thefinalboss commited on
Commit
1855527
·
verified ·
1 Parent(s): 0ee9303

results/hashing: SPIRA Hashing scripts, reports, artefacts (final push) (part 5)

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. results/hashing/code_scale_v1/drafts_r3_cs1/patch_verdict_r3_draft.py +75 -0
  2. results/hashing/code_scale_v1/drafts_r3_cs1/test_train_patch_r3.py +27 -0
  3. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/RESULTS_SYNTH_r3.json +1323 -0
  4. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/RESULTS_SYNTH_r3.md +44 -0
  5. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/aux/SYNTH_GO_CS1 +1 -0
  6. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/aux/SYNTH_recount64_cs1_mix_sample.json +30 -0
  7. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/aux/SYNTH_recount64_v1b_sample.json +30 -0
  8. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/aux/SYNTH_scan64_code_scale_v1.json +29 -0
  9. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/case.json +12 -0
  10. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/runs_cand/results.jsonl +1 -0
  11. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/runs_ref/results.jsonl +1 -0
  12. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/aux/SYNTH_GO_CS1 +1 -0
  13. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/aux/SYNTH_recount64_cs1_mix_sample.json +30 -0
  14. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/aux/SYNTH_recount64_v1b_sample.json +30 -0
  15. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/aux/SYNTH_scan64_code_scale_v1.json +29 -0
  16. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/case.json +10 -0
  17. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/runs_cand/results.jsonl +1 -0
  18. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/runs_ref/results.jsonl +1 -0
  19. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/aux/SYNTH_GO_CS1 +1 -0
  20. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/aux/SYNTH_recount64_cs1_mix_sample.json +30 -0
  21. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/aux/SYNTH_recount64_v1b_sample.json +30 -0
  22. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/aux/SYNTH_scan64_code_scale_v1.json +29 -0
  23. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/case.json +10 -0
  24. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/runs_cand/results.jsonl +1 -0
  25. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/runs_ref/results.jsonl +1 -0
  26. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/aux/SYNTH_GO_CS1 +1 -0
  27. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/aux/SYNTH_recount64_cs1_mix_sample.json +30 -0
  28. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/aux/SYNTH_recount64_v1b_sample.json +30 -0
  29. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/aux/SYNTH_scan64_code_scale_v1.json +29 -0
  30. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/case.json +10 -0
  31. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/runs_cand/results.jsonl +1 -0
  32. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/runs_ref/results.jsonl +1 -0
  33. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/aux/SYNTH_GO_CS1 +1 -0
  34. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/aux/SYNTH_recount64_cs1_mix_sample.json +30 -0
  35. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/aux/SYNTH_recount64_v1b_sample.json +30 -0
  36. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/aux/SYNTH_scan64_code_scale_v1.json +29 -0
  37. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/case.json +10 -0
  38. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/runs_cand/results.jsonl +1 -0
  39. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/runs_ref/results.jsonl +1 -0
  40. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/aux/SYNTH_GO_CS1 +1 -0
  41. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/aux/SYNTH_recount64_cs1_mix_sample.json +30 -0
  42. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/aux/SYNTH_recount64_v1b_sample.json +30 -0
  43. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/aux/SYNTH_scan64_code_scale_v1.json +29 -0
  44. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/case.json +10 -0
  45. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/runs_cand/results.jsonl +1 -0
  46. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/runs_ref/results.jsonl +1 -0
  47. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_c1_aucun_gain/aux/SYNTH_GO_CS1 +1 -0
  48. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_c1_aucun_gain/aux/SYNTH_recount64_cs1_mix_sample.json +30 -0
  49. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_c1_aucun_gain/aux/SYNTH_recount64_v1b_sample.json +30 -0
  50. results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_c1_aucun_gain/aux/SYNTH_scan64_code_scale_v1.json +29 -0
results/hashing/code_scale_v1/drafts_r3_cs1/patch_verdict_r3_draft.py ADDED
@@ -0,0 +1,75 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ """Fabrique verdict_CS1.r3_draft.py à partir de cs1/verdict_CS1.py r2 (fc8f0861…, lu seulement, NON modifié) par remplacements exacts."""
3
+ import hashlib, os
4
+ D = os.path.dirname(os.path.abspath(__file__)); SRC = f"{D}/../cs1/verdict_CS1.py"
5
+ s = open(SRC).read(); assert hashlib.sha256(s.encode()).hexdigest() == "fc8f0861aa054f4673f96f9d64e49b46d69118b1469bf30607e85614da931749"
6
+ def rep(old, new, n=1):
7
+ global s; assert s.count(old) == n, (s.count(old), old[:80]); s = s.replace(old, new)
8
+ rep('''(r2-f) bras imposés ctl/ctl (--ref-arm/--cand-arm n'acceptent plus « cm ») ; (r2-g) ALPHA = 0,05 nommé, même valeur partout."""''',
9
+ '''(r2-f) bras imposés ctl/ctl (--ref-arm/--cand-arm n'acceptent plus « cm ») ; (r2-g) ALPHA = 0,05 nommé, même valeur partout.
10
+ RÉVISION r3 — BROUILLON (24 oct. 2026, 14 h 40-15 h 00, Toronto ; NON FIGÉ, non audité ; r2 = fc8f0861… inchangée) — motif : taux d'appariement
11
+ ATTENDUS r2 0,21-0,33 < MATCH_MIN 0,50 sur les 5 fraîches (expected_match_CS1 r2) ⇒ NON MESURABLE quasi certain. Changements, liste complète :
12
+ (r3-a) TOKENISATION COMMUNE : les deux bras sont entraînés ET évalués avec UNE SEULE EPT, ajustée une fois sur l'échantillon du bras ctl
13
+ (v1b_sample.bin[:2 000 000], calibrée sur [2 000 000, 2 400 000), recette SPIRA.build) ; empreinte déclarée en dur :
14
+ sha256(log2p float32 256×256) = COMMON_EPT_LOG2P_SHA, ept_params = COMMON_EPT_PARAMS. Le script relit l'EPT dans le ckpt de CHAQUE bras
15
+ (<runs>/<arm>.npz, clés log2p / ept_params, écrit par SPIRA.save) ; le sha256 du ckpt doit être égal au ckpt_sha256 de results.jsonl du bras.
16
+ Ckpt absent, illisible, empreinte différente ou sha ≠ results.jsonl ⇒ NON MESURABLE.
17
+ (r3-b) contrôle par fenêtre : sur la plage d'octets commune, les cibles (tgt_start, tgt_len) des deux bras doivent être IDENTIQUES (même ordre,
18
+ même nombre) ; sinon ⇒ NON MESURABLE (anomalie : avec une EPT commune la segmentation est déterministe ; on refuse d'exclure en silence
19
+ une fenêtre, qui pourrait porter une perte). Avec r3-a/r3-b, les contextes tokenisés sont aussi identiques (la « limite connue » r2 disparaît).
20
+ (r3-c) INCHANGÉS : MATCH_MIN = 0,50 (même définition plage commune r2, même ≥ inclusif ; attendu non contraignant, taux attendu 1,000),
21
+ règle B, ALPHA, garde seen non apparié (Fisher ; seen = préfixe [:200 000] de l'échantillon de chaque bras, toujours non appariable),
22
+ scan de tous les npz, dénominateur plage commune, n_clean par fenêtre, contrôles sha des fenêtres et des recomptes, ctl/ctl imposé.
23
+ Le scan r2-a couvre déjà <arm>.npz (listé dans inputs_sha256)."""''')
24
+ rep('''ALPHA = 0.05 # seuil unique : McNemar (gain/perte) ET Fisher seen (r2)''',
25
+ '''ALPHA = 0.05 # seuil unique : McNemar (gain/perte) ET Fisher seen (r2)
26
+ COMMON_EPT_LOG2P_SHA = "bc3e74d8b2812c4248ccd97bb73fa2e05f9c945fdd79b63487cba1533cf71d2a" # r3-a : EPT commune (fit v1b_sample [:2 Mo], recette SPIRA.build)
27
+ COMMON_EPT_PARAMS = [16.04248046875, 2.0, 32.0, 0.5] # r3-a : [budget_bits, min_len, max_len, alpha]''')
28
+ # r3-a : contrôle EPT commune, après le bloc budgets
29
+ rep('''TRAINED_MIN = {x: (min(min(TRAINED[x]), BUD["snn_tokens"][x]) if TRAINED[x] and isinstance(BUD["snn_tokens"][x], int) else None) for x in ("ref", "cand")}''',
30
+ '''TRAINED_MIN = {x: (min(min(TRAINED[x]), BUD["snn_tokens"][x]) if TRAINED[x] and isinstance(BUD["snn_tokens"][x], int) else None) for x in ("ref", "cand")}
31
+ # ---- r3-a : tokenisation commune — EPT relue dans le ckpt de chaque bras, comparée à l'empreinte déclarée
32
+ def ept_check(side):
33
+ p = f"{DR if side == 'ref' else DC}/{ARM[side]}.npz"; k = f"{side}:{ARM[side]}.npz"; r = {"ckpt": p, "present": k in NPZ_SCAN}
34
+ if not r["present"]: return {**r, "ok": False, "reason": "ckpt absent"}
35
+ with np.load(p, allow_pickle=False) as Z:
36
+ if "log2p" not in Z.files or "ept_params" not in Z.files: return {**r, "ok": False, "reason": "log2p / ept_params absents du ckpt"}
37
+ lp = Z["log2p"]; pr = [float(x) for x in np.asarray(Z["ept_params"]).ravel()]
38
+ r["log2p_dtype_shape"] = [str(lp.dtype), list(lp.shape)]
39
+ r["log2p_sha256"] = hashlib.sha256(np.ascontiguousarray(lp).tobytes()).hexdigest() if lp.dtype == np.float32 and lp.shape == (256, 256) else None
40
+ r["ept_params"] = pr; r["ckpt_sha256_file"] = NPZ_SCAN[k]["sha256_file"]; r["ckpt_sha256_results_jsonl"] = arm_field(side, "ckpt_sha256")
41
+ r["log2p_ok"] = r["log2p_sha256"] == COMMON_EPT_LOG2P_SHA; r["params_ok"] = pr == COMMON_EPT_PARAMS
42
+ r["ckpt_is_run_ckpt"] = r["ckpt_sha256_file"] == r["ckpt_sha256_results_jsonl"]
43
+ r["ok"] = bool(r["log2p_ok"] and r["params_ok"] and r["ckpt_is_run_ckpt"])
44
+ if not r["ok"]: r["reason"] = ", ".join(n for n, f in (("log2p ≠ EPT commune", r["log2p_ok"]), ("ept_params ≠ EPT commune", r["params_ok"]),
45
+ ("sha ckpt ≠ ckpt_sha256 results.jsonl", r["ckpt_is_run_ckpt"])) if not f)
46
+ return r
47
+ EPT_CHECK = {x: ept_check(x) for x in ("ref", "cand")}
48
+ EPT_OK = all(EPT_CHECK[x]["ok"] for x in ("ref", "cand"))''')
49
+ # r3-b : identité des cibles sur la plage commune, par fenêtre
50
+ rep(''' same_tok = na == nb_ and np.array_equal(A["tgt_start"], B["tgt_start"]) and np.array_equal(A["tgt_len"], B["tgt_len"])''',
51
+ ''' same_tok = na == nb_ and np.array_equal(A["tgt_start"], B["tgt_start"]) and np.array_equal(A["tgt_len"], B["tgt_len"])
52
+ same_common = bool(np.array_equal(A["tgt_start"][ina], B["tgt_start"][inb]) and np.array_equal(A["tgt_len"][ina], B["tgt_len"][inb])) # r3-b''')
53
+ rep('''"match_ok": rate >= MATCH_MIN, "tokenisation_identical": bool(same_tok)}''',
54
+ '''"match_ok": rate >= MATCH_MIN, "tokenisation_identical": bool(same_tok),
55
+ "tokenisation_identical_common_range_r3": same_common}''')
56
+ rep('''FV = [w for w in FRESH if R[w]["valid"]]''',
57
+ '''TOK_ID = {w: MATCH[w]["tokenisation_identical_common_range_r3"] for w in EVALW}
58
+ TOK_OK = bool(EPT_OK and all(TOK_ID.values())) # r3-a + r3-b
59
+ COMMON_TOK = {"declared": {"log2p_sha256": COMMON_EPT_LOG2P_SHA, "ept_params": COMMON_EPT_PARAMS, "fitted_on": "v1b_sample.bin (bras ctl / témoin) [:2 000 000], calibrée [2 000 000, 2 400 000)"},
60
+ "per_arm": EPT_CHECK, "ept_ok_both": EPT_OK, "targets_identical_common_range_per_window": TOK_ID, "ok": TOK_OK}
61
+ FV = [w for w in FRESH if R[w]["valid"]]''')
62
+ rep('''if not SEEN_OK: why.append("seen d'un bras avec trained_frac < 0,99 ou non vérifiable")''',
63
+ '''if not SEEN_OK: why.append("seen d'un bras avec trained_frac < 0,99 ou non vérifiable")
64
+ if not EPT_OK: why.append("tokenisation commune non vérifiée (r3-a) : " + " / ".join(f"{x} {EPT_CHECK[x].get('reason')}" for x in ("ref", "cand") if not EPT_CHECK[x]["ok"]))
65
+ if not all(TOK_ID.values()): why.append("cibles non identiques sur la plage commune (r3-b) : " + ", ".join(w for w in EVALW if not TOK_ID[w]))''')
66
+ rep('''if BUDGET_OK and SEEN_OK and SCALE_OK and FV:''', '''if BUDGET_OK and SEEN_OK and SCALE_OK and TOK_OK and FV:''')
67
+ rep('''"revision": "r2", "alpha": ALPHA,''', '''"revision": "r3-draft (NON FIGÉ)", "common_tokenisation": COMMON_TOK, "alpha": ALPHA,''')
68
+ rep('''ou aucune fraîche valide",
69
+ "bootstrap"''', '''ou aucune fraîche valide ; r3 : NON MESURABLE aussi si la tokenisation commune n'est pas vérifiée (EPT des ckpts ≠ empreinte déclarée, ckpt ≠ run) ou si les cibles diffèrent sur la plage commune d'une fenêtre",
70
+ "bootstrap"''')
71
+ rep(''' f"Budgets : {json.dumps(BUD)}''', ''' f"Tokenisation commune (r3) : EPT {COMMON_EPT_LOG2P_SHA[:16]}… {COMMON_EPT_PARAMS} — réf {EPT_CHECK['ref']['ok']} / cand {EPT_CHECK['cand']['ok']} ; cibles identiques (plage commune) : {json.dumps(TOK_ID)} → {'OK' if TOK_OK else 'NON OK'}.",
72
+ f"Budgets : {json.dumps(BUD)}''')
73
+ rep('''"Précision fused APPARIÉE (%) sur les cibles identiques (tgt_start, tgt_len) in-vocab dans les DEUX bras (v4.2).", "",''',
74
+ '''"Précision fused APPARIÉE (%) sur les cibles identiques (tgt_start, tgt_len) in-vocab dans les DEUX bras (v4.2) ; r3 : tokenisation commune (même EPT, mêmes cibles, mêmes contextes).", "",''')
75
+ open(f"{D}/verdict_CS1.r3_draft.py", "w").write(s); print("écrit", hashlib.sha256(s.encode()).hexdigest())
results/hashing/code_scale_v1/drafts_r3_cs1/test_train_patch_r3.py ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ """BROUILLON r3 — test LÉGER du remplacement d'EPT de train_cs1_r3_draft.py (aucun entraînement, aucun SPIRA.build complet) :
3
+ reproduit les 2 appels EPT de SPIRA.build (fit [:2 Mo], calibrate [2 ; 2,4 Mo)) sur les 2,4 premiers Mo de chaque échantillon.
4
+ T1 : sans patch, EPT(v1b) == EPT commune (le témoin n'est pas modifié par r3). T2 : avec patch, EPT(mix) == EPT commune.
5
+ T3 : sans patch, EPT(mix) != EPT commune (contrôle : le patch est nécessaire). T4 : segmentation d'une fenêtre identique entre bras patchés."""
6
+ import sys, os, hashlib, importlib.util, numpy as np
7
+ sys.path.insert(0, "/workspace/SPIRA/python"); from spira.ept import EPT; from spira.model import SPIRAConfig
8
+ D = os.path.dirname(os.path.abspath(__file__)); S = "/workspace/spira_shared"
9
+ spec = importlib.util.spec_from_file_location("tr3", f"{D}/train_cs1_r3_draft.py"); tr3 = importlib.util.module_from_spec(spec); spec.loader.exec_module(tr3)
10
+ cfg = SPIRAConfig(vocab_size=16384, d_in=1024, snn_tokens=600000, am_tokens=1200000, epochs=2)
11
+ V1B, MIX = f"{S}/hashing_gen/m8_v1b/sample/v1b_sample.bin", f"{S}/hashing_gen/code_scale_v1/sample/cs1_mix_sample.bin"
12
+ def as_build(path):
13
+ h = open(path, "rb").read(2_400_000); e = EPT(budget_bits=16.0, min_len=cfg.min_len, max_len=cfg.max_len, alpha=cfg.alpha)
14
+ e.fit(h[:2_000_000]); e.calibrate(h[2_000_000:2_400_000], target_len=cfg.target_avg_len); return e
15
+ fp = lambda e: (hashlib.sha256(np.ascontiguousarray(e.log2p, np.float32).tobytes()).hexdigest(), [float(e.budget_bits), int(e.min_len), int(e.max_len), float(e.alpha)])
16
+ C = (tr3.COMMON_LOG2P_SHA, tr3.COMMON_PARAMS); R = {}
17
+ R["T1_ctl_sans_patch_egal_commune"] = fp(as_build(V1B)) == C
18
+ R["T3_mix_sans_patch_different"] = fp(as_build(MIX)) != C
19
+ of, oc = EPT.fit, EPT.calibrate; tr3.install_common_ept()
20
+ emix, ev1b = as_build(MIX), as_build(V1B)
21
+ R["T2_mix_avec_patch_egal_commune"] = fp(emix) == C
22
+ R["T2b_ctl_avec_patch_egal_commune"] = fp(ev1b) == C
23
+ wd = open(f"{S}/spira_m7/windows/fresh_fr.bin", "rb").read()
24
+ R["T4_fresh_fr_segmentation_identique"] = bool(np.array_equal(emix.boundaries(wd), ev1b.boundaries(wd)))
25
+ EPT.fit, EPT.calibrate = of, oc
26
+ for k, v in R.items(): print(("PASS " if v else "FAIL ") + k)
27
+ print(f"{sum(R.values())}/{len(R)} PASS"); sys.exit(0 if all(R.values()) else 1)
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/RESULTS_SYNTH_r3.json ADDED
@@ -0,0 +1,1323 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "case": "a_gain_frais",
4
+ "desc": "gain significatif sur fc2, aucune autre différence ; r3 : taux 1,000 partout",
5
+ "attendu": "ADOPTÉ",
6
+ "obtenu": "ADOPTÉ",
7
+ "checks": {
8
+ "all_rates_1": true
9
+ },
10
+ "conforme": true,
11
+ "raisons": "",
12
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
13
+ "revision": "r3-draft (NON FIGÉ)",
14
+ "fresh_valid": [
15
+ "fc2",
16
+ "fresh_en",
17
+ "fresh_fr",
18
+ "test20k",
19
+ "test_v3"
20
+ ],
21
+ "gains": [
22
+ "fc2"
23
+ ],
24
+ "losses": [],
25
+ "rates": {
26
+ "fc2": 1.0,
27
+ "fresh_en": 1.0,
28
+ "fresh_fr": 1.0,
29
+ "test20k": 1.0,
30
+ "test_v3": 1.0,
31
+ "queries": 1.0
32
+ },
33
+ "tok_ok": true,
34
+ "ept_ok": {
35
+ "ref": true,
36
+ "cand": true
37
+ },
38
+ "seen_fisher_p": 0.41236097478755496,
39
+ "n_inputs_npz": 16
40
+ },
41
+ {
42
+ "case": "b1_gain_perte_fraiche",
43
+ "desc": "gain fc2 + perte significative fresh_en",
44
+ "attendu": "REJETÉ",
45
+ "obtenu": "REJETÉ",
46
+ "checks": {},
47
+ "conforme": true,
48
+ "raisons": "perte(s) significative(s) : fresh_en",
49
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
50
+ "revision": "r3-draft (NON FIGÉ)",
51
+ "fresh_valid": [
52
+ "fc2",
53
+ "fresh_en",
54
+ "fresh_fr",
55
+ "test20k",
56
+ "test_v3"
57
+ ],
58
+ "gains": [
59
+ "fc2"
60
+ ],
61
+ "losses": [
62
+ "fresh_en"
63
+ ],
64
+ "rates": {
65
+ "fc2": 1.0,
66
+ "fresh_en": 1.0,
67
+ "fresh_fr": 1.0,
68
+ "test20k": 1.0,
69
+ "test_v3": 1.0,
70
+ "queries": 1.0
71
+ },
72
+ "tok_ok": true,
73
+ "ept_ok": {
74
+ "ref": true,
75
+ "cand": true
76
+ },
77
+ "seen_fisher_p": 0.41236097478755496,
78
+ "n_inputs_npz": 16
79
+ },
80
+ {
81
+ "case": "b2_gain_perte_seen",
82
+ "desc": "gain fc2 + seen cand à 0 % (garde seen non apparié)",
83
+ "attendu": "REJETÉ",
84
+ "obtenu": "REJETÉ",
85
+ "checks": {},
86
+ "conforme": true,
87
+ "raisons": "perte(s) significative(s) : seen (non apparié, Fisher)",
88
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
89
+ "revision": "r3-draft (NON FIGÉ)",
90
+ "fresh_valid": [
91
+ "fc2",
92
+ "fresh_en",
93
+ "fresh_fr",
94
+ "test20k",
95
+ "test_v3"
96
+ ],
97
+ "gains": [
98
+ "fc2"
99
+ ],
100
+ "losses": [
101
+ "seen (non apparié, Fisher)"
102
+ ],
103
+ "rates": {
104
+ "fc2": 1.0,
105
+ "fresh_en": 1.0,
106
+ "fresh_fr": 1.0,
107
+ "test20k": 1.0,
108
+ "test_v3": 1.0,
109
+ "queries": 1.0
110
+ },
111
+ "tok_ok": true,
112
+ "ept_ok": {
113
+ "ref": true,
114
+ "cand": true
115
+ },
116
+ "seen_fisher_p": 3.6910103552267424e-149,
117
+ "n_inputs_npz": 16
118
+ },
119
+ {
120
+ "case": "b3_gain_perte_queries",
121
+ "desc": "gain fc2 + perte significative queries (hors règle)",
122
+ "attendu": "ADOPTÉ",
123
+ "obtenu": "ADOPTÉ",
124
+ "checks": {},
125
+ "conforme": true,
126
+ "raisons": "",
127
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
128
+ "revision": "r3-draft (NON FIGÉ)",
129
+ "fresh_valid": [
130
+ "fc2",
131
+ "fresh_en",
132
+ "fresh_fr",
133
+ "test20k",
134
+ "test_v3"
135
+ ],
136
+ "gains": [
137
+ "fc2"
