.py``.  Pattern
-identique aux autres rendus : server-side, pas de JS, anti-
-injection systématique.
-
-Vue
----
-Tableau récapitulatif des modules utilisés dans une pipeline
-composée, chacun avec :
-
-- Statut d'audit (✓ vert si tous les checks passent, ✗ rouge
-  sinon, avec compte des échecs) ;
-- Métadonnées : version, auteur, licence ;
-- Citation académique si fournie ;
-- Lien vers la homepage si fourni.
-
-Adaptive : ``""`` si la liste est vide.
-
-Note d'intégration
-------------------
-Module pur — l'utilisateur compose la liste depuis sa
-``PipelineSpec`` augmentée des ``ModuleManifest`` :
-
-.. code-block:: python
-
-    from picarones.measurements.module_policy import audit_module
-    from picarones.report.module_audit_render import build_module_audit_html
-
-    audits = []
-    for step in pipeline.steps:
-        manifest = step.module.manifest  # convention applicative
-        result = audit_module(step.module, manifest)
-        audits.append({
-            "manifest": manifest.as_dict(),
-            "audit": result.as_dict(),
-        })
-    html = build_module_audit_html(audits, labels)
+Canonique : :mod:`picarones.reports_v2.html.renderers.module_audit`.
+Phase 5.C du retrait du legacy.
 """
 
 from __future__ import annotations
 
-from html import escape as _e
-from typing import Optional
-
-
-def _passed_badge(passed: bool, n_failed: int, label_pass: str,
-                  label_fail: str) -> str:
-    if passed:
-        return (
-            f'<span style="color:#16a34a;font-weight:700">'
-            f'✓ {_e(label_pass)}</span>'
-        )
-    return (
-        f'<span style="color:#dc2626;font-weight:700">'
-        f'✗ {_e(label_fail)} ({n_failed})</span>'
-    )
-
-
-def build_module_audit_html(
-    audits: Optional[list],
-    labels: Optional[dict[str, str]] = None,
-) -> str:
-    """Construit la vue HTML « Modules audités ».
-
-    Parameters
-    ----------
-    audits:
-        Liste de dicts ``{"manifest": ManifestDict, "audit":
-        AuditResultDict}``.  Si vide ou ``None``, retourne ``""``.
-    labels:
-        Dict i18n.  Clés sous le préfixe ``audit_*``.
-    """
-    if not audits:
-        return ""
-    rows = [
-        a for a in audits
-        if isinstance(a, dict)
-        and isinstance(a.get("manifest"), dict)
-        and isinstance(a.get("audit"), dict)
-    ]
-    if not rows:
-        return ""
-    labels = labels or {}
-    title = labels.get("audit_title", "Modules audités")
-    note = labels.get(
-        "audit_note",
-        "Récapitulatif des modules utilisés dans la pipeline "
-        "composée. Un module qui ne passe pas l'audit n'est "
-        "pas exécutable. Métadonnées issues du manifest fourni "
-        "par le contributeur (auteur, licence, citation).",
-    )
-    label_pass = labels.get("audit_pass", "audit OK")
-    label_fail = labels.get("audit_fail", "checks échoués")
-    h_module = labels.get("audit_module", "Module")
-    h_status = labels.get("audit_status", "Audit")
-    h_version = labels.get("audit_version", "Version")
-    h_author = labels.get("audit_author", "Auteur")
-    h_license = labels.get("audit_license", "Licence")
-    h_io = labels.get("audit_io", "Entrée → sortie")
-    h_citation = labels.get("audit_citation", "Citation")
-    h_homepage = labels.get("audit_homepage", "Page projet")
-
-    parts = [
-        '<section class="audit-section" style="margin:1rem 0">',
-        f'<h3 style="margin:0 0 .3rem 0">{_e(title)}</h3>',
-        f'<div style="font-size:.85rem;opacity:.75;margin-bottom:.5rem">'
-        f'{_e(note)}</div>',
-        '<table style="border-collapse:collapse;width:100%;'
-        'font-size:.9rem">',
-        '<thead><tr>',
-    ]
-    for col in (h_module, h_status, h_version, h_author,
-                h_license, h_io, h_citation, h_homepage):
-        parts.append(
-            f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:left;'
-            f'border-bottom:1px solid #ccc;font-weight:600">'
-            f'{_e(col)}</th>'
-        )
-    parts.append("</tr></thead><tbody>")
-
-    for entry in rows:
-        manifest = entry["manifest"]
-        audit = entry["audit"]
-        name = str(manifest.get("name") or "?")
-        version = str(manifest.get("version") or "—")
-        author = str(manifest.get("author") or "—")
-        license_ = str(manifest.get("license") or "—")
-        in_types = ", ".join(manifest.get("input_types") or []) or "—"
-        out_types = ", ".join(manifest.get("output_types") or []) or "—"
-        citation = manifest.get("citation") or ""
-        homepage = manifest.get("homepage") or ""
-        passed = bool(audit.get("passed"))
-        n_failed = int(audit.get("n_failed") or 0)
-        status_cell = _passed_badge(
-            passed, n_failed, label_pass, label_fail,
-        )
-        # Citation : tronqué si trop long
-        citation_str = str(citation)[:120]
-        if len(str(citation)) > 120:
-            citation_str += "…"
-        citation_cell = (
-            _e(citation_str) if citation_str.strip() else "—"
-        )
-        # Homepage : on n'auto-link **pas** (anti-injection +
-        # honnêteté : l'URL peut pointer ailleurs).  On affiche
-        # le texte échappé tel quel.
-        homepage_cell = (
-            _e(str(homepage))[:80] + ("…" if len(str(homepage)) > 80 else "")
-        ) if str(homepage).strip() else "—"
-        parts.append(
-            f'<tr>'
-            f'<td style="padding:.4rem .6rem;font-family:monospace">'
-            f'{_e(name)}</td>'
-            f'<td style="padding:.4rem .6rem">{status_cell}</td>'
-            f'<td style="padding:.4rem .6rem;font-family:monospace">'
-            f'{_e(version)}</td>'
-            f'<td style="padding:.4rem .6rem">{_e(author)}</td>'
-            f'<td style="padding:.4rem .6rem;font-family:monospace">'
-            f'{_e(license_)}</td>'
-            f'<td style="padding:.4rem .6rem;font-family:monospace;'
-            f'font-size:.8rem">{_e(in_types)} → {_e(out_types)}</td>'
-            f'<td style="padding:.4rem .6rem;font-size:.8rem;'
-            f'opacity:.85">{citation_cell}</td>'
-            f'<td style="padding:.4rem .6rem;font-family:monospace;'
-            f'font-size:.8rem">{homepage_cell}</td>'
-            f'</tr>'
-        )
-    parts.append("</tbody></table></section>")
-    return "".join(parts)
+import warnings
 
+from picarones.reports_v2.html.renderers.module_audit import *  # noqa: F401, F403
 
-__all__ = ["build_module_audit_html"]
+warnings.warn(
+    "picarones.report.module_audit_render is deprecated and will be removed in 2.0.  "
+    "Import from picarones.reports_v2.html.renderers.module_audit instead.",
+    DeprecationWarning,
+    stacklevel=2,
+)

picarones/report/ner_render.py

@@ -1,222 +1,18 @@
-"""Rendu HTML server-side de la section NER (Sprint 41).
+"""``picarones.report.ner_render`` — shim re-export (déprécié, suppression 2.0).
 
