"""v307 -- the data watchdog: a decision pipeline over the frozen registry. v84 freezes the predictions and v100 censuses the grammar; this module is the operational DECISION layer that an adversarial reviewer's "experimental signature ledger" and "kill-test board" call for. It reads predictions_frozen.json and, for every prediction with a current best measurement, computes the live tension n_sigma and classifies it PASS / WATCH / TENSION / KILL against a fixed threshold, plus the conditional bands (r, n_s) and the watchdog fronts (w) -- so the theory's standing against today's data is one machine-checked table, refreshed on every suite run. [N] 1. COVERAGE: every frozen prediction + assigned texture + conditional band + F_transfer bridge readout (rare kaon) is classified (count matches the JSON). [N] 6. RARE-KAON BRIDGE: BR(K+)=9.45e-11 now lands ON the NA62 2016-2024 combination (9.6 +1.9/-1.8)e-11 (+0.08 sigma; was ~1.2 sigma vs the 2016-2022 13.0e-11) -- a strong consistency hit, but an F_transfer bridge (SM