138
+ ],
139
+ "losses": [],
140
+ "rates": {
141
+ "fc2": 1.0,
142
+ "fresh_en": 1.0,
143
+ "fresh_fr": 1.0,
144
+ "test20k": 1.0,
145
+ "test_v3": 1.0,
146
+ "queries": 1.0
147
+ },
148
+ "tok_ok": true,
149
+ "ept_ok": {
150
+ "ref": true,
151
+ "cand": true
152
+ },
153
+ "seen_fisher_p": 0.41236097478755496,
154
+ "n_inputs_npz": 16
155
+ },
156
+ {
157
+ "case": "b4_gain_perte_test20k",
158
+ "desc": "gain fc2 + perte significative test20k strict",
159
+ "attendu": "REJETÉ",
160
+ "obtenu": "REJETÉ",
161
+ "checks": {},
162
+ "conforme": true,
163
+ "raisons": "perte(s) significative(s) : test20k",
164
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
165
+ "revision": "r3-draft (NON FIGÉ)",
166
+ "fresh_valid": [
167
+ "fc2",
168
+ "fresh_en",
169
+ "fresh_fr",
170
+ "test20k",
171
+ "test_v3"
172
+ ],
173
+ "gains": [
174
+ "fc2"
175
+ ],
176
+ "losses": [
177
+ "test20k"
178
+ ],
179
+ "rates": {
180
+ "fc2": 1.0,
181
+ "fresh_en": 1.0,
182
+ "fresh_fr": 1.0,
183
+ "test20k": 1.0,
184
+ "test_v3": 1.0,
185
+ "queries": 1.0
186
+ },
187
+ "tok_ok": true,
188
+ "ept_ok": {
189
+ "ref": true,
190
+ "cand": true
191
+ },
192
+ "seen_fisher_p": 0.41236097478755496,
193
+ "n_inputs_npz": 16
194
+ },
195
+ {
196
+ "case": "b5_seen_un_peu_bas_ns",
197
+ "desc": "gain fc2 + seen cand −20 hits / 1 000, non significatif",
198
+ "attendu": "ADOPTÉ",
199
+ "obtenu": "ADOPTÉ",
200
+ "checks": {},
201
+ "conforme": true,
202
+ "raisons": "",
203
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
204
+ "revision": "r3-draft (NON FIGÉ)",
205
+ "fresh_valid": [
206
+ "fc2",
207
+ "fresh_en",
208
+ "fresh_fr",
209
+ "test20k",
210
+ "test_v3"
211
+ ],
212
+ "gains": [
213
+ "fc2"
214
+ ],
215
+ "losses": [],
216
+ "rates": {
217
+ "fc2": 1.0,
218
+ "fresh_en": 1.0,
219
+ "fresh_fr": 1.0,
220
+ "test20k": 1.0,
221
+ "test_v3": 1.0,
222
+ "queries": 1.0
223
+ },
224
+ "tok_ok": true,
225
+ "ept_ok": {
226
+ "ref": true,
227
+ "cand": true
228
+ },
229
+ "seen_fisher_p": 0.38678602351324315,
230
+ "n_inputs_npz": 16
231
+ },
232
+ {
233
+ "case": "b6_seen_cand_plus_haut",
234
+ "desc": "gain fc2 + seen cand +200 hits / 1 000",
235
+ "attendu": "ADOPTÉ",
236
+ "obtenu": "ADOPTÉ",
237
+ "checks": {},
238
+ "conforme": true,
239
+ "raisons": "",
240
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
241
+ "revision": "r3-draft (NON FIGÉ)",
242
+ "fresh_valid": [
243
+ "fc2",
244
+ "fresh_en",
245
+ "fresh_fr",
246
+ "test20k",
247
+ "test_v3"
248
+ ],
249
+ "gains": [
250
+ "fc2"
251
+ ],
252
+ "losses": [],
253
+ "rates": {
254
+ "fc2": 1.0,
255
+ "fresh_en": 1.0,
256
+ "fresh_fr": 1.0,
257
+ "test20k": 1.0,
258
+ "test_v3": 1.0,
259
+ "queries": 1.0
260
+ },
261
+ "tok_ok": true,
262
+ "ept_ok": {
263
+ "ref": true,
264
+ "cand": true
265
+ },
266
+ "seen_fisher_p": 4.141338000811789e-19,
267
+ "n_inputs_npz": 16
268
+ },
269
+ {
270
+ "case": "b7_seen_perte_sans_gain",
271
+ "desc": "aucun gain + seen cand à 0 %",
272
+ "attendu": "REJETÉ",
273
+ "obtenu": "REJETÉ",
274
+ "checks": {},
275
+ "conforme": true,
276
+ "raisons": "aucun gain significatif sur une fenêtre fraîche valide; perte(s) significative(s) : seen (non apparié, Fisher)",
277
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
278
+ "revision": "r3-draft (NON FIGÉ)",
279
+ "fresh_valid": [
280
+ "fc2",
281
+ "fresh_en",
282
+ "fresh_fr",
283
+ "test20k",
284
+ "test_v3"
285
+ ],
286
+ "gains": [],
287
+ "losses": [
288
+ "seen (non apparié, Fisher)"
289
+ ],
290
+ "rates": {
291
+ "fc2": 1.0,
292
+ "fresh_en": 1.0,
293
+ "fresh_fr": 1.0,
294
+ "test20k": 1.0,
295
+ "test_v3": 1.0,
296
+ "queries": 1.0
297
+ },
298
+ "tok_ok": true,
299
+ "ept_ok": {
300
+ "ref": true,
301
+ "cand": true
302
+ },
303
+ "seen_fisher_p": 3.6910103552267424e-149,
304
+ "n_inputs_npz": 16
305
+ },
306
+ {
307
+ "case": "c1_aucun_gain",
308
+ "desc": "bras identiques",
309
+ "attendu": "REJETÉ",
310
+ "obtenu": "REJETÉ",
311
+ "checks": {},
312
+ "conforme": true,
313
+ "raisons": "aucun gain significatif sur une fenêtre fraîche valide",
314
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
315
+ "revision": "r3-draft (NON FIGÉ)",
316
+ "fresh_valid": [
317
+ "fc2",
318
+ "fresh_en",
319
+ "fresh_fr",
320
+ "test20k",
321
+ "test_v3"
322
+ ],
323
+ "gains": [],
324
+ "losses": [],
325
+ "rates": {
326
+ "fc2": 1.0,
327
+ "fresh_en": 1.0,
328
+ "fresh_fr": 1.0,
329
+ "test20k": 1.0,
330
+ "test_v3": 1.0,
331
+ "queries": 1.0
332
+ },
333
+ "tok_ok": true,
334
+ "ept_ok": {
335
+ "ref": true,
336
+ "cand": true
337
+ },
338
+ "seen_fisher_p": 0.41236097478755496,
339
+ "n_inputs_npz": 16
340
+ },
341
+ {
342
+ "case": "c2_gain_queries_seul",
343
+ "desc": "gain seulement sur queries (hors règle)",
344
+ "attendu": "REJETÉ",
345
+ "obtenu": "REJETÉ",
346
+ "checks": {},
347
+ "conforme": true,
348
+ "raisons": "aucun gain significatif sur une fenêtre fraîche valide",
349
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
350
+ "revision": "r3-draft (NON FIGÉ)",
351
+ "fresh_valid": [
352
+ "fc2",
353
+ "fresh_en",
354
+ "fresh_fr",
355
+ "test20k",
356
+ "test_v3"
357
+ ],
358
+ "gains": [],
359
+ "losses": [],
360
+ "rates": {
361
+ "fc2": 1.0,
362
+ "fresh_en": 1.0,
363
+ "fresh_fr": 1.0,
364
+ "test20k": 1.0,
365
+ "test_v3": 1.0,
366
+ "queries": 1.0
367
+ },
368
+ "tok_ok": true,
369
+ "ept_ok": {
370
+ "ref": true,
371
+ "cand": true
372
+ },
373
+ "seen_fisher_p": 0.41236097478755496,
374
+ "n_inputs_npz": 16
375
+ },
376
+ {
377
+ "case": "d1_scale_ref_bas",
378
+ "desc": "gain fc2, fusion.scale réf 0,02",
379
+ "attendu": "NON MESURABLE",
380
+ "obtenu": "NON MESURABLE",
381
+ "checks": {},
382
+ "conforme": true,
383
+ "raisons": "fusion.scale hors [0.03 ; 0.11] : réf 0.02 / cand 0.07",
384
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
385
+ "revision": "r3-draft (NON FIGÉ)",
386
+ "fresh_valid": [
387
+ "fc2",
388
+ "fresh_en",
389
+ "fresh_fr",
390
+ "test20k",
391
+ "test_v3"
392
+ ],
393
+ "gains": [
394
+ "fc2"
395
+ ],
396
+ "losses": [],
397
+ "rates": {
398
+ "fc2": 1.0,
399
+ "fresh_en": 1.0,
400
+ "fresh_fr": 1.0,
401
+ "test20k": 1.0,
402
+ "test_v3": 1.0,
403
+ "queries": 1.0
404
+ },
405
+ "tok_ok": true,
406
+ "ept_ok": {
407
+ "ref": true,
408
+ "cand": true
409
+ },
410
+ "seen_fisher_p": 0.41236097478755496,
411
+ "n_inputs_npz": 16
412
+ },
413
+ {
414
+ "case": "d2_scale_cand_haut",
415
+ "desc": "gain fc2, fusion.scale cand 0,12",
416
+ "attendu": "NON MESURABLE",
417
+ "obtenu": "NON MESURABLE",
418
+ "checks": {},
419
+ "conforme": true,
420
+ "raisons": "fusion.scale hors [0.03 ; 0.11] : réf 0.07 / cand 0.12",
421
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
422
+ "revision": "r3-draft (NON FIGÉ)",
423
+ "fresh_valid": [
424
+ "fc2",
425
+ "fresh_en",
426
+ "fresh_fr",
427
+ "test20k",
428
+ "test_v3"
429
+ ],
430
+ "gains": [
431
+ "fc2"
432
+ ],
433
+ "losses": [],
434
+ "rates": {
435
+ "fc2": 1.0,
436
+ "fresh_en": 1.0,
437
+ "fresh_fr": 1.0,
438
+ "test20k": 1.0,
439
+ "test_v3": 1.0,
440
+ "queries": 1.0
441
+ },
442
+ "tok_ok": true,
443
+ "ept_ok": {
444
+ "ref": true,
445
+ "cand": true
446
+ },
447
+ "seen_fisher_p": 0.41236097478755496,
448
+ "n_inputs_npz": 16
449
+ },
450
+ {
451
+ "case": "d3_scale_bornes",
452
+ "desc": "gain fc2, fusion.scale 0,03 / 0,11",
453
+ "attendu": "ADOPTÉ",
454
+ "obtenu": "ADOPTÉ",
455
+ "checks": {},
456
+ "conforme": true,
457
+ "raisons": "",
458
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
459
+ "revision": "r3-draft (NON FIGÉ)",
460
+ "fresh_valid": [
461
+ "fc2",
462
+ "fresh_en",
463
+ "fresh_fr",
464
+ "test20k",
465
+ "test_v3"
466
+ ],
467
+ "gains": [
468
+ "fc2"
469
+ ],
470
+ "losses": [],
471
+ "rates": {
472
+ "fc2": 1.0,
473
+ "fresh_en": 1.0,
474
+ "fresh_fr": 1.0,
475
+ "test20k": 1.0,
476
+ "test_v3": 1.0,
477
+ "queries": 1.0
478
+ },
479
+ "tok_ok": true,
480
+ "ept_ok": {
481
+ "ref": true,
482
+ "cand": true
483
+ },
484
+ "seen_fisher_p": 0.41236097478755496,
485
+ "n_inputs_npz": 16
486
+ },
487
+ {
488
+ "case": "d4_scale_absent",
489
+ "desc": "gain fc2, cal.scale absent côté cand",
490
+ "attendu": "NON MESURABLE",
491
+ "obtenu": "NON MESURABLE",
492
+ "checks": {},
493
+ "conforme": true,
494
+ "raisons": "fusion.scale hors [0.03 ; 0.11] : réf 0.07 / cand None",
495
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
496
+ "revision": "r3-draft (NON FIGÉ)",
497
+ "fresh_valid": [
498
+ "fc2",
499
+ "fresh_en",
500
+ "fresh_fr",
501
+ "test20k",
502
+ "test_v3"
503
+ ],
504
+ "gains": [
505
+ "fc2"
506
+ ],
507
+ "losses": [],
508
+ "rates": {
509
+ "fc2": 1.0,
510
+ "fresh_en": 1.0,
511
+ "fresh_fr": 1.0,
512
+ "test20k": 1.0,
513
+ "test_v3": 1.0,
514
+ "queries": 1.0
515
+ },
516
+ "tok_ok": true,
517
+ "ept_ok": {
518
+ "ref": true,
519
+ "cand": true
520
+ },
521
+ "seen_fisher_p": 0.41236097478755496,
522
+ "n_inputs_npz": 16
523
+ },
524
+ {
525
+ "case": "e1_appariement_60",
526
+ "desc": "cibles fc2 différentes (taux plage commune 0,60), gain fc2",
527
+ "attendu": "NON MESURABLE",
528
+ "obtenu": "NON MESURABLE",
529
+ "checks": {
530
+ "reason_contains": true,
531
+ "window_reason_absent": true
532
+ },
533
+ "conforme": true,
534
+ "raisons": "cibles non identiques sur la plage commune (r3-b) : fc2",
535
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
536
+ "revision": "r3-draft (NON FIGÉ)",
537
+ "fresh_valid": [
538
+ "fc2",
539
+ "fresh_en",
540
+ "fresh_fr",
541
+ "test20k",
542
+ "test_v3"
543
+ ],
544
+ "gains": [
545
+ "fc2"
546
+ ],
547
+ "losses": [],
548
+ "rates": {
549
+ "fc2": 0.6004,
550
+ "fresh_en": 1.0,
551
+ "fresh_fr": 1.0,
552
+ "test20k": 1.0,
553
+ "test_v3": 1.0,
554
+ "queries": 1.0
555
+ },
556
+ "tok_ok": false,
557
+ "ept_ok": {
558
+ "ref": true,
559
+ "cand": true
560
+ },
561
+ "seen_fisher_p": 0.41236097478755496,
562
+ "n_inputs_npz": 16
563
+ },
564
+ {
565
+ "case": "e2_appariement_40",
566
+ "desc": "cibles fc2 différentes (taux 0,40), gain fc2",
567
+ "attendu": "NON MESURABLE",
568
+ "obtenu": "NON MESURABLE",
569
+ "checks": {
570
+ "reason_contains": true,
571
+ "window_reason_contains": true
572
+ },
573
+ "conforme": true,
574
+ "raisons": "cibles non identiques sur la plage commune (r3-b) : fc2",
575
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
576
+ "revision": "r3-draft (NON FIGÉ)",
577
+ "fresh_valid": [
578
+ "fresh_en",
579
+ "fresh_fr",
580
+ "test20k",
581
+ "test_v3"
582
+ ],
583
+ "gains": [],
584
+ "losses": [],
585
+ "rates": {
586
+ "fc2": 0.4,
587
+ "fresh_en": 1.0,
588
+ "fresh_fr": 1.0,
589
+ "test20k": 1.0,
590
+ "test_v3": 1.0,
591
+ "queries": 1.0
592
+ },
593
+ "tok_ok": false,
594
+ "ept_ok": {
595
+ "ref": true,
596
+ "cand": true
597
+ },
598
+ "seen_fisher_p": 0.41236097478755496,
599
+ "n_inputs_npz": 16
600
+ },
601
+ {
602
+ "case": "e3_appariement_50",
603
+ "desc": "cibles fc2 différentes (taux 0,5002), gain fc2",
604
+ "attendu": "NON MESURABLE",
605
+ "obtenu": "NON MESURABLE",
606
+ "checks": {
607
+ "reason_contains": true,
608
+ "window_reason_absent": true
609
+ },
610
+ "conforme": true,
611
+ "raisons": "cibles non identiques sur la plage commune (r3-b) : fc2",
612
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
613
+ "revision": "r3-draft (NON FIGÉ)",
614
+ "fresh_valid": [
615
+ "fc2",
616
+ "fresh_en",
617
+ "fresh_fr",
618
+ "test20k",
619
+ "test_v3"
620
+ ],
621
+ "gains": [
622
+ "fc2"
623
+ ],
624
+ "losses": [],
625
+ "rates": {
626
+ "fc2": 0.5002,
627
+ "fresh_en": 1.0,
628
+ "fresh_fr": 1.0,
629
+ "test20k": 1.0,
630
+ "test_v3": 1.0,
631
+ "queries": 1.0
632
+ },
633
+ "tok_ok": false,
634
+ "ept_ok": {
635
+ "ref": true,
636
+ "cand": true
637
+ },
638
+ "seen_fisher_p": 0.41236097478755496,
639
+ "n_inputs_npz": 16
640
+ },
641
+ {
642
+ "case": "e3b_appariement_4996",
643
+ "desc": "cibles fc2 différentes (taux 0,4996), gain fc2",
644
+ "attendu": "NON MESURABLE",
645
+ "obtenu": "NON MESURABLE",
646
+ "checks": {
647
+ "reason_contains": true,
648
+ "window_reason_contains": true
649
+ },
650
+ "conforme": true,
651
+ "raisons": "cibles non identiques sur la plage commune (r3-b) : fc2",
652
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
653
+ "revision": "r3-draft (NON FIGÉ)",
654
+ "fresh_valid": [
655
+ "fresh_en",
656
+ "fresh_fr",
657
+ "test20k",
658
+ "test_v3"
659
+ ],
660
+ "gains": [],
661
+ "losses": [],
662
+ "rates": {
663
+ "fc2": 0.4996,
664
+ "fresh_en": 1.0,
665
+ "fresh_fr": 1.0,
666
+ "test20k": 1.0,
667
+ "test_v3": 1.0,
668
+ "queries": 1.0
669
+ },
670
+ "tok_ok": false,
671
+ "ept_ok": {
672
+ "ref": true,
673
+ "cand": true
674
+ },
675
+ "seen_fisher_p": 0.41236097478755496,
676
+ "n_inputs_npz": 16
677
+ },
678
+ {
679
+ "case": "e4_appariement_45_denominateur",
680
+ "desc": "cibles fc2 différentes (taux 0,62), perte fc2 + gain fresh_en",
681
+ "attendu": "NON MESURABLE",
682
+ "obtenu": "NON MESURABLE",
683
+ "checks": {
684
+ "reason_contains": true
685
+ },
686
+ "conforme": true,
687
+ "raisons": "cibles non identiques sur la plage commune (r3-b) : fc2",
688
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
689
+ "revision": "r3-draft (NON FIGÉ)",
690
+ "fresh_valid": [
691
+ "fc2",
692
+ "fresh_en",
693
+ "fresh_fr",
694
+ "test20k",
695
+ "test_v3"
696
+ ],
697
+ "gains": [
698
+ "fresh_en"
699
+ ],
700
+ "losses": [
701
+ "fc2"
702
+ ],
703
+ "rates": {
704
+ "fc2": 0.6207,
705
+ "fresh_en": 1.0,
706
+ "fresh_fr": 1.0,
707
+ "test20k": 1.0,
708
+ "test_v3": 1.0,
709
+ "queries": 1.0
710
+ },
711
+ "tok_ok": false,
712
+ "ept_ok": {
713
+ "ref": true,
714
+ "cand": true
715
+ },
716
+ "seen_fisher_p": 0.41236097478755496,
717
+ "n_inputs_npz": 16
718
+ },
719
+ {
720
+ "case": "e5_tout_sous_50",
721
+ "desc": "taux 0,40 sur les 5 fraîches",
722
+ "attendu": "NON MESURABLE",
723
+ "obtenu": "NON MESURABLE",
724
+ "checks": {
725
+ "reason_contains": true
726
+ },
727
+ "conforme": true,
728
+ "raisons": "cibles non identiques sur la plage commune (r3-b) : fc2, fresh_en, fresh_fr, test20k, test_v3; aucune fenêtre fraîche valide",
729
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
730
+ "revision": "r3-draft (NON FIGÉ)",
731
+ "fresh_valid": [],
732
+ "gains": [],
733
+ "losses": [],
734
+ "rates": {
735
+ "fc2": 0.4,
736
+ "fresh_en": 0.4,
737
+ "fresh_fr": 0.4,
738
+ "test20k": 0.4,
739
+ "test_v3": 0.4,
740
+ "queries": 1.0
741
+ },
742
+ "tok_ok": false,
743
+ "ept_ok": {
744
+ "ref": true,
745
+ "cand": true
746
+ },
747
+ "seen_fisher_p": 0.41236097478755496,
748
+ "n_inputs_npz": 16
749
+ },
750
+ {
751
+ "case": "f1_broad_dans_eval",
752
+ "desc": "npz fc2 de sha e89725bb…",
753
+ "attendu": "AssertionError",
754
+ "obtenu": "AssertionError",
755
+ "checks": {},
756
+ "conforme": true,
757
+ "raisons": "ref:/workspace/spira_shared/hashing_gen/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_f1_broad_dans_eval/runs_ref/ctl__fc2.npz: window_sha256 e89725bb… (broad_*_test) — interdit en CS1"
758
+ },
759
+ {
760
+ "case": "f2_broad_fichier_en_plus",
761
+ "desc": "ctl__broad_local_v1_test.npz (e89725bb…) en plus",
762
+ "attendu": "AssertionError",
763
+ "obtenu": "AssertionError",
764
+ "checks": {},
765
+ "conforme": true,
766
+ "raisons": "ref:/workspace/spira_shared/hashing_gen/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_f2_broad_fichier_en_plus/runs_ref/ctl__broad_local_v1_test.npz: nom de fenêtre EXCLUE (broad_*_test) — interdit en CS1"
767
+ },
768
+ {
769
+ "case": "f3_nom_broad_sha_neutre",
770
+ "desc": "ctl__broad_gutenberg_v1_test.npz (nom exclu, sha fc2) côté cand",
771
+ "attendu": "AssertionError",
772
+ "obtenu": "AssertionError",
773
+ "checks": {},
774
+ "conforme": true,
775
+ "raisons": "cand:/workspace/spira_shared/hashing_gen/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_f3_nom_broad_sha_neutre/runs_cand/ctl__broad_gutenberg_v1_test.npz: nom de fenêtre EXCLUE (broad_*_test) — interdit en CS1"
776
+ },
777
+ {
778
+ "case": "f4_broad_sous_dossier",
779
+ "desc": "npz e89725bb… dans runs_ref/ckpt_ctl/x.npz",
780
+ "attendu": "AssertionError",
781
+ "obtenu": "AssertionError",
782
+ "checks": {},
783
+ "conforme": true,
784
+ "raisons": "ref:/workspace/spira_shared/hashing_gen/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_f4_broad_sous_dossier/runs_ref/ckpt_ctl/x.npz: window_sha256 e89725bb… (broad_*_test) — interdit en CS1"
785
+ },
786
+ {
787
+ "case": "f5_npz_modele_present",
788
+ "desc": "ctl.npz (ckpt EPT commune) + ckpt_ctl/epoch1.npz dans les deux dossiers, gain fc2",
789
+ "attendu": "ADOPTÉ",
790
+ "obtenu": "ADOPTÉ",
791
+ "checks": {
792
+ "n_inputs_npz": true
793
+ },
794
+ "conforme": true,
795
+ "raisons": "",
796
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
797
+ "revision": "r3-draft (NON FIGÉ)",
798
+ "fresh_valid": [
799
+ "fc2",
800
+ "fresh_en",
801
+ "fresh_fr",
802
+ "test20k",
803
+ "test_v3"
804
+ ],
805
+ "gains": [
806
+ "fc2"
807
+ ],
808
+ "losses": [],
809
+ "rates": {
810
+ "fc2": 1.0,
811
+ "fresh_en": 1.0,
812
+ "fresh_fr": 1.0,
813
+ "test20k": 1.0,
814
+ "test_v3": 1.0,
815
+ "queries": 1.0
816
+ },
817
+ "tok_ok": true,
818
+ "ept_ok": {
819
+ "ref": true,
820
+ "cand": true
821
+ },
822
+ "seen_fisher_p": 0.41236097478755496,
823
+ "n_inputs_npz": 18
824
+ },
825
+ {
826
+ "case": "g1_recompte_mix_absent",
827
+ "desc": "gain fc2, recompte mix absent",
828
+ "attendu": "NON MESURABLE",
829
+ "obtenu": "NON MESURABLE",
830
+ "checks": {},
831
+ "conforme": true,
832
+ "raisons": "aucune fenêtre fraîche valide",
833
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
834
+ "revision": "r3-draft (NON FIGÉ)",
835
+ "fresh_valid": [],
836
+ "gains": [],
837
+ "losses": [],
838
+ "rates": {
839
+ "fc2": 1.0,
840
+ "fresh_en": 1.0,
841
+ "fresh_fr": 1.0,
842
+ "test20k": 1.0,
843
+ "test_v3": 1.0,
844
+ "queries": 1.0
845
+ },
846
+ "tok_ok": true,
847
+ "ept_ok": {
848
+ "ref": true,
849
+ "cand": true
850
+ },
851
+ "seen_fisher_p": 0.41236097478755496,
852
+ "n_inputs_npz": 16
853
+ },
854
+ {
855
+ "case": "g2_recompte_mix_mauvais_sha",
856
+ "desc": "gain fc2, recompte mix sha faux",
857
+ "attendu": "NON MESURABLE",
858
+ "obtenu": "NON MESURABLE",
859
+ "checks": {},
860
+ "conforme": true,
861
+ "raisons": "aucune fenêtre fraîche valide",
862
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
863
+ "revision": "r3-draft (NON FIGÉ)",
864
+ "fresh_valid": [],
865
+ "gains": [],
866
+ "losses": [],
867
+ "rates": {
868
+ "fc2": 1.0,
869
+ "fresh_en": 1.0,
870
+ "fresh_fr": 1.0,
871
+ "test20k": 1.0,
872
+ "test_v3": 1.0,
873
+ "queries": 1.0
874
+ },
875
+ "tok_ok": true,
876
+ "ept_ok": {
877
+ "ref": true,
878
+ "cand": true
879
+ },
880
+ "seen_fisher_p": 0.41236097478755496,
881
+ "n_inputs_npz": 16
882
+ },
883
+ {
884
+ "case": "g3_recompte_mix_sha_fenetre_faux",
885
+ "desc": "gain fc2, window_sha256 fc2 faux au recompte mix",
886
+ "attendu": "AssertionError",
887
+ "obtenu": "AssertionError",
888
+ "checks": {},
889