-Suite directe des Sprints 38-40 : la couche de calcul, le backend
-extracteur et le câblage runner sont en place ; ce module produit les
-blocs HTML qui remontent ces données dans le rapport.
-
-- ``build_ner_summary_html`` — encart factuel par moteur : F1 global,
-  precision/recall, total entités, hallucinations, missed.
-- ``build_ner_per_category_html`` — table heatmap moteur × catégorie,
-  cellules colorées par F1 (rouge → vert).
-
-Principe — cohérent avec ``inter_engine_render`` (Sprint 37) : rendu
-server-side, pas de JavaScript, déterministe.  Si aucun moteur n'a de
-``aggregated_ner``, les fonctions retournent une chaîne vide — la vue
-est silencieusement omise (rapport adaptatif).
-
-Anti-injection : tous les noms de moteurs et catégories sont passés à
-``html.escape`` avant insertion.
+Canonique : :mod:`picarones.reports_v2.html.renderers.ner`.  Phase
+5.C du retrait du legacy.
 """
 
 from __future__ import annotations
 
-from html import escape as _e
-from typing import Optional
-
-from picarones.reports_v2._helpers.render_helpers import color_traffic_light
-
-
-def _engines_with_ner(engines_summary: list[dict]) -> list[dict]:
-    """Filtre les moteurs qui ont une analyse NER agrégée."""
-    return [e for e in engines_summary if e.get("aggregated_ner")]
-
-
-def build_ner_summary_html(
-    engines_summary: list[dict],
-    labels: Optional[dict[str, str]] = None,
-) -> str:
-    """Construit l'encart résumé NER : F1 global par moteur + totaux.
-
-    Parameters
-    ----------
-    engines_summary:
-        Liste de dicts moteur (au moins ``name`` et ``aggregated_ner``).
-    labels:
-        Dict d'étiquettes i18n.
-
-    Returns
-    -------
-    str
-        HTML ``<div>...</div>`` ou ``""`` si aucun moteur n'a de NER.
-    """
-    relevant = _engines_with_ner(engines_summary)
-    if not relevant:
-        return ""
-
-    labels = labels or {}
-    caption = labels.get("ner_summary_caption", "Précision sur entités nommées")
-    engine_label = labels.get("ner_engine_label", "Moteur")
-    f1_label = labels.get("ner_f1_label", "F1 global")
-    p_label = labels.get("ner_precision_label", "Précision")
-    r_label = labels.get("ner_recall_label", "Rappel")
-    docs_label = labels.get("ner_doc_count_label", "Docs évalués")
-    halluc_label = labels.get("ner_hallucinated_label", "Hallucinations")
-    missed_label = labels.get("ner_missed_label", "Entités manquées")
-
-    parts: list[str] = []
-    parts.append('<div class="ner-summary">')
-    parts.append(
-        f'<div class="ner-summary-caption" style="font-weight:600;'
-        f'margin-bottom:.4rem">{_e(caption)}</div>'
-    )
-    parts.append(
-        '<table class="ner-summary-table" '
-        'style="border-collapse:collapse;font-size:.85rem;width:100%">'
-    )
-    parts.append("<thead><tr>")
-    for hdr in (engine_label, f1_label, p_label, r_label,
-                docs_label, halluc_label, missed_label):
-        parts.append(
-            f'<th scope=\"col\" style="padding:.3rem .5rem;text-align:left;'
-            f'border-bottom:1px solid var(--border);font-weight:600">'
-            f'{_e(hdr)}</th>'
-        )
-    parts.append("</tr></thead><tbody>")
-    for engine in relevant:
-        agg = engine["aggregated_ner"]
-        global_stats = agg.get("global", {}) or {}
-        f1 = float(global_stats.get("f1") or 0.0)
-        precision = float(global_stats.get("precision") or 0.0)
-        recall = float(global_stats.get("recall") or 0.0)
-        doc_count = int(agg.get("doc_count") or 0)
-        hallucinated = int(agg.get("hallucinated_total") or 0)
-        missed = int(agg.get("missed_total") or 0)
-        bg = color_traffic_light(f1)
-        parts.append("<tr>")
-        parts.append(
-            f'<td style="padding:.3rem .5rem;font-weight:600">'
-            f'{_e(engine.get("name", ""))}</td>'
-        )
-        parts.append(
-            f'<td style="padding:.3rem .5rem;background:{bg};'
-            f'font-variant-numeric:tabular-nums">{f1 * 100:.1f} %</td>'
-        )
-        parts.append(
-            f'<td style="padding:.3rem .5rem;font-variant-numeric:tabular-nums">'
-            f'{precision * 100:.1f} %</td>'
-        )
-        parts.append(
-            f'<td style="padding:.3rem .5rem;font-variant-numeric:tabular-nums">'
-            f'{recall * 100:.1f} %</td>'
-        )
-        parts.append(
-            f'<td style="padding:.3rem .5rem;font-variant-numeric:tabular-nums">'
-            f'{doc_count}</td>'
-        )
-        parts.append(
-            f'<td style="padding:.3rem .5rem;font-variant-numeric:tabular-nums">'
-            f'{hallucinated}</td>'
-        )
-        parts.append(
-            f'<td style="padding:.3rem .5rem;font-variant-numeric:tabular-nums">'
-            f'{missed}</td>'
-        )
-        parts.append("</tr>")
-    parts.append("</tbody></table></div>")
-    return "".join(parts)
-
-
-def build_ner_per_category_html(
-    engines_summary: list[dict],
-    labels: Optional[dict[str, str]] = None,
-) -> str:
-    """Construit la heatmap NER moteur × catégorie d'entité.
-
-    Lignes = moteurs, colonnes = catégories (PER, LOC, ORG, DATE,
-    MISC…).  Cellules colorées par F1 (rouge → vert).  La cellule
-    affiche le F1 en pourcentage.  Cellules vides quand la catégorie
-    n'a pas été observée pour le moteur.
-
-    Returns
-    -------
-    str
-        HTML ``<div>...</div>`` ou ``""`` si pas de données.
-    """
-    relevant = _engines_with_ner(engines_summary)
-    if not relevant:
-        return ""
-
-    # Catégories : union sur tous les moteurs, ordre alphabétique
-    all_categories: set[str] = set()
-    for engine in relevant:
-        per_cat = (engine["aggregated_ner"] or {}).get("per_category") or {}
-        all_categories.update(per_cat.keys())
-    if not all_categories:
-        return ""
-    categories = sorted(all_categories)
-
-    labels = labels or {}
-    caption = labels.get(
-        "ner_per_category_caption",
-        "F1 par catégorie d'entité (heatmap)",
-    )
-    engine_label = labels.get("ner_engine_label", "Moteur")
-    no_data = labels.get("ner_no_data_label", "—")
-
-    parts: list[str] = []
-    parts.append('<div class="ner-per-category">')
-    parts.append(
-        f'<div class="ner-per-category-caption" '
-        f'style="font-weight:600;margin-bottom:.4rem">{_e(caption)}</div>'
-    )
-    parts.append(
-        '<table class="ner-per-category-table" '
-        'style="border-collapse:collapse;font-size:.8rem">'
-    )
-    parts.append("<thead><tr>")
-    parts.append(
-        f'<th scope=\"col\" style="padding:.3rem .5rem;text-align:left;'
-        f'border-bottom:1px solid var(--border)">{_e(engine_label)}</th>'
-    )
-    for cat in categories:
-        parts.append(
-            f'<th scope=\"col\" style="padding:.3rem .5rem;text-align:center;'
-            f'border-bottom:1px solid var(--border)">{_e(cat)}</th>'
-        )
-    parts.append("</tr></thead><tbody>")
-    for engine in relevant:
-        per_cat = (engine["aggregated_ner"] or {}).get("per_category") or {}
-        parts.append("<tr>")
-        parts.append(
-            f'<th scope=\"col\" style="padding:.3rem .5rem;text-align:right;'
-            f'border-right:1px solid var(--border);font-weight:600">'
-            f'{_e(engine.get("name", ""))}</th>'
-        )
-        for cat in categories:
-            stats = per_cat.get(cat)
-            if not stats or int(stats.get("support", 0)) == 0:
-                parts.append(
-                    f'<td style="padding:.3rem .5rem;text-align:center;'
-                    f'background:#f4f4f4;color:var(--text-muted);'
-                    f'font-style:italic">{_e(no_data)}</td>'
-                )
-            else:
-                f1 = float(stats.get("f1") or 0.0)
-                support = int(stats.get("support", 0))
-                bg = color_traffic_light(f1)
-                parts.append(
-                    f'<td style="padding:.3rem .5rem;text-align:center;'
-                    f'background:{bg};color:#222;'
-                    f'font-variant-numeric:tabular-nums" '
-                    f'title="support={support}">'
-                    f'{f1 * 100:.1f} %</td>'
-                )
-        parts.append("</tr>")
-    parts.append("</tbody></table></div>")
-    return "".join(parts)
+import warnings
 
+from picarones.reports_v2.html.renderers.ner import *  # noqa: F401, F403
 
-__all__ = [
-    "build_ner_summary_html",
-    "build_ner_per_category_html",
-]
+warnings.warn(
+    "picarones.report.ner_render is deprecated and will be removed in 2.0.  "
+    "Import from picarones.reports_v2.html.renderers.ner instead.",
+    DeprecationWarning,
+    stacklevel=2,
+)

picarones/report/views/diagnostics.py

@@ -144,7 +144,7 @@ def build_diagnostics_view_html(
             from picarones.measurements.image_predictive import (
                 aggregate_corpus_predictive,
             )
-            from picarones.report.image_predictive_render import (
+            from picarones.reports_v2.html.renderers.image_predictive import (
                 build_image_predictive_html,
             )
             aggregated = aggregate_corpus_predictive(image_qualities)

picarones/report/views/pipeline.py

@@ -148,7 +148,7 @@ def build_pipeline_view_html(
     # Sous-section 3 : absorption d'erreur par jonction
     if junctions:
         try:
-            from picarones.report.error_absorption_render import (
+            from picarones.reports_v2.html.renderers.error_absorption import (
                 build_error_absorption_html,
             )
             html = build_error_absorption_html(junctions, labels=labels)
@@ -171,7 +171,7 @@ def build_pipeline_view_html(
             from picarones.measurements.incremental_comparison import (
                 compare_isolated_effect,
             )
-            from picarones.report.incremental_comparison_render import (
+            from picarones.reports_v2.html.renderers.incremental_comparison import (
                 build_incremental_comparison_html,
             )
             comparison = compare_isolated_effect(
@@ -200,7 +200,7 @@ def build_pipeline_view_html(
     # Sous-section 5 : audit des modules contribués
     if module_audits:
         try:
-            from picarones.report.module_audit_render import (
+            from picarones.reports_v2.html.renderers.module_audit import (
                 build_module_audit_html,
             )
             html = build_module_audit_html(module_audits, labels=labels)

picarones/reports_v2/html/renderers/error_absorption.py

@@ -0,0 +1,216 @@
+"""Rendu HTML « Absorption d'erreur » — Sprint 94 (B.3).
+
+Phase 5.C — module relocalisé depuis
+``picarones.report.error_absorption_render`` vers
+``picarones.reports_v2.html.renderers.error_absorption``.  Le chemin
+legacy reste disponible via un shim avec ``DeprecationWarning`` ;
+suppression prévue en 2.0.
+
+Suite directe ``picarones/core/error_absorption.py``.  Pattern
+identique aux autres rendus : server-side, pas de JS, anti-
+injection systématique.
+
+Vue
+---
+Tableau résumé des jonctions du pipeline ; chaque ligne décrit
+un module post-correction et présente :
+
+- erreurs en entrée vs en sortie ;
+- nb corrigées (gradient vert), nb introduites (gradient rouge) ;
+- taux de correction (gradient vert), taux d'introduction
+  (gradient rouge) ;
+- amélioration nette (n_corrected - n_introduced) — coloré.
+- éventuellement un échantillon de tokens corrigés/introduits.
+
+Adaptive : ``""`` si la liste est vide.
+
+Note d'intégration
+------------------
+Module pur — la liste ``junctions`` est composée par
+l'utilisateur depuis son benchmark de pipeline composée :
+
+.. code-block:: python
+
+    from picarones.measurements.error_absorption import (
+        compute_error_absorption, aggregate_error_absorption,
+    )
+    from picarones.reports_v2.html.renderers.error_absorption import (
+        build_error_absorption_html,
+    )
+
+    junctions = []
+    for step in pipeline.steps_with_text_output:
+        per_doc = [
+            compute_error_absorption(doc.gt_text, doc.before_text,
+                                     doc.after_text)
+            for doc in benchmark.docs
+        ]
+        agg = aggregate_error_absorption(per_doc)
+        if agg is not None:
+            agg["junction_name"] = step.name
+            junctions.append(agg)
+    html = build_error_absorption_html(junctions, labels)
+"""
+
+from __future__ import annotations
+
+from html import escape as _e
+from typing import Optional
+
+from picarones.reports_v2._helpers.render_helpers import color_diverging, color_traffic_light
+
+
+# Palette « net improvement » : vert clair au centre, vert profond
+# si favorable (net > 0), rouge si défavorable (net < 0).  Centrée
+# sur le vert clair car un delta nul est déjà « pas de régression ».
+_NET_NEUTRAL_RGB = (167, 240, 167)
+_NET_POSITIVE_RGB = (90, 200, 90)
+_NET_NEGATIVE_RGB = (220, 50, 50)
+
+
+def build_error_absorption_html(
+    junctions: Optional[list],
+    labels: Optional[dict[str, str]] = None,
+    *,
+    sample_max: int = 8,
+) -> str:
+    """Construit la vue HTML « Absorption d'erreur ».
+
+    Parameters
+    ----------
+    junctions:
+        Liste de dicts (un par jonction de pipeline), enrichis
+        d'un ``junction_name``.  Si vide ou ``None``, retourne
+        ``""``.
+    labels:
+        Dict i18n.  Clés sous le préfixe ``absorption_*``.
+    sample_max:
+        Nombre maximal de tokens corrigés/introduits affichés
+        en cellule échantillon.
+    """
+    if not junctions:
+        return ""
+    rows = [
+        j for j in junctions
+        if isinstance(j, dict) and j.get("junction_name")
+    ]
+    if not rows:
+        return ""
+    labels = labels or {}
+    title = labels.get(
+        "absorption_title", "Absorption d'erreur par jonction",
+    )
+    note = labels.get(
+        "absorption_note",
+        "À chaque jonction du pipeline, deux flux sont mesurés "
+        "indépendamment : combien d'erreurs sont corrigées et "
+        "combien sont introduites. Une jonction qui corrige "
+        "beaucoup mais introduit aussi beaucoup absorbe les "
+        "différences amont au lieu de les améliorer.",
+    )
+    h_junction = labels.get("absorption_junction", "Jonction")
+    h_errors_before = labels.get("absorption_errors_before", "Erreurs avant")
+    h_errors_after = labels.get("absorption_errors_after", "Erreurs après")
+    h_corrected = labels.get("absorption_corrected", "Corrigées")
+    h_introduced = labels.get("absorption_introduced", "Introduites")
+    h_corr_rate = labels.get("absorption_corr_rate", "% corrigées")
+    h_intro_rate = labels.get("absorption_intro_rate", "% introduites")
+    h_net = labels.get("absorption_net", "Amélioration nette")
+    h_sample = labels.get("absorption_sample", "Échantillon (intro)")
+
+    # Saturation pour le gradient « net »
+    max_abs_net = max(
+        (abs(int(r.get("net_improvement") or 0)) for r in rows), default=1,
+    ) or 1
+
+    parts = [
+        '<section class="absorption-section" style="margin:1rem 0">',
+        f'<h3 style="margin:0 0 .3rem 0">{_e(title)}</h3>',
+        f'<div style="font-size:.85rem;opacity:.75;margin-bottom:.6rem">'
+        f'{_e(note)}</div>',
+        '<table style="border-collapse:collapse;width:100%;'
+        'font-size:.9rem">',
+        '<thead><tr>',
+    ]
+    for col in (h_junction, h_errors_before, h_errors_after,
+                h_corrected, h_introduced, h_corr_rate,
+                h_intro_rate, h_net, h_sample):
+        parts.append(
+            f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:left;'
+            f'border-bottom:1px solid #ccc;font-weight:600">'
+            f'{_e(col)}</th>'
+        )
+    parts.append("</tr></thead><tbody>")
+    for entry in rows:
+        name = str(entry.get("junction_name") or "?")
+        n_eb = int(entry.get("n_errors_before") or 0)
+        n_ea = int(entry.get("n_errors_after") or 0)
+        n_corr = int(entry.get("n_corrected") or 0)
+        n_intro = int(entry.get("n_introduced") or 0)
+        net = int(entry.get("net_improvement") or 0)
+        corr_rate = entry.get("correction_rate")
+        intro_rate = entry.get("introduction_rate")
+        if isinstance(corr_rate, (int, float)):
+            corr_rate_str = f"{corr_rate * 100:.1f}%"
+            corr_color = color_traffic_light(float(corr_rate))
+            corr_cell = (
+                f'<td style="padding:.4rem .6rem;text-align:right;'
+                f'background:{corr_color};font-family:monospace;'
+                f'font-weight:600">{corr_rate_str}</td>'
+            )
+        else:
+            corr_cell = (
+                '<td style="padding:.4rem .6rem;text-align:right;'
+                'opacity:.4">—</td>'
+            )
+        if isinstance(intro_rate, (int, float)):
+            intro_rate_str = f"{intro_rate * 100:.1f}%"
+            intro_color = color_traffic_light(float(intro_rate), low_is_good=True)
+            intro_cell = (
+                f'<td style="padding:.4rem .6rem;text-align:right;'
+                f'background:{intro_color};font-family:monospace;'
+                f'font-weight:600">{intro_rate_str}</td>'
+            )
+        else:
+            intro_cell = (
+                '<td style="padding:.4rem .6rem;text-align:right;'
+                'opacity:.4">—</td>'
+            )
+        net_color = color_diverging(
+            float(net),
+            max_abs=float(max_abs_net) if max_abs_net else 1.0,
+            neutral_rgb=_NET_NEUTRAL_RGB,
+            positive_rgb=_NET_POSITIVE_RGB,
+            negative_rgb=_NET_NEGATIVE_RGB,
+        )
+        intro_sample = entry.get("introduced_tokens_sample") or []
+        sample_cell_text = ", ".join(
+            _e(str(t)) for t in intro_sample[:sample_max]
+        ) or "—"
+        if len(intro_sample) > sample_max:
+            sample_cell_text += " …"
+        parts.append(
+            f'<tr>'
+            f'<td style="padding:.4rem .6rem">{_e(name)}</td>'
+            f'<td style="padding:.4rem .6rem;text-align:right;'
+            f'font-family:monospace">{n_eb}</td>'
+            f'<td style="padding:.4rem .6rem;text-align:right;'
+            f'font-family:monospace">{n_ea}</td>'
+            f'<td style="padding:.4rem .6rem;text-align:right;'
+            f'font-family:monospace">{n_corr}</td>'
+            f'<td style="padding:.4rem .6rem;text-align:right;'
+            f'font-family:monospace">{n_intro}</td>'
+            f'{corr_cell}'
+            f'{intro_cell}'
+            f'<td style="padding:.4rem .6rem;text-align:right;'
+            f'background:{net_color};font-family:monospace;'
+            f'font-weight:600">{net:+d}</td>'
+            f'<td style="padding:.4rem .6rem;font-family:monospace;'
+            f'font-size:.8rem">{sample_cell_text}</td>'
+            f'</tr>'
+        )
+    parts.append("</tbody></table></section>")
+    return "".join(parts)
+
+
+__all__ = ["build_error_absorption_html"]