+ "conforme": true,
890
+ "raisons": "fc2: window_sha256 recompte mix != npz (r2-e)"
891
+ },
892
+ {
893
+ "case": "h1_test_v3_sans_GO",
894
+ "desc": "gain test_v3 seul, GO_CS1 absent",
895
+ "attendu": "REJETÉ",
896
+ "obtenu": "REJETÉ",
897
+ "checks": {},
898
+ "conforme": true,
899
+ "raisons": "aucun gain significatif sur une fenêtre fraîche valide",
900
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
901
+ "revision": "r3-draft (NON FIGÉ)",
902
+ "fresh_valid": [
903
+ "fc2",
904
+ "fresh_en",
905
+ "fresh_fr",
906
+ "test20k"
907
+ ],
908
+ "gains": [],
909
+ "losses": [],
910
+ "rates": {
911
+ "fc2": 1.0,
912
+ "fresh_en": 1.0,
913
+ "fresh_fr": 1.0,
914
+ "test20k": 1.0,
915
+ "test_v3": 1.0,
916
+ "queries": 1.0
917
+ },
918
+ "tok_ok": true,
919
+ "ept_ok": {
920
+ "ref": true,
921
+ "cand": true
922
+ },
923
+ "seen_fisher_p": 0.41236097478755496,
924
+ "n_inputs_npz": 16
925
+ },
926
+ {
927
+ "case": "h2_budget_inegal",
928
+ "desc": "gain fc2, am_tokens cand 1 000 000",
929
+ "attendu": "NON MESURABLE",
930
+ "obtenu": "NON MESURABLE",
931
+ "checks": {},
932
+ "conforme": true,
933
+ "raisons": "budgets réf/cand inégaux : am_tokens",
934
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
935
+ "revision": "r3-draft (NON FIGÉ)",
936
+ "fresh_valid": [
937
+ "fc2",
938
+ "fresh_en",
939
+ "fresh_fr",
940
+ "test20k",
941
+ "test_v3"
942
+ ],
943
+ "gains": [
944
+ "fc2"
945
+ ],
946
+ "losses": [],
947
+ "rates": {
948
+ "fc2": 1.0,
949
+ "fresh_en": 1.0,
950
+ "fresh_fr": 1.0,
951
+ "test20k": 1.0,
952
+ "test_v3": 1.0,
953
+ "queries": 1.0
954
+ },
955
+ "tok_ok": true,
956
+ "ept_ok": {
957
+ "ref": true,
958
+ "cand": true
959
+ },
960
+ "seen_fisher_p": 0.41236097478755496,
961
+ "n_inputs_npz": 16
962
+ },
963
+ {
964
+ "case": "i1_bras_cm_refuse",
965
+ "desc": "--cand-arm cm",
966
+ "attendu": "SystemExit",
967
+ "obtenu": "SystemExit",
968
+ "checks": {},
969
+ "conforme": true,
970
+ "raisons": "2"
971
+ },
972
+ {
973
+ "case": "j1_ckpt_cand_absent",
974
+ "desc": "gain fc2, ckpt cand absent (EPT invérifiable)",
975
+ "attendu": "NON MESURABLE",
976
+ "obtenu": "NON MESURABLE",
977
+ "checks": {
978
+ "reason_contains": true
979
+ },
980
+ "conforme": true,
981
+ "raisons": "tokenisation commune non vérifiée (r3-a) : cand ckpt absent",
982
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
983
+ "revision": "r3-draft (NON FIGÉ)",
984
+ "fresh_valid": [
985
+ "fc2",
986
+ "fresh_en",
987
+ "fresh_fr",
988
+ "test20k",
989
+ "test_v3"
990
+ ],
991
+ "gains": [
992
+ "fc2"
993
+ ],
994
+ "losses": [],
995
+ "rates": {
996
+ "fc2": 1.0,
997
+ "fresh_en": 1.0,
998
+ "fresh_fr": 1.0,
999
+ "test20k": 1.0,
1000
+ "test_v3": 1.0,
1001
+ "queries": 1.0
1002
+ },
1003
+ "tok_ok": false,
1004
+ "ept_ok": {
1005
+ "ref": true,
1006
+ "cand": false
1007
+ },
1008
+ "seen_fisher_p": 0.41236097478755496,
1009
+ "n_inputs_npz": 15
1010
+ },
1011
+ {
1012
+ "case": "j2_ept_cand_propre_mix",
1013
+ "desc": "gain fc2, cand entraîné avec SON EPT (ajustée sur cs1_mix_sample, comme r2)",
1014
+ "attendu": "NON MESURABLE",
1015
+ "obtenu": "NON MESURABLE",
1016
+ "checks": {
1017
+ "reason_contains": true
1018
+ },
1019
+ "conforme": true,
1020
+ "raisons": "tokenisation commune non vérifiée (r3-a) : cand log2p ≠ EPT commune, ept_params ≠ EPT commune",
1021
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
1022
+ "revision": "r3-draft (NON FIGÉ)",
1023
+ "fresh_valid": [
1024
+ "fc2",
1025
+ "fresh_en",
1026
+ "fresh_fr",
1027
+ "test20k",
1028
+ "test_v3"
1029
+ ],
1030
+ "gains": [
1031
+ "fc2"
1032
+ ],
1033
+ "losses": [],
1034
+ "rates": {
1035
+ "fc2": 1.0,
1036
+ "fresh_en": 1.0,
1037
+ "fresh_fr": 1.0,
1038
+ "test20k": 1.0,
1039
+ "test_v3": 1.0,
1040
+ "queries": 1.0
1041
+ },
1042
+ "tok_ok": false,
1043
+ "ept_ok": {
1044
+ "ref": true,
1045
+ "cand": false
1046
+ },
1047
+ "seen_fisher_p": 0.41236097478755496,
1048
+ "n_inputs_npz": 16
1049
+ },
1050
+ {
1051
+ "case": "j3_budget_bits_cand_different",
1052
+ "desc": "gain fc2, log2p commune mais budget_bits cand 15,489",
1053
+ "attendu": "NON MESURABLE",
1054
+ "obtenu": "NON MESURABLE",
1055
+ "checks": {
1056
+ "reason_contains": true
1057
+ },
1058
+ "conforme": true,
1059
+ "raisons": "tokenisation commune non vérifiée (r3-a) : cand ept_params ≠ EPT commune",
1060
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
1061
+ "revision": "r3-draft (NON FIGÉ)",
1062
+ "fresh_valid": [
1063
+ "fc2",
1064
+ "fresh_en",
1065
+ "fresh_fr",
1066
+ "test20k",
1067
+ "test_v3"
1068
+ ],
1069
+ "gains": [
1070
+ "fc2"
1071
+ ],
1072
+ "losses": [],
1073
+ "rates": {
1074
+ "fc2": 1.0,
1075
+ "fresh_en": 1.0,
1076
+ "fresh_fr": 1.0,
1077
+ "test20k": 1.0,
1078
+ "test_v3": 1.0,
1079
+ "queries": 1.0
1080
+ },
1081
+ "tok_ok": false,
1082
+ "ept_ok": {
1083
+ "ref": true,
1084
+ "cand": false
1085
+ },
1086
+ "seen_fisher_p": 0.41236097478755496,
1087
+ "n_inputs_npz": 16
1088
+ },
1089
+ {
1090
+ "case": "j4_ckpt_sha_different_results",
1091
+ "desc": "gain fc2, sha du ckpt cand ≠ ckpt_sha256 de results.jsonl",
1092
+ "attendu": "NON MESURABLE",
1093
+ "obtenu": "NON MESURABLE",
1094
+ "checks": {
1095
+ "reason_contains": true
1096
+ },
1097
+ "conforme": true,
1098
+ "raisons": "tokenisation commune non vérifiée (r3-a) : cand sha ckpt ≠ ckpt_sha256 results.jsonl",
1099
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
1100
+ "revision": "r3-draft (NON FIGÉ)",
1101
+ "fresh_valid": [
1102
+ "fc2",
1103
+ "fresh_en",
1104
+ "fresh_fr",
1105
+ "test20k",
1106
+ "test_v3"
1107
+ ],
1108
+ "gains": [
1109
+ "fc2"
1110
+ ],
1111
+ "losses": [],
1112
+ "rates": {
1113
+ "fc2": 1.0,
1114
+ "fresh_en": 1.0,
1115
+ "fresh_fr": 1.0,
1116
+ "test20k": 1.0,
1117
+ "test_v3": 1.0,
1118
+ "queries": 1.0
1119
+ },
1120
+ "tok_ok": false,
1121
+ "ept_ok": {
1122
+ "ref": true,
1123
+ "cand": false
1124
+ },
1125
+ "seen_fisher_p": 0.41236097478755496,
1126
+ "n_inputs_npz": 16
1127
+ },
1128
+ {
1129
+ "case": "j5_ckpt_ref_sans_log2p",
1130
+ "desc": "gain fc2, ckpt réf sans log2p / ept_params",
1131
+ "attendu": "NON MESURABLE",
1132
+ "obtenu": "NON MESURABLE",
1133
+ "checks": {
1134
+ "reason_contains": true
1135
+ },
1136
+ "conforme": true,
1137
+ "raisons": "tokenisation commune non vérifiée (r3-a) : ref log2p / ept_params absents du ckpt",
1138
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
1139
+ "revision": "r3-draft (NON FIGÉ)",
1140
+ "fresh_valid": [
1141
+ "fc2",
1142
+ "fresh_en",
1143
+ "fresh_fr",
1144
+ "test20k",
1145
+ "test_v3"
1146
+ ],
1147
+ "gains": [
1148
+ "fc2"
1149
+ ],
1150
+ "losses": [],
1151
+ "rates": {
1152
+ "fc2": 1.0,
1153
+ "fresh_en": 1.0,
1154
+ "fresh_fr": 1.0,
1155
+ "test20k": 1.0,
1156
+ "test_v3": 1.0,
1157
+ "queries": 1.0
1158
+ },
1159
+ "tok_ok": false,
1160
+ "ept_ok": {
1161
+ "ref": false,
1162
+ "cand": true
1163
+ },
1164
+ "seen_fisher_p": 0.41236097478755496,
1165
+ "n_inputs_npz": 16
1166
+ },
1167
+ {
1168
+ "case": "j6_log2p_float64",
1169
+ "desc": "gain fc2, log2p réf en float64 (mêmes valeurs ; SPIRA.save écrit float32)",
1170
+ "attendu": "NON MESURABLE",
1171
+ "obtenu": "NON MESURABLE",
1172
+ "checks": {
1173
+ "reason_contains": true
1174
+ },
1175
+ "conforme": true,
1176
+ "raisons": "tokenisation commune non vérifiée (r3-a) : ref log2p ≠ EPT commune",
1177
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
1178
+ "revision": "r3-draft (NON FIGÉ)",
1179
+ "fresh_valid": [
1180
+ "fc2",
1181
+ "fresh_en",
1182
+ "fresh_fr",
1183
+ "test20k",
1184
+ "test_v3"
1185
+ ],
1186
+ "gains": [
1187
+ "fc2"
1188
+ ],
1189
+ "losses": [],
1190
+ "rates": {
1191
+ "fc2": 1.0,
1192
+ "fresh_en": 1.0,
1193
+ "fresh_fr": 1.0,
1194
+ "test20k": 1.0,
1195
+ "test_v3": 1.0,
1196
+ "queries": 1.0
1197
+ },
1198
+ "tok_ok": false,
1199
+ "ept_ok": {
1200
+ "ref": false,
1201
+ "cand": true
1202
+ },
1203
+ "seen_fisher_p": 0.41236097478755496,
1204
+ "n_inputs_npz": 16
1205
+ },
1206
+ {
1207
+ "case": "j7_eval_cand_tronquee_3000",
1208
+ "desc": "gain fc2, éval cand fc2 tronquée à 3 000 positions (même préfixe) : plage commune, taux 1,000",
1209
+ "attendu": "ADOPTÉ",
1210
+ "obtenu": "ADOPTÉ",
1211
+ "checks": {
1212
+ "all_rates_1": true
1213
+ },
1214
+ "conforme": true,
1215
+ "raisons": "",
1216
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
1217
+ "revision": "r3-draft (NON FIGÉ)",
1218
+ "fresh_valid": [
1219
+ "fc2",
1220
+ "fresh_en",
1221
+ "fresh_fr",
1222
+ "test20k",
1223
+ "test_v3"
1224
+ ],
1225
+ "gains": [
1226
+ "fc2"
1227
+ ],
1228
+ "losses": [],
1229
+ "rates": {
1230
+ "fc2": 1.0,
1231
+ "fresh_en": 1.0,
1232
+ "fresh_fr": 1.0,
1233
+ "test20k": 1.0,
1234
+ "test_v3": 1.0,
1235
+ "queries": 1.0
1236
+ },
1237
+ "tok_ok": true,
1238
+ "ept_ok": {
1239
+ "ref": true,
1240
+ "cand": true
1241
+ },
1242
+ "seen_fisher_p": 0.41236097478755496,
1243
+ "n_inputs_npz": 16
1244
+ },
1245
+ {
1246
+ "case": "j8_cibles_queries_differentes",
1247
+ "desc": "gain fc2, cibles différentes sur queries seulement (hors règle)",
1248
+ "attendu": "NON MESURABLE",
1249
+ "obtenu": "NON MESURABLE",
1250
+ "checks": {
1251
+ "reason_contains": true
1252
+ },
1253
+ "conforme": true,
1254
+ "raisons": "cibles non identiques sur la plage commune (r3-b) : queries",
1255
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
1256
+ "revision": "r3-draft (NON FIGÉ)",
1257
+ "fresh_valid": [
1258
+ "fc2",
1259
+ "fresh_en",
1260
+ "fresh_fr",
1261
+ "test20k",
1262
+ "test_v3"
1263
+ ],
1264
+ "gains": [
1265
+ "fc2"
1266
+ ],
1267
+ "losses": [],
1268
+ "rates": {
1269
+ "fc2": 1.0,
1270
+ "fresh_en": 1.0,
1271
+ "fresh_fr": 1.0,
1272
+ "test20k": 1.0,
1273
+ "test_v3": 1.0,
1274
+ "queries": 0.6004
1275
+ },
1276
+ "tok_ok": false,
1277
+ "ept_ok": {
1278
+ "ref": true,
1279
+ "cand": true
1280
+ },
1281
+ "seen_fisher_p": 0.41236097478755496,
1282
+ "n_inputs_npz": 16
1283
+ },
1284
+ {
1285
+ "case": "j9_ept_commune_les_deux_perte",
1286
+ "desc": "gain fc2 + perte fresh_fr, EPT commune OK",
1287
+ "attendu": "REJETÉ",
1288
+ "obtenu": "REJETÉ",
1289
+ "checks": {},
1290
+ "conforme": true,
1291
+ "raisons": "perte(s) significative(s) : fresh_fr",
1292
+ "script_sha256": "31f3d1374248d5aed17064862dbb29737af331d437985253cf7823ba7a0b3430",
1293
+ "revision": "r3-draft (NON FIGÉ)",
1294
+ "fresh_valid": [
1295
+ "fc2",
1296
+ "fresh_en",
1297
+ "fresh_fr",
1298
+ "test20k",
1299
+ "test_v3"
1300
+ ],
1301
+ "gains": [
1302
+ "fc2"
1303
+ ],
1304
+ "losses": [
1305
+ "fresh_fr"
1306
+ ],
1307
+ "rates": {
1308
+ "fc2": 1.0,
1309
+ "fresh_en": 1.0,
1310
+ "fresh_fr": 1.0,
1311
+ "test20k": 1.0,
1312
+ "test_v3": 1.0,
1313
+ "queries": 1.0
1314
+ },
1315
+ "tok_ok": true,
1316
+ "ept_ok": {
1317
+ "ref": true,
1318
+ "cand": true
1319
+ },
1320
+ "seen_fisher_p": 0.41236097478755496,
1321
+ "n_inputs_npz": 16
1322
+ }
1323
+ ]
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/RESULTS_SYNTH_r3.md ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Tests SYNTHÉTIQUES verdict_CS1.r3_draft.py — BROUILLON r3 — 40/40 conformes
2
+
3
+ | cas | description | attendu r3 | obtenu | contrôles en plus | conforme | raisons (script) |
4
+ |---|---|---|---|---|---|---|
5
+ | a_gain_frais | gain significatif sur fc2, aucune autre différence ; r3 : taux 1,000 partout | ADOPTÉ | **ADOPTÉ** | all_rates_1=True | oui | |
6
+ | b1_gain_perte_fraiche | gain fc2 + perte significative fresh_en | REJETÉ | **REJETÉ** | — | oui | perte(s) significative(s) : fresh_en |
7
+ | b2_gain_perte_seen | gain fc2 + seen cand à 0 % (garde seen non apparié) | REJETÉ | **REJETÉ** | — | oui | perte(s) significative(s) : seen (non apparié, Fisher) |
8
+ | b3_gain_perte_queries | gain fc2 + perte significative queries (hors règle) | ADOPTÉ | **ADOPTÉ** | — | oui | |
9
+ | b4_gain_perte_test20k | gain fc2 + perte significative test20k strict | REJETÉ | **REJETÉ** | — | oui | perte(s) significative(s) : test20k |
10
+ | b5_seen_un_peu_bas_ns | gain fc2 + seen cand −20 hits / 1 000, non significatif | ADOPTÉ | **ADOPTÉ** | — | oui | |
11
+ | b6_seen_cand_plus_haut | gain fc2 + seen cand +200 hits / 1 000 | ADOPTÉ | **ADOPTÉ** | — | oui | |
12
+ | b7_seen_perte_sans_gain | aucun gain + seen cand à 0 % | REJETÉ | **REJETÉ** | — | oui | aucun gain significatif sur une fenêtre fraîche valide; perte(s) significative(s) : seen (non apparié, Fisher) |
13
+ | c1_aucun_gain | bras identiques | REJETÉ | **REJETÉ** | — | oui | aucun gain significatif sur une fenêtre fraîche valide |
14
+ | c2_gain_queries_seul | gain seulement sur queries (hors règle) | REJETÉ | **REJETÉ** | — | oui | aucun gain significatif sur une fenêtre fraîche valide |
15
+ | d1_scale_ref_bas | gain fc2, fusion.scale réf 0,02 | NON MESURABLE | **NON MESURABLE** | — | oui | fusion.scale hors [0.03 ; 0.11] : réf 0.02 / cand 0.07 |
16
+ | d2_scale_cand_haut | gain fc2, fusion.scale cand 0,12 | NON MESURABLE | **NON MESURABLE** | — | oui | fusion.scale hors [0.03 ; 0.11] : réf 0.07 / cand 0.12 |
17
+ | d3_scale_bornes | gain fc2, fusion.scale 0,03 / 0,11 | ADOPTÉ | **ADOPTÉ** | — | oui | |
18
+ | d4_scale_absent | gain fc2, cal.scale absent côté cand | NON MESURABLE | **NON MESURABLE** | — | oui | fusion.scale hors [0.03 ; 0.11] : réf 0.07 / cand None |
19
+ | e1_appariement_60 | cibles fc2 différentes (taux plage commune 0,60), gain fc2 | NON MESURABLE | **NON MESURABLE** | reason_contains=True, window_reason_absent=True | oui | cibles non identiques sur la plage commune (r3-b) : fc2 |
20
+ | e2_appariement_40 | cibles fc2 différentes (taux 0,40), gain fc2 | NON MESURABLE | **NON MESURABLE** | reason_contains=True, window_reason_contains=True | oui | cibles non identiques sur la plage commune (r3-b) : fc2 |
21
+ | e3_appariement_50 | cibles fc2 différentes (taux 0,5002), gain fc2 | NON MESURABLE | **NON MESURABLE** | reason_contains=True, window_reason_absent=True | oui | cibles non identiques sur la plage commune (r3-b) : fc2 |
22
+ | e3b_appariement_4996 | cibles fc2 différentes (taux 0,4996), gain fc2 | NON MESURABLE | **NON MESURABLE** | reason_contains=True, window_reason_contains=True | oui | cibles non identiques sur la plage commune (r3-b) : fc2 |
23
+ | e4_appariement_45_denominateur | cibles fc2 différentes (taux 0,62), perte fc2 + gain fresh_en | NON MESURABLE | **NON MESURABLE** | reason_contains=True | oui | cibles non identiques sur la plage commune (r3-b) : fc2 |
24
+ | e5_tout_sous_50 | taux 0,40 sur les 5 fraîches | NON MESURABLE | **NON MESURABLE** | reason_contains=True | oui | cibles non identiques sur la plage commune (r3-b) : fc2, fresh_en, fresh_fr, test20k, test_v3; aucune fenêtre fraîche valide |
25
+ | f1_broad_dans_eval | npz fc2 de sha e89725bb… | AssertionError | **AssertionError** | — | oui | ref:/workspace/spira_shared/hashing_gen/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_f1_broad_dans_eval/runs_ref/ctl__fc2.npz: window_sha256 e89725bb… (broad_*_ |
26
+ | f2_broad_fichier_en_plus | ctl__broad_local_v1_test.npz (e89725bb…) en plus | AssertionError | **AssertionError** | — | oui | ref:/workspace/spira_shared/hashing_gen/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_f2_broad_fichier_en_plus/runs_ref/ctl__broad_local_v1_test.npz: nom de fenê |
27
+ | f3_nom_broad_sha_neutre | ctl__broad_gutenberg_v1_test.npz (nom exclu, sha fc2) côté cand | AssertionError | **AssertionError** | — | oui | cand:/workspace/spira_shared/hashing_gen/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_f3_nom_broad_sha_neutre/runs_cand/ctl__broad_gutenberg_v1_test.npz: nom de |
28
+ | f4_broad_sous_dossier | npz e89725bb… dans runs_ref/ckpt_ctl/x.npz | AssertionError | **AssertionError** | — | oui | ref:/workspace/spira_shared/hashing_gen/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_f4_broad_sous_dossier/runs_ref/ckpt_ctl/x.npz: window_sha256 e89725bb… (bro |
29
+ | f5_npz_modele_present | ctl.npz (ckpt EPT commune) + ckpt_ctl/epoch1.npz dans les deux dossiers, gain fc2 | ADOPTÉ | **ADOPTÉ** | n_inputs_npz=True | oui | |
30
+ | g1_recompte_mix_absent | gain fc2, recompte mix absent | NON MESURABLE | **NON MESURABLE** | — | oui | aucune fenêtre fraîche valide |
31
+ | g2_recompte_mix_mauvais_sha | gain fc2, recompte mix sha faux | NON MESURABLE | **NON MESURABLE** | — | oui | aucune fenêtre fraîche valide |
32
+ | g3_recompte_mix_sha_fenetre_faux | gain fc2, window_sha256 fc2 faux au recompte mix | AssertionError | **AssertionError** | — | oui | fc2: window_sha256 recompte mix != npz (r2-e) |
33
+ | h1_test_v3_sans_GO | gain test_v3 seul, GO_CS1 absent | REJETÉ | **REJETÉ** | — | oui | aucun gain significatif sur une fenêtre fraîche valide |
34
+ | h2_budget_inegal | gain fc2, am_tokens cand 1 000 000 | NON MESURABLE | **NON MESURABLE** | — | oui | budgets réf/cand inégaux : am_tokens |
35
+ | i1_bras_cm_refuse | --cand-arm cm | SystemExit | **SystemExit** | — | oui | 2 |
36
+ | j1_ckpt_cand_absent | gain fc2, ckpt cand absent (EPT invérifiable) | NON MESURABLE | **NON MESURABLE** | reason_contains=True | oui | tokenisation commune non vérifiée (r3-a) : cand ckpt absent |
37
+ | j2_ept_cand_propre_mix | gain fc2, cand entraîné avec SON EPT (ajustée sur cs1_mix_sample, comme r2) | NON MESURABLE | **NON MESURABLE** | reason_contains=True | oui | tokenisation commune non vérifiée (r3-a) : cand log2p ≠ EPT commune, ept_params ≠ EPT commune |
38
+ | j3_budget_bits_cand_different | gain fc2, log2p commune mais budget_bits cand 15,489 | NON MESURABLE | **NON MESURABLE** | reason_contains=True | oui | tokenisation commune non vérifiée (r3-a) : cand ept_params ≠ EPT commune |
39
+ | j4_ckpt_sha_different_results | gain fc2, sha du ckpt cand ≠ ckpt_sha256 de results.jsonl | NON MESURABLE | **NON MESURABLE** | reason_contains=True | oui | tokenisation commune non vérifiée (r3-a) : cand sha ckpt ≠ ckpt_sha256 results.jsonl |
40
+ | j5_ckpt_ref_sans_log2p | gain fc2, ckpt réf sans log2p / ept_params | NON MESURABLE | **NON MESURABLE** | reason_contains=True | oui | tokenisation commune non vérifiée (r3-a) : ref log2p / ept_params absents du ckpt |
41
+ | j6_log2p_float64 | gain fc2, log2p réf en float64 (mêmes valeurs ; SPIRA.save écrit float32) | NON MESURABLE | **NON MESURABLE** | reason_contains=True | oui | tokenisation commune non vérifiée (r3-a) : ref log2p ≠ EPT commune |
42
+ | j7_eval_cand_tronquee_3000 | gain fc2, éval cand fc2 tronquée à 3 000 positions (même préfixe) : plage commune, taux 1,000 | ADOPTÉ | **ADOPTÉ** | all_rates_1=True | oui | |
43
+ | j8_cibles_queries_differentes | gain fc2, cibles différentes sur queries seulement (hors règle) | NON MESURABLE | **NON MESURABLE** | reason_contains=True | oui | cibles non identiques sur la plage commune (r3-b) : queries |
44
+ | j9_ept_commune_les_deux_perte | gain fc2 + perte fresh_fr, EPT commune OK | REJETÉ | **REJETÉ** | — | oui | perte(s) significative(s) : fresh_fr |
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/aux/SYNTH_GO_CS1 ADDED
@@ -0,0 +1 @@
 