picarones/reports_v2/html/renderers/image_predictive.py

@@ -0,0 +1,213 @@
+"""Rendu HTML « Profil d'image du corpus » — Sprint 93 (A.II.7).
+
+Phase 5.C — module relocalisé depuis
+``picarones.report.image_predictive_render`` vers
+``picarones.reports_v2.html.renderers.image_predictive``.  Le chemin
+legacy reste disponible via un shim avec ``DeprecationWarning`` ;
+suppression prévue en 2.0.
+
+Suite directe ``picarones/core/image_predictive.py``.  Pattern
+identique aux autres rendus : server-side, pas de JS, anti-
+injection systématique.
+
+Vue
+---
+Deux blocs dans une section unique :
+
+1. **Complexité paléographique** : moyenne, médiane, min, max,
+   écart-type sur l'ensemble du corpus.
+2. **Homogénéité du corpus** : score combiné + détail par
+   feature (mean, stdev, contribution normalisée).
+
+Adaptive : ``""`` si pas de données.
+
+Note d'intégration
+------------------
+Module pur — l'utilisateur compose :
+
+.. code-block:: python
+
+    from picarones.measurements.image_predictive import aggregate_corpus_predictive
+    from picarones.reports_v2.html.renderers.image_predictive import (
+        build_image_predictive_html,
+    )
+
+    qualities = [doc.image_quality.as_dict() for doc in benchmark.docs]
+    agg = aggregate_corpus_predictive(qualities)
+    html = build_image_predictive_html(agg, labels)
+"""
+
+from __future__ import annotations
+
+from html import escape as _e
+from typing import Optional
+
+from picarones.reports_v2._helpers.render_helpers import color_traffic_light
+
+
+_FEATURE_LABEL_KEYS = {
+    "noise_level": "imgpred_feat_noise",
+    "sharpness_score": "imgpred_feat_sharpness",
+    "contrast_score": "imgpred_feat_contrast",
+    "rotation_degrees": "imgpred_feat_rotation",
+}
+
+
+def _render_complexity_block(
+    aggregated: dict, labels: dict[str, str],
+) -> str:
+    h_complex = labels.get(
+        "imgpred_complexity", "Complexité paléographique",
+    )
+    h_mean = labels.get("imgpred_mean", "Moyenne")
+    h_median = labels.get("imgpred_median", "Médiane")
+    h_min = labels.get("imgpred_min", "Min")
+    h_max = labels.get("imgpred_max", "Max")
+    h_stdev = labels.get("imgpred_stdev", "Écart-type")
+    h_docs = labels.get("imgpred_docs", "Docs")
+    mean = float(aggregated.get("complexity_mean") or 0.0)
+    median = float(aggregated.get("complexity_median") or 0.0)
+    mn = float(aggregated.get("complexity_min") or 0.0)
+    mx = float(aggregated.get("complexity_max") or 0.0)
+    sd = float(aggregated.get("complexity_stdev") or 0.0)
+    n_docs = int(aggregated.get("n_docs") or 0)
+    color_mean = color_traffic_light(mean, low_is_good=True)
+    return (
+        f'<div style="font-weight:600;margin:.4rem 0 .3rem 0">'
+        f'{_e(h_complex)}</div>'
+        '<table style="border-collapse:collapse;width:100%;'
+        'font-size:.9rem;margin-bottom:.8rem">'
+        f'<thead><tr>'
+        f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:right;'
+        f'border-bottom:1px solid #ccc;font-weight:600">{_e(h_mean)}</th>'
+        f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:right;'
+        f'border-bottom:1px solid #ccc;font-weight:600">{_e(h_median)}</th>'
+        f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:right;'
+        f'border-bottom:1px solid #ccc;font-weight:600">{_e(h_min)}</th>'
+        f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:right;'
+        f'border-bottom:1px solid #ccc;font-weight:600">{_e(h_max)}</th>'
+        f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:right;'
+        f'border-bottom:1px solid #ccc;font-weight:600">{_e(h_stdev)}</th>'
+        f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:right;'
+        f'border-bottom:1px solid #ccc;font-weight:600">{_e(h_docs)}</th>'
+        f'</tr></thead>'
+        f'<tbody><tr>'
+        f'<td style="padding:.4rem .6rem;text-align:right;'
+        f'background:{color_mean};font-family:monospace;font-weight:600">'
+        f'{mean:.3f}</td>'
+        f'<td style="padding:.4rem .6rem;text-align:right;'
+        f'font-family:monospace">{median:.3f}</td>'
+        f'<td style="padding:.4rem .6rem;text-align:right;'
+        f'font-family:monospace">{mn:.3f}</td>'
+        f'<td style="padding:.4rem .6rem;text-align:right;'
+        f'font-family:monospace">{mx:.3f}</td>'
+        f'<td style="padding:.4rem .6rem;text-align:right;'
+        f'font-family:monospace">{sd:.3f}</td>'
+        f'<td style="padding:.4rem .6rem;text-align:right;'
+        f'font-family:monospace">{n_docs}</td>'
+        f'</tr></tbody></table>'
+    )
+
+
+def _render_homogeneity_block(
+    homogeneity: dict, labels: dict[str, str],
+) -> str:
+    h_homo = labels.get(
+        "imgpred_homogeneity", "Homogénéité du corpus",
+    )
+    h_feat = labels.get("imgpred_feature", "Feature")
+    h_mean = labels.get("imgpred_feat_mean", "Moyenne")
+    h_stdev = labels.get("imgpred_feat_stdev", "Écart-type")
+    h_norm = labels.get(
+        "imgpred_feat_norm", "Contribution normalisée",
+    )
+    score = float(homogeneity.get("score") or 0.0)
+    color = color_traffic_light(score, low_is_good=True)
+    parts = [
+        f'<div style="font-weight:600;margin:.4rem 0 .3rem 0">'
+        f'{_e(h_homo)} : '
+        f'<span style="background:{color};padding:.1rem .4rem;'
+        f'border-radius:.3rem;font-family:monospace">{score:.3f}</span>'
+        f'</div>',
+        '<table style="border-collapse:collapse;width:100%;'
+        'font-size:.9rem">',
+        '<thead><tr>',
+    ]
+    for col in (h_feat, h_mean, h_stdev, h_norm):
+        parts.append(
+            f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:left;'
+            f'border-bottom:1px solid #ccc;font-weight:600">'
+            f'{_e(col)}</th>'
+        )
+    parts.append("</tr></thead><tbody>")
+    per_feat = homogeneity.get("per_feature") or {}
+    for key, label_key in _FEATURE_LABEL_KEYS.items():
+        if key not in per_feat:
+            continue
+        slot = per_feat[key]
+        feat_label = labels.get(label_key, key)
+        feat_mean = float(slot.get("mean") or 0.0)
+        feat_stdev = float(slot.get("stdev") or 0.0)
+        feat_norm = float(slot.get("normalised") or 0.0)
+        norm_color = color_traffic_light(feat_norm, low_is_good=True)
+        parts.append(
+            f'<tr>'
+            f'<td style="padding:.4rem .6rem">{_e(feat_label)}</td>'
+            f'<td style="padding:.4rem .6rem;text-align:right;'
+            f'font-family:monospace">{feat_mean:.3f}</td>'
+            f'<td style="padding:.4rem .6rem;text-align:right;'
+            f'font-family:monospace">{feat_stdev:.3f}</td>'
+            f'<td style="padding:.4rem .6rem;text-align:right;'
+            f'background:{norm_color};font-family:monospace">'
+            f'{feat_norm:.3f}</td>'
+            f'</tr>'
+        )
+    parts.append("</tbody></table>")
+    return "".join(parts)
+
+
+def build_image_predictive_html(
+    aggregated: Optional[dict],
+    labels: Optional[dict[str, str]] = None,
+) -> str:
+    """Construit la vue HTML « Profil d'image du corpus ».
+
+    Parameters
+    ----------
+    aggregated:
+        Sortie de ``aggregate_corpus_predictive``.  Si ``None``
+        ou ``n_docs == 0``, retourne ``""``.
+    labels:
+        Dict i18n.  Clés sous le préfixe ``imgpred_*``.
+    """
+    if not aggregated:
+        return ""
+    if not aggregated.get("n_docs"):
+        return ""
+    labels = labels or {}
+    title = labels.get(
+        "imgpred_title", "Profil d'image du corpus",
+    )
+    note = labels.get(
+        "imgpred_note",
+        "Score de complexité paléographique combinant bruit, "
+        "flou, faible contraste et rotation. Le score "
+        "d'homogénéité signale si la moyenne globale est fiable "
+        "(corpus uniforme) ou trompeuse (corpus hétérogène — "
+        "voir alors la vue stratifiée).",
+    )
+    parts = [
+        '<section class="imgpred-section" style="margin:1rem 0">',
+        f'<h3 style="margin:0 0 .3rem 0">{_e(title)}</h3>',
+        f'<div style="font-size:.85rem;opacity:.75;margin-bottom:.6rem">'
+        f'{_e(note)}</div>',
+    ]
+    parts.append(_render_complexity_block(aggregated, labels))
+    homo = aggregated.get("homogeneity")
+    if isinstance(homo, dict):
+        parts.append(_render_homogeneity_block(homo, labels))
+    parts.append("</section>")
+    return "".join(parts)
+
+
+__all__ = ["build_image_predictive_html"]

picarones/reports_v2/html/renderers/incremental_comparison.py

@@ -0,0 +1,207 @@
+"""Rendu HTML « Comparaison contrôlée » — Sprint 96 (B.5).
+
+Phase 5.C — module relocalisé depuis
+``picarones.report.incremental_comparison_render`` vers
+``picarones.reports_v2.html.renderers.incremental_comparison``.
+Le chemin legacy reste disponible via un shim avec
+``DeprecationWarning`` ; suppression prévue en 2.0.
+
+Suite directe ``picarones/core/incremental_comparison.py``.
+Pattern identique aux autres rendus : server-side, pas de JS,
+anti-injection systématique.
+
+Vue
+---
+Tableau ANOVA-like : pour chaque valeur du slot variant, mean
+± stdev, rang moyen, n_observations.  Mean colorée en
+gradient vert (meilleur) → rouge (pire), normalisée sur la
+plage des moyennes observées.
+
+Adaptive : ``""`` si ``analysis`` est ``None``.
+
+Note d'intégration
+------------------
+Module pur — l'utilisateur compose :
+
+.. code-block:: python
+
+    from picarones.measurements.incremental_comparison import (
+        PipelineRun, compare_isolated_effect,
+    )
+    from picarones.reports_v2.html.renderers.incremental_comparison import (
+        build_incremental_comparison_html,
+    )
+
+    runs = [
+        PipelineRun(name=p.name,
+                    slots={"ocr": p.ocr, "llm": p.llm},
+                    score=p.cer_mean)
+        for p in benchmark.pipelines
+    ]
+    analysis = compare_isolated_effect(runs, "llm")
+    html = build_incremental_comparison_html(analysis, labels)
+"""
+
+from __future__ import annotations
+
+from html import escape as _e
+from typing import Optional
+
+from picarones.reports_v2._helpers.render_helpers import color_traffic_light
+
+
+def _bg_for_relative_score(
+    score: float, low: float, high: float, higher_is_better: bool,
+) -> str:
+    """Mappe ``score`` sur une plage [low, high] et retourne une cellule
+    colorée traffic-light.
+
+    Si ``higher_is_better=True``, ``score=high`` est vert ; sinon
+    ``score=low`` est vert.
+    """
+    if high == low:
+        return color_traffic_light(1.0)  # neutre vert clair
+    return color_traffic_light(
+        score,
+        low_is_good=not higher_is_better,
+        scale_min=low,
+        scale_max=high,
+    )
+
+
+def _format_score(value: Optional[float]) -> str:
+    if value is None:
+        return "—"
+    if abs(value) < 1.0:
+        return f"{value * 100:.2f}%"
+    return f"{value:.3f}"
+
+
+def build_incremental_comparison_html(
+    analysis: Optional[dict],
+    labels: Optional[dict[str, str]] = None,
+) -> str:
+    """Construit la vue HTML « Comparaison contrôlée ».
+
+    Parameters
+    ----------
+    analysis:
+        Sortie de ``compare_isolated_effect``.  ``None`` ou
+        ``per_value`` vide → retourne ``""``.
+    labels:
+        Dict i18n.  Clés sous le préfixe ``incr_*``.
+    """
+    if not analysis:
+        return ""
+    per_value = analysis.get("per_value") or {}
+    if not per_value:
+        return ""
+    labels = labels or {}
+    title = labels.get(
+        "incr_title", "Comparaison contrôlée par slot",
+    )
+    note = labels.get(
+        "incr_note",
+        "Effet isolé du module variant sur les pipelines en "
+        "contrôlant les autres slots. Pour chaque valeur du "
+        "slot, moyenne ± écart-type, rang moyen sur les groupes "
+        "fixes, et nombre d'observations. Type design "
+        "d'expérience pour des comparaisons honnêtes.",
+    )
+    slot_label = labels.get("incr_slot_label", "Slot variant")
+    h_value = labels.get("incr_value", "Valeur")
+    h_mean = labels.get("incr_mean", "Score moyen")
+    h_stdev = labels.get("incr_stdev", "± σ")
+    h_rank = labels.get("incr_rank", "Rang moyen")
+    h_n_obs = labels.get("incr_n_obs", "Observations")
+    h_groups = labels.get("incr_groups", "Groupes fixes")
+    higher_is_better = bool(analysis.get("higher_is_better", False))
+
+    # Plage de moyennes pour le code couleur
+    means = [
+        d["mean"] for d in per_value.values() if d.get("mean") is not None
+    ]
+    low = min(means) if means else 0.0
+    high = max(means) if means else 0.0
+
+    varying_slot = str(analysis.get("varying_slot") or "?")
+    n_groups = int(analysis.get("n_groups") or 0)
+    n_runs = int(analysis.get("n_runs") or 0)
+    best = analysis.get("best_value")
+    worst = analysis.get("worst_value")
+
+    parts = [
+        '<section class="incr-section" style="margin:1rem 0">',
+        f'<h3 style="margin:0 0 .3rem 0">{_e(title)}</h3>',
+        f'<div style="font-size:.85rem;opacity:.75;margin-bottom:.5rem">'
+        f'{_e(note)}</div>',
+        f'<div style="font-size:.85rem;margin-bottom:.5rem">'
+        f'<strong>{_e(slot_label)} :</strong> '
+        f'<code>{_e(varying_slot)}</code> &nbsp; '
+        f'<span style="opacity:.75">'
+        f'{n_runs} runs, {n_groups} {_e(h_groups.lower())}'
+        f'</span></div>',
+        '<table style="border-collapse:collapse;width:100%;'
+        'font-size:.9rem">',
+        '<thead><tr>',
+    ]
+    for col in (h_value, h_mean, h_stdev, h_rank, h_n_obs):
+        parts.append(
+            f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:left;'
+            f'border-bottom:1px solid #ccc;font-weight:600">'
+            f'{_e(col)}</th>'
+        )
+    parts.append("</tr></thead><tbody>")
+
+    # Tri par rang moyen ascendant
+    rows = sorted(
+        per_value.items(),
+        key=lambda kv: (kv[1].get("mean_rank") or float("inf")),
+    )
+    for value, d in rows:
+        mean = d.get("mean")
+        stdev = d.get("stdev")
+        rank = d.get("mean_rank")
+        n_obs = int(d.get("n_observations") or 0)
+        if isinstance(mean, (int, float)):
+            color = _bg_for_relative_score(
+                float(mean), low, high, higher_is_better,
+            )
+            mean_cell = (
+                f'<td style="padding:.4rem .6rem;text-align:right;'
+                f'background:{color};font-family:monospace;'
+                f'font-weight:600">{_format_score(mean)}</td>'
+            )
+        else:
+            mean_cell = (
+                '<td style="padding:.4rem .6rem;text-align:right;'
+                'opacity:.4">—</td>'
+            )
+        stdev_str = (
+            f"± {_format_score(stdev)}"
+            if isinstance(stdev, (int, float)) else "—"
+        )
+        rank_str = f"{rank:.2f}" if isinstance(rank, (int, float)) else "—"
+        marker = ""
+        if value == best:
+            marker = ' <span style="color:#16a34a">★</span>'
+        elif value == worst:
+            marker = ' <span style="color:#dc2626">▼</span>'
+        parts.append(
+            f'<tr>'
+            f'<td style="padding:.4rem .6rem;font-family:monospace">'
+            f'{_e(str(value))}{marker}</td>'
+            f'{mean_cell}'
+            f'<td style="padding:.4rem .6rem;text-align:right;'
+            f'font-family:monospace">{stdev_str}</td>'
+            f'<td style="padding:.4rem .6rem;text-align:right;'
+            f'font-family:monospace">{rank_str}</td>'
+            f'<td style="padding:.4rem .6rem;text-align:right;'
+            f'font-family:monospace">{n_obs}</td>'
+            f'</tr>'
+        )
+    parts.append("</tbody></table></section>")
+    return "".join(parts)
+
+
+__all__ = ["build_incremental_comparison_html"]