 
1
+ SYNTHETIQUE — drapeau factice de test
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/aux/SYNTH_recount64_cs1_mix_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "85e977700bbfc4ad571d4be15623450eaa4360fdfd1fe9451b0706f81bfc1453",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/aux/SYNTH_recount64_v1b_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "9ea562e44f708adda470c53144dcae748796d77b98039b5fc52ac962d877370b",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/aux/SYNTH_scan64_code_scale_v1.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "PASS": true,
4
+ "test_mode": false,
5
+ "stage_A": {
6
+ "code_delta": {
7
+ "per_window": {
8
+ "fc2": {
9
+ "covered_window_byte_positions": 0
10
+ },
11
+ "m7_fresh_en": {
12
+ "covered_window_byte_positions": 0
13
+ },
14
+ "m7_fresh_fr": {
15
+ "covered_window_byte_positions": 0
16
+ },
17
+ "m7_test20k": {
18
+ "covered_window_byte_positions": 0
19
+ },
20
+ "corpus_v3_test": {
21
+ "covered_window_byte_positions": 0
22
+ },
23
+ "m7_queries": {
24
+ "covered_window_byte_positions": 0
25
+ }
26
+ }
27
+ }
28
+ }
29
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/case.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "case": "a_gain_frais",
3
+ "description": "gain significatif sur fc2, aucune autre différence ; r3 : taux 1,000 partout",
4
+ "expected_consigne": "ADOPTÉ",
5
+ "expected_predecl": "ADOPTÉ",
6
+ "SYNTHETIQUE": true,
7
+ "extra_args": [],
8
+ "checks": {
9
+ "all_rates_1": true
10
+ },
11
+ "r3": true
12
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/runs_cand/results.jsonl ADDED
@@ -0,0 +1 @@
 