picarones/reports_v2/html/renderers/module_audit.py

@@ -0,0 +1,179 @@
+"""Rendu HTML « Modules audités » — Sprint 97 (B.6).
+
+Phase 5.C — module relocalisé depuis
+``picarones.report.module_audit_render`` vers
+``picarones.reports_v2.html.renderers.module_audit``.  Le chemin
+legacy reste disponible via un shim avec ``DeprecationWarning`` ;
+suppression prévue en 2.0.
+
+Suite directe ``picarones/core/module_policy.py``.  Pattern
+identique aux autres rendus : server-side, pas de JS, anti-
+injection systématique.
+
+Vue
+---
+Tableau récapitulatif des modules utilisés dans une pipeline
+composée, chacun avec :
+
+- Statut d'audit (✓ vert si tous les checks passent, ✗ rouge
+  sinon, avec compte des échecs) ;
+- Métadonnées : version, auteur, licence ;
+- Citation académique si fournie ;
+- Lien vers la homepage si fourni.
+
+Adaptive : ``""`` si la liste est vide.
+
+Note d'intégration
+------------------
+Module pur — l'utilisateur compose la liste depuis sa
+``PipelineSpec`` augmentée des ``ModuleManifest`` :
+
+.. code-block:: python
+
+    from picarones.measurements.module_policy import audit_module
+    from picarones.reports_v2.html.renderers.module_audit import build_module_audit_html
+
+    audits = []
+    for step in pipeline.steps:
+        manifest = step.module.manifest  # convention applicative
+        result = audit_module(step.module, manifest)
+        audits.append({
+            "manifest": manifest.as_dict(),
+            "audit": result.as_dict(),
+        })
+    html = build_module_audit_html(audits, labels)
+"""
+
+from __future__ import annotations
+
+from html import escape as _e
+from typing import Optional
+
+
+def _passed_badge(passed: bool, n_failed: int, label_pass: str,
+                  label_fail: str) -> str:
+    if passed:
+        return (
+            f'<span style="color:#16a34a;font-weight:700">'
+            f'✓ {_e(label_pass)}</span>'
+        )
+    return (
+        f'<span style="color:#dc2626;font-weight:700">'
+        f'✗ {_e(label_fail)} ({n_failed})</span>'
+    )
+
+
+def build_module_audit_html(
+    audits: Optional[list],
+    labels: Optional[dict[str, str]] = None,
+) -> str:
+    """Construit la vue HTML « Modules audités ».
+
+    Parameters
+    ----------
+    audits:
+        Liste de dicts ``{"manifest": ManifestDict, "audit":
+        AuditResultDict}``.  Si vide ou ``None``, retourne ``""``.
+    labels:
+        Dict i18n.  Clés sous le préfixe ``audit_*``.
+    """
+    if not audits:
+        return ""
+    rows = [
+        a for a in audits
+        if isinstance(a, dict)
+        and isinstance(a.get("manifest"), dict)
+        and isinstance(a.get("audit"), dict)
+    ]
+    if not rows:
+        return ""
+    labels = labels or {}
+    title = labels.get("audit_title", "Modules audités")
+    note = labels.get(
+        "audit_note",
+        "Récapitulatif des modules utilisés dans la pipeline "
+        "composée. Un module qui ne passe pas l'audit n'est "
+        "pas exécutable. Métadonnées issues du manifest fourni "
+        "par le contributeur (auteur, licence, citation).",
+    )
+    label_pass = labels.get("audit_pass", "audit OK")
+    label_fail = labels.get("audit_fail", "checks échoués")
+    h_module = labels.get("audit_module", "Module")
+    h_status = labels.get("audit_status", "Audit")
+    h_version = labels.get("audit_version", "Version")
+    h_author = labels.get("audit_author", "Auteur")
+    h_license = labels.get("audit_license", "Licence")
+    h_io = labels.get("audit_io", "Entrée → sortie")
+    h_citation = labels.get("audit_citation", "Citation")
+    h_homepage = labels.get("audit_homepage", "Page projet")
+
+    parts = [
+        '<section class="audit-section" style="margin:1rem 0">',
+        f'<h3 style="margin:0 0 .3rem 0">{_e(title)}</h3>',
+        f'<div style="font-size:.85rem;opacity:.75;margin-bottom:.5rem">'
+        f'{_e(note)}</div>',
+        '<table style="border-collapse:collapse;width:100%;'
+        'font-size:.9rem">',
+        '<thead><tr>',
+    ]
+    for col in (h_module, h_status, h_version, h_author,
+                h_license, h_io, h_citation, h_homepage):
+        parts.append(
+            f'<th scope=\"col\" style="padding:.4rem .6rem;text-align:left;'
+            f'border-bottom:1px solid #ccc;font-weight:600">'
+            f'{_e(col)}</th>'
+        )
+    parts.append("</tr></thead><tbody>")
+
+    for entry in rows:
+        manifest = entry["manifest"]
+        audit = entry["audit"]
+        name = str(manifest.get("name") or "?")
+        version = str(manifest.get("version") or "—")
+        author = str(manifest.get("author") or "—")
+        license_ = str(manifest.get("license") or "—")
+        in_types = ", ".join(manifest.get("input_types") or []) or "—"
+        out_types = ", ".join(manifest.get("output_types") or []) or "—"
+        citation = manifest.get("citation") or ""
+        homepage = manifest.get("homepage") or ""
+        passed = bool(audit.get("passed"))
+        n_failed = int(audit.get("n_failed") or 0)
+        status_cell = _passed_badge(
+            passed, n_failed, label_pass, label_fail,
+        )
+        # Citation : tronqué si trop long
+        citation_str = str(citation)[:120]
+        if len(str(citation)) > 120:
+            citation_str += "…"
+        citation_cell = (
+            _e(citation_str) if citation_str.strip() else "—"
+        )
+        # Homepage : on n'auto-link **pas** (anti-injection +
+        # honnêteté : l'URL peut pointer ailleurs).  On affiche
+        # le texte échappé tel quel.
+        homepage_cell = (
+            _e(str(homepage))[:80] + ("…" if len(str(homepage)) > 80 else "")
+        ) if str(homepage).strip() else "—"
+        parts.append(
+            f'<tr>'
+            f'<td style="padding:.4rem .6rem;font-family:monospace">'
+            f'{_e(name)}</td>'
+            f'<td style="padding:.4rem .6rem">{status_cell}</td>'
+            f'<td style="padding:.4rem .6rem;font-family:monospace">'
+            f'{_e(version)}</td>'
+            f'<td style="padding:.4rem .6rem">{_e(author)}</td>'
+            f'<td style="padding:.4rem .6rem;font-family:monospace">'
+            f'{_e(license_)}</td>'
+            f'<td style="padding:.4rem .6rem;font-family:monospace;'
+            f'font-size:.8rem">{_e(in_types)} → {_e(out_types)}</td>'
+            f'<td style="padding:.4rem .6rem;font-size:.8rem;'
+            f'opacity:.85">{citation_cell}</td>'
+            f'<td style="padding:.4rem .6rem;font-family:monospace;'
+            f'font-size:.8rem">{homepage_cell}</td>'
+            f'</tr>'
+        )
+    parts.append("</tbody></table></section>")
+    return "".join(parts)
+
+
+__all__ = ["build_module_audit_html"]

picarones/reports_v2/html/renderers/ner.py

@@ -0,0 +1,227 @@
+"""Rendu HTML server-side de la section NER (Sprint 41).
+
+Phase 5.C — module relocalisé depuis ``picarones.report.ner_render``
+vers ``picarones.reports_v2.html.renderers.ner``.  Le chemin legacy
+reste disponible via un shim avec ``DeprecationWarning`` ;
+suppression prévue en 2.0.
+
+Suite directe des Sprints 38-40 : la couche de calcul, le backend
+extracteur et le câblage runner sont en place ; ce module produit les
+blocs HTML qui remontent ces données dans le rapport.
+
+- ``build_ner_summary_html`` — encart factuel par moteur : F1 global,
+  precision/recall, total entités, hallucinations, missed.
+- ``build_ner_per_category_html`` — table heatmap moteur × catégorie,
+  cellules colorées par F1 (rouge → vert).
+
+Principe — cohérent avec ``inter_engine_render`` (Sprint 37) : rendu
+server-side, pas de JavaScript, déterministe.  Si aucun moteur n'a de
+``aggregated_ner``, les fonctions retournent une chaîne vide — la vue
+est silencieusement omise (rapport adaptatif).
+
+Anti-injection : tous les noms de moteurs et catégories sont passés à
+``html.escape`` avant insertion.
+"""
+
+from __future__ import annotations
+
+from html import escape as _e
+from typing import Optional
+
+from picarones.reports_v2._helpers.render_helpers import color_traffic_light
+
+
+def _engines_with_ner(engines_summary: list[dict]) -> list[dict]:
+    """Filtre les moteurs qui ont une analyse NER agrégée."""
+    return [e for e in engines_summary if e.get("aggregated_ner")]
+
+
+def build_ner_summary_html(
+    engines_summary: list[dict],
+    labels: Optional[dict[str, str]] = None,
+) -> str:
+    """Construit l'encart résumé NER : F1 global par moteur + totaux.
+
+    Parameters
+    ----------
+    engines_summary:
+        Liste de dicts moteur (au moins ``name`` et ``aggregated_ner``).
+    labels:
+        Dict d'étiquettes i18n.
+
+    Returns
+    -------
+    str
+        HTML ``<div>...</div>`` ou ``""`` si aucun moteur n'a de NER.
+    """
+    relevant = _engines_with_ner(engines_summary)
+    if not relevant:
+        return ""
+
+    labels = labels or {}
+    caption = labels.get("ner_summary_caption", "Précision sur entités nommées")
+    engine_label = labels.get("ner_engine_label", "Moteur")
+    f1_label = labels.get("ner_f1_label", "F1 global")
+    p_label = labels.get("ner_precision_label", "Précision")
+    r_label = labels.get("ner_recall_label", "Rappel")
+    docs_label = labels.get("ner_doc_count_label", "Docs évalués")
+    halluc_label = labels.get("ner_hallucinated_label", "Hallucinations")
+    missed_label = labels.get("ner_missed_label", "Entités manquées")
+
+    parts: list[str] = []
+    parts.append('<div class="ner-summary">')
+    parts.append(
+        f'<div class="ner-summary-caption" style="font-weight:600;'
+        f'margin-bottom:.4rem">{_e(caption)}</div>'
+    )
+    parts.append(
+        '<table class="ner-summary-table" '
+        'style="border-collapse:collapse;font-size:.85rem;width:100%">'
+    )
+    parts.append("<thead><tr>")
+    for hdr in (engine_label, f1_label, p_label, r_label,
+                docs_label, halluc_label, missed_label):
+        parts.append(
+            f'<th scope=\"col\" style="padding:.3rem .5rem;text-align:left;'
+            f'border-bottom:1px solid var(--border);font-weight:600">'
+            f'{_e(hdr)}</th>'
+        )
+    parts.append("</tr></thead><tbody>")
+    for engine in relevant:
+        agg = engine["aggregated_ner"]
+        global_stats = agg.get("global", {}) or {}
+        f1 = float(global_stats.get("f1") or 0.0)
+        precision = float(global_stats.get("precision") or 0.0)
+        recall = float(global_stats.get("recall") or 0.0)
+        doc_count = int(agg.get("doc_count") or 0)
+        hallucinated = int(agg.get("hallucinated_total") or 0)
+        missed = int(agg.get("missed_total") or 0)
+        bg = color_traffic_light(f1)
+        parts.append("<tr>")
+        parts.append(
+            f'<td style="padding:.3rem .5rem;font-weight:600">'
+            f'{_e(engine.get("name", ""))}</td>'
+        )
+        parts.append(
+            f'<td style="padding:.3rem .5rem;background:{bg};'
+            f'font-variant-numeric:tabular-nums">{f1 * 100:.1f} %</td>'
+        )
+        parts.append(
+            f'<td style="padding:.3rem .5rem;font-variant-numeric:tabular-nums">'
+            f'{precision * 100:.1f} %</td>'
+        )
+        parts.append(
+            f'<td style="padding:.3rem .5rem;font-variant-numeric:tabular-nums">'
+            f'{recall * 100:.1f} %</td>'
+        )
+        parts.append(
+            f'<td style="padding:.3rem .5rem;font-variant-numeric:tabular-nums">'
+            f'{doc_count}</td>'
+        )
+        parts.append(
+            f'<td style="padding:.3rem .5rem;font-variant-numeric:tabular-nums">'
+            f'{hallucinated}</td>'
+        )
+        parts.append(
+            f'<td style="padding:.3rem .5rem;font-variant-numeric:tabular-nums">'
+            f'{missed}</td>'
+        )
+        parts.append("</tr>")
+    parts.append("</tbody></table></div>")
+    return "".join(parts)
+
+
+def build_ner_per_category_html(
+    engines_summary: list[dict],