 
1
+ {"arm": "ctl", "snn_tokens": 600000, "am_tokens": 1200000, "epochs": 2, "ticks_per_epoch": [600000, 600000], "data_sha256": "85e977700bbfc4ad571d4be15623450eaa4360fdfd1fe9451b0706f81bfc1453", "cal": {"scale": 0.07}, "SYNTHETIQUE": "r\u00e9sultat d'entra\u00eenement FACTICE", "ckpt_sha256": "4b7cbbd3458ffe94acbb78549ec87e3ea9f34170bafd289aa19e6ca4f50e7a51"}
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_a_gain_frais/runs_ref/results.jsonl ADDED
@@ -0,0 +1 @@
 
 
1
+ {"arm": "ctl", "snn_tokens": 600000, "am_tokens": 1200000, "epochs": 2, "ticks_per_epoch": [600000, 600000], "data_sha256": "9ea562e44f708adda470c53144dcae748796d77b98039b5fc52ac962d877370b", "cal": {"scale": 0.07}, "SYNTHETIQUE": "r\u00e9sultat d'entra\u00eenement FACTICE", "ckpt_sha256": "4b7cbbd3458ffe94acbb78549ec87e3ea9f34170bafd289aa19e6ca4f50e7a51"}
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/aux/SYNTH_GO_CS1 ADDED
@@ -0,0 +1 @@
 
 
1
+ SYNTHETIQUE — drapeau factice de test
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/aux/SYNTH_recount64_cs1_mix_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "85e977700bbfc4ad571d4be15623450eaa4360fdfd1fe9451b0706f81bfc1453",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/aux/SYNTH_recount64_v1b_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "9ea562e44f708adda470c53144dcae748796d77b98039b5fc52ac962d877370b",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/aux/SYNTH_scan64_code_scale_v1.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "PASS": true,
4
+ "test_mode": false,
5
+ "stage_A": {
6
+ "code_delta": {
7
+ "per_window": {
8
+ "fc2": {
9
+ "covered_window_byte_positions": 0
10
+ },
11
+ "m7_fresh_en": {
12
+ "covered_window_byte_positions": 0
13
+ },
14
+ "m7_fresh_fr": {
15
+ "covered_window_byte_positions": 0
16
+ },
17
+ "m7_test20k": {
18
+ "covered_window_byte_positions": 0
19
+ },
20
+ "corpus_v3_test": {
21
+ "covered_window_byte_positions": 0
22
+ },
23
+ "m7_queries": {
24
+ "covered_window_byte_positions": 0
25
+ }
26
+ }
27
+ }
28
+ }
29
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/case.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "case": "b1_gain_perte_fraiche",
3
+ "description": "gain fc2 + perte significative fresh_en",
4
+ "expected_consigne": "REJETÉ",
5
+ "expected_predecl": "REJETÉ",
6
+ "SYNTHETIQUE": true,
7
+ "extra_args": [],
8
+ "checks": {},
9
+ "r3": true
10
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/runs_cand/results.jsonl ADDED
@@ -0,0 +1 @@
 