+    labels: Optional[dict[str, str]] = None,
+) -> str:
+    """Construit la heatmap NER moteur × catégorie d'entité.
+
+    Lignes = moteurs, colonnes = catégories (PER, LOC, ORG, DATE,
+    MISC…).  Cellules colorées par F1 (rouge → vert).  La cellule
+    affiche le F1 en pourcentage.  Cellules vides quand la catégorie
+    n'a pas été observée pour le moteur.
+
+    Returns
+    -------
+    str
+        HTML ``<div>...</div>`` ou ``""`` si pas de données.
+    """
+    relevant = _engines_with_ner(engines_summary)
+    if not relevant:
+        return ""
+
+    # Catégories : union sur tous les moteurs, ordre alphabétique
+    all_categories: set[str] = set()
+    for engine in relevant:
+        per_cat = (engine["aggregated_ner"] or {}).get("per_category") or {}
+        all_categories.update(per_cat.keys())
+    if not all_categories:
+        return ""
+    categories = sorted(all_categories)
+
+    labels = labels or {}
+    caption = labels.get(
+        "ner_per_category_caption",
+        "F1 par catégorie d'entité (heatmap)",
+    )
+    engine_label = labels.get("ner_engine_label", "Moteur")
+    no_data = labels.get("ner_no_data_label", "—")
+
+    parts: list[str] = []
+    parts.append('<div class="ner-per-category">')
+    parts.append(
+        f'<div class="ner-per-category-caption" '
+        f'style="font-weight:600;margin-bottom:.4rem">{_e(caption)}</div>'
+    )
+    parts.append(
+        '<table class="ner-per-category-table" '
+        'style="border-collapse:collapse;font-size:.8rem">'
+    )
+    parts.append("<thead><tr>")
+    parts.append(
+        f'<th scope=\"col\" style="padding:.3rem .5rem;text-align:left;'
+        f'border-bottom:1px solid var(--border)">{_e(engine_label)}</th>'
+    )
+    for cat in categories:
+        parts.append(
+            f'<th scope=\"col\" style="padding:.3rem .5rem;text-align:center;'
+            f'border-bottom:1px solid var(--border)">{_e(cat)}</th>'
+        )
+    parts.append("</tr></thead><tbody>")
+    for engine in relevant:
+        per_cat = (engine["aggregated_ner"] or {}).get("per_category") or {}
+        parts.append("<tr>")
+        parts.append(
+            f'<th scope=\"col\" style="padding:.3rem .5rem;text-align:right;'
+            f'border-right:1px solid var(--border);font-weight:600">'
+            f'{_e(engine.get("name", ""))}</th>'
+        )
+        for cat in categories:
+            stats = per_cat.get(cat)
+            if not stats or int(stats.get("support", 0)) == 0:
+                parts.append(
+                    f'<td style="padding:.3rem .5rem;text-align:center;'
+                    f'background:#f4f4f4;color:var(--text-muted);'
+                    f'font-style:italic">{_e(no_data)}</td>'
+                )
+            else:
+                f1 = float(stats.get("f1") or 0.0)
+                support = int(stats.get("support", 0))
+                bg = color_traffic_light(f1)
+                parts.append(
+                    f'<td style="padding:.3rem .5rem;text-align:center;'
+                    f'background:{bg};color:#222;'
+                    f'font-variant-numeric:tabular-nums" '
+                    f'title="support={support}">'
+                    f'{f1 * 100:.1f} %</td>'
+                )
+        parts.append("</tr>")
+    parts.append("</tbody></table></div>")
+    return "".join(parts)
+
+
+__all__ = [
+    "build_ner_summary_html",
+    "build_ner_per_category_html",
+]

tests/integration/test_sprint94_error_absorption.py

@@ -29,7 +29,7 @@ from picarones.measurements.error_absorption import (
     aggregate_error_absorption,
     compute_error_absorption,
 )
-from picarones.report.error_absorption_render import (
+from picarones.reports_v2.html.renderers.error_absorption import (
     build_error_absorption_html,
 )
 

tests/measurements/test_sprint93_image_predictive.py

@@ -35,7 +35,7 @@ from picarones.measurements.image_predictive import (
     compute_corpus_homogeneity,
     compute_paleographic_complexity,
 )
-from picarones.report.image_predictive_render import (
+from picarones.reports_v2.html.renderers.image_predictive import (
     build_image_predictive_html,
 )
 

tests/measurements/test_sprint96_incremental_comparison.py

@@ -31,7 +31,7 @@ from picarones.measurements.incremental_comparison import (
     PipelineRun,
     compare_isolated_effect,
 )
-from picarones.report.incremental_comparison_render import (
+from picarones.reports_v2.html.renderers.incremental_comparison import (
     build_incremental_comparison_html,
 )
 

tests/measurements/test_sprint97_module_policy.py

@@ -36,7 +36,7 @@ from picarones.measurements.module_policy import (
     validate_manifest,
 )
 from picarones.core.modules import ArtifactType, BaseModule
-from picarones.report.module_audit_render import (
+from picarones.reports_v2.html.renderers.module_audit import (
     build_module_audit_html,
 )
 

tests/report/test_sprint41_ner_html.py

@@ -23,7 +23,7 @@ import pytest
 
 from picarones.fixtures import generate_sample_benchmark
 from picarones.report.generator import ReportGenerator
-from picarones.report.ner_render import (
+from picarones.reports_v2.html.renderers.ner import (
     build_ner_per_category_html,
     build_ner_summary_html,
 )