 
1
+ {"arm": "ctl", "snn_tokens": 600000, "am_tokens": 1200000, "epochs": 2, "ticks_per_epoch": [600000, 600000], "data_sha256": "85e977700bbfc4ad571d4be15623450eaa4360fdfd1fe9451b0706f81bfc1453", "cal": {"scale": 0.07}, "SYNTHETIQUE": "r\u00e9sultat d'entra\u00eenement FACTICE", "ckpt_sha256": "4b7cbbd3458ffe94acbb78549ec87e3ea9f34170bafd289aa19e6ca4f50e7a51"}
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b1_gain_perte_fraiche/runs_ref/results.jsonl ADDED
@@ -0,0 +1 @@
 
 
1
+ {"arm": "ctl", "snn_tokens": 600000, "am_tokens": 1200000, "epochs": 2, "ticks_per_epoch": [600000, 600000], "data_sha256": "9ea562e44f708adda470c53144dcae748796d77b98039b5fc52ac962d877370b", "cal": {"scale": 0.07}, "SYNTHETIQUE": "r\u00e9sultat d'entra\u00eenement FACTICE", "ckpt_sha256": "4b7cbbd3458ffe94acbb78549ec87e3ea9f34170bafd289aa19e6ca4f50e7a51"}
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/aux/SYNTH_GO_CS1 ADDED
@@ -0,0 +1 @@
 
 
1
+ SYNTHETIQUE — drapeau factice de test
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/aux/SYNTH_recount64_cs1_mix_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "85e977700bbfc4ad571d4be15623450eaa4360fdfd1fe9451b0706f81bfc1453",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/aux/SYNTH_recount64_v1b_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "9ea562e44f708adda470c53144dcae748796d77b98039b5fc52ac962d877370b",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/aux/SYNTH_scan64_code_scale_v1.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "PASS": true,
4
+ "test_mode": false,
5
+ "stage_A": {
6
+ "code_delta": {
7
+ "per_window": {
8
+ "fc2": {
9
+ "covered_window_byte_positions": 0
10
+ },
11
+ "m7_fresh_en": {
12
+ "covered_window_byte_positions": 0
13
+ },
14
+ "m7_fresh_fr": {
15
+ "covered_window_byte_positions": 0
16
+ },
17
+ "m7_test20k": {
18
+ "covered_window_byte_positions": 0
19
+ },
20
+ "corpus_v3_test": {
21
+ "covered_window_byte_positions": 0
22
+ },
23
+ "m7_queries": {
24
+ "covered_window_byte_positions": 0
25
+ }
26
+ }
27
+ }
28
+ }
29
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/case.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "case": "b2_gain_perte_seen",
3
+ "description": "gain fc2 + seen cand à 0 % (garde seen non apparié)",
4
+ "expected_consigne": "REJETÉ",
5
+ "expected_predecl": "REJETÉ",
6
+ "SYNTHETIQUE": true,
7
+ "extra_args": [],
8
+ "checks": {},
9
+ "r3": true
10
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/runs_cand/results.jsonl ADDED
@@ -0,0 +1 @@
 
 
1
+ {"arm": "ctl", "snn_tokens": 600000, "am_tokens": 1200000, "epochs": 2, "ticks_per_epoch": [600000, 600000], "data_sha256": "85e977700bbfc4ad571d4be15623450eaa4360fdfd1fe9451b0706f81bfc1453", "cal": {"scale": 0.07}, "SYNTHETIQUE": "r\u00e9sultat d'entra\u00eenement FACTICE", "ckpt_sha256": "4b7cbbd3458ffe94acbb78549ec87e3ea9f34170bafd289aa19e6ca4f50e7a51"}
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b2_gain_perte_seen/runs_ref/results.jsonl ADDED
@@ -0,0 +1 @@
 
 
1
+ {"arm": "ctl", "snn_tokens": 600000, "am_tokens": 1200000, "epochs": 2, "ticks_per_epoch": [600000, 600000], "data_sha256": "9ea562e44f708adda470c53144dcae748796d77b98039b5fc52ac962d877370b", "cal": {"scale": 0.07}, "SYNTHETIQUE": "r\u00e9sultat d'entra\u00eenement FACTICE", "ckpt_sha256": "4b7cbbd3458ffe94acbb78549ec87e3ea9f34170bafd289aa19e6ca4f50e7a51"}
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/aux/SYNTH_GO_CS1 ADDED
@@ -0,0 +1 @@
 
 
1
+ SYNTHETIQUE — drapeau factice de test
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/aux/SYNTH_recount64_cs1_mix_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "85e977700bbfc4ad571d4be15623450eaa4360fdfd1fe9451b0706f81bfc1453",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/aux/SYNTH_recount64_v1b_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "9ea562e44f708adda470c53144dcae748796d77b98039b5fc52ac962d877370b",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/aux/SYNTH_scan64_code_scale_v1.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "PASS": true,
4
+ "test_mode": false,
5
+ "stage_A": {
6
+ "code_delta": {
7
+ "per_window": {
8
+ "fc2": {
9
+ "covered_window_byte_positions": 0
10
+ },
11
+ "m7_fresh_en": {
12
+ "covered_window_byte_positions": 0
13
+ },
14
+ "m7_fresh_fr": {
15
+ "covered_window_byte_positions": 0
16
+ },
17
+ "m7_test20k": {
18
+ "covered_window_byte_positions": 0
19
+ },
20
+ "corpus_v3_test": {
21
+ "covered_window_byte_positions": 0
22
+ },
23
+ "m7_queries": {
24
+ "covered_window_byte_positions": 0
25
+ }
26
+ }
27
+ }
28
+ }
29
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/case.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "case": "b5_seen_un_peu_bas_ns",
3
+ "description": "gain fc2 + seen cand −20 hits / 1 000, non significatif",
4
+ "expected_consigne": "ADOPTÉ",
5
+ "expected_predecl": "ADOPTÉ",
6
+ "SYNTHETIQUE": true,
7
+ "extra_args": [],
8
+ "checks": {},
9
+ "r3": true
10
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/runs_cand/results.jsonl ADDED
@@ -0,0 +1 @@
 
 
1
+ {"arm": "ctl", "snn_tokens": 600000, "am_tokens": 1200000, "epochs": 2, "ticks_per_epoch": [600000, 600000], "data_sha256": "85e977700bbfc4ad571d4be15623450eaa4360fdfd1fe9451b0706f81bfc1453", "cal": {"scale": 0.07}, "SYNTHETIQUE": "r\u00e9sultat d'entra\u00eenement FACTICE", "ckpt_sha256": "4b7cbbd3458ffe94acbb78549ec87e3ea9f34170bafd289aa19e6ca4f50e7a51"}
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b5_seen_un_peu_bas_ns/runs_ref/results.jsonl ADDED
@@ -0,0 +1 @@
 
 
1
+ {"arm": "ctl", "snn_tokens": 600000, "am_tokens": 1200000, "epochs": 2, "ticks_per_epoch": [600000, 600000], "data_sha256": "9ea562e44f708adda470c53144dcae748796d77b98039b5fc52ac962d877370b", "cal": {"scale": 0.07}, "SYNTHETIQUE": "r\u00e9sultat d'entra\u00eenement FACTICE", "ckpt_sha256": "4b7cbbd3458ffe94acbb78549ec87e3ea9f34170bafd289aa19e6ca4f50e7a51"}
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/aux/SYNTH_GO_CS1 ADDED
@@ -0,0 +1 @@
 
 
1
+ SYNTHETIQUE — drapeau factice de test
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/aux/SYNTH_recount64_cs1_mix_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "85e977700bbfc4ad571d4be15623450eaa4360fdfd1fe9451b0706f81bfc1453",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/aux/SYNTH_recount64_v1b_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "9ea562e44f708adda470c53144dcae748796d77b98039b5fc52ac962d877370b",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/aux/SYNTH_scan64_code_scale_v1.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "PASS": true,
4
+ "test_mode": false,
5
+ "stage_A": {
6
+ "code_delta": {
7
+ "per_window": {
8
+ "fc2": {
9
+ "covered_window_byte_positions": 0
10
+ },
11
+ "m7_fresh_en": {
12
+ "covered_window_byte_positions": 0
13
+ },
14
+ "m7_fresh_fr": {
15
+ "covered_window_byte_positions": 0
16
+ },
17
+ "m7_test20k": {
18
+ "covered_window_byte_positions": 0
19
+ },
20
+ "corpus_v3_test": {
21
+ "covered_window_byte_positions": 0
22
+ },
23
+ "m7_queries": {
24
+ "covered_window_byte_positions": 0
25
+ }
26
+ }
27
+ }
28
+ }
29
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/case.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "case": "b6_seen_cand_plus_haut",
3
+ "description": "gain fc2 + seen cand +200 hits / 1 000",
4
+ "expected_consigne": "ADOPTÉ",
5
+ "expected_predecl": "ADOPTÉ",
6
+ "SYNTHETIQUE": true,
7
+ "extra_args": [],
8
+ "checks": {},
9
+ "r3": true
10
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/runs_cand/results.jsonl ADDED
@@ -0,0 +1 @@
 
 
1
+ {"arm": "ctl", "snn_tokens": 600000, "am_tokens": 1200000, "epochs": 2, "ticks_per_epoch": [600000, 600000], "data_sha256": "85e977700bbfc4ad571d4be15623450eaa4360fdfd1fe9451b0706f81bfc1453", "cal": {"scale": 0.07}, "SYNTHETIQUE": "r\u00e9sultat d'entra\u00eenement FACTICE", "ckpt_sha256": "4b7cbbd3458ffe94acbb78549ec87e3ea9f34170bafd289aa19e6ca4f50e7a51"}
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b6_seen_cand_plus_haut/runs_ref/results.jsonl ADDED
@@ -0,0 +1 @@
 
 
1
+ {"arm": "ctl", "snn_tokens": 600000, "am_tokens": 1200000, "epochs": 2, "ticks_per_epoch": [600000, 600000], "data_sha256": "9ea562e44f708adda470c53144dcae748796d77b98039b5fc52ac962d877370b", "cal": {"scale": 0.07}, "SYNTHETIQUE": "r\u00e9sultat d'entra\u00eenement FACTICE", "ckpt_sha256": "4b7cbbd3458ffe94acbb78549ec87e3ea9f34170bafd289aa19e6ca4f50e7a51"}
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/aux/SYNTH_GO_CS1 ADDED
@@ -0,0 +1 @@
 
 
1
+ SYNTHETIQUE — drapeau factice de test
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/aux/SYNTH_recount64_cs1_mix_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "85e977700bbfc4ad571d4be15623450eaa4360fdfd1fe9451b0706f81bfc1453",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/aux/SYNTH_recount64_v1b_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "9ea562e44f708adda470c53144dcae748796d77b98039b5fc52ac962d877370b",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/aux/SYNTH_scan64_code_scale_v1.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "PASS": true,
4
+ "test_mode": false,
5
+ "stage_A": {
6
+ "code_delta": {
7
+ "per_window": {
8
+ "fc2": {
9
+ "covered_window_byte_positions": 0
10
+ },
11
+ "m7_fresh_en": {
12
+ "covered_window_byte_positions": 0
13
+ },
14
+ "m7_fresh_fr": {
15
+ "covered_window_byte_positions": 0
16
+ },
17
+ "m7_test20k": {
18
+ "covered_window_byte_positions": 0
19
+ },
20
+ "corpus_v3_test": {
21
+ "covered_window_byte_positions": 0
22
+ },
23
+ "m7_queries": {
24
+ "covered_window_byte_positions": 0
25
+ }
26
+ }
27
+ }
28
+ }
29
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/case.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "case": "b7_seen_perte_sans_gain",
3
+ "description": "aucun gain + seen cand à 0 %",
4
+ "expected_consigne": "REJETÉ",
5
+ "expected_predecl": "REJETÉ",
6
+ "SYNTHETIQUE": true,
7
+ "extra_args": [],
8
+ "checks": {},
9
+ "r3": true
10
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/runs_cand/results.jsonl ADDED
@@ -0,0 +1 @@
 
 
1
+ {"arm": "ctl", "snn_tokens": 600000, "am_tokens": 1200000, "epochs": 2, "ticks_per_epoch": [600000, 600000], "data_sha256": "85e977700bbfc4ad571d4be15623450eaa4360fdfd1fe9451b0706f81bfc1453", "cal": {"scale": 0.07}, "SYNTHETIQUE": "r\u00e9sultat d'entra\u00eenement FACTICE", "ckpt_sha256": "4b7cbbd3458ffe94acbb78549ec87e3ea9f34170bafd289aa19e6ca4f50e7a51"}
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_b7_seen_perte_sans_gain/runs_ref/results.jsonl ADDED
@@ -0,0 +1 @@
 
 
1
+ {"arm": "ctl", "snn_tokens": 600000, "am_tokens": 1200000, "epochs": 2, "ticks_per_epoch": [600000, 600000], "data_sha256": "9ea562e44f708adda470c53144dcae748796d77b98039b5fc52ac962d877370b", "cal": {"scale": 0.07}, "SYNTHETIQUE": "r\u00e9sultat d'entra\u00eenement FACTICE", "ckpt_sha256": "4b7cbbd3458ffe94acbb78549ec87e3ea9f34170bafd289aa19e6ca4f50e7a51"}
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_c1_aucun_gain/aux/SYNTH_GO_CS1 ADDED
@@ -0,0 +1 @@
 
 
1
+ SYNTHETIQUE — drapeau factice de test
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_c1_aucun_gain/aux/SYNTH_recount64_cs1_mix_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "85e977700bbfc4ad571d4be15623450eaa4360fdfd1fe9451b0706f81bfc1453",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_c1_aucun_gain/aux/SYNTH_recount64_v1b_sample.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "sha256": "9ea562e44f708adda470c53144dcae748796d77b98039b5fc52ac962d877370b",
4
+ "windows": {
5
+ "fc2": {
6
+ "confirmed_window_positions": 0,
7
+ "window_sha256": "b052096b0c1e2b301cd79921502f85e61912f209fc0027151bbf82237de8c2e3"
8
+ },
9
+ "m7_fresh_en": {
10
+ "confirmed_window_positions": 0,
11
+ "window_sha256": "67ff20b26ce5e3401cea57c817b1a8417e95f6cd131fd8fb631cd5cd8625331a"
12
+ },
13
+ "m7_fresh_fr": {
14
+ "confirmed_window_positions": 0,
15
+ "window_sha256": "875f0906a320e3f7b87e446aab646cf49c74786cb1a0db78b54fa4c62e872715"
16
+ },
17
+ "m7_test20k": {
18
+ "confirmed_window_positions": 0,
19
+ "window_sha256": "ce0526c62048f8e590a96cca5393a87c2e631f194eb4a2e974c584fd443b6a22"
20
+ },
21
+ "corpus_v3_test": {
22
+ "confirmed_window_positions": 0,
23
+ "window_sha256": "a6730bbe87555555555555555555555555555555555555555555555555555555"
24
+ },
25
+ "m7_queries": {
26
+ "confirmed_window_positions": 0,
27
+ "window_sha256": "5aab38d62b2f000df6a913d2574ccd314300e350ccc76e71e16768ed5c73b706"
28
+ }
29
+ }
30
+ }
results/hashing/code_scale_v1/drafts_r3_cs1/tests_synth_r3/cases/SYNTH_c1_aucun_gain/aux/SYNTH_scan64_code_scale_v1.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "SYNTHETIQUE": true,
3
+ "PASS": true,
4
+ "test_mode": false,
5
+ "stage_A": {
6
+ "code_delta": {
7
+ "per_window": {
8
+ "fc2": {
9
+ "covered_window_byte_positions": 0
10
+ },
11
+ "m7_fresh_en": {
12
+ "covered_window_byte_positions": 0
13
+ },
14
+ "m7_fresh_fr": {
15
+ "covered_window_byte_positions": 0
16
+ },
17
+ "m7_test20k": {
18
+ "covered_window_byte_positions": 0
19
+ },
20
+ "corpus_v3_test": {
21
+ "covered_window_byte_positions": 0
22
+ },
23
+ "m7_queries": {
24
+ "covered_window_byte_positions": 0
25
+ }
26
+ }
27
+ }
28
+ }
29
+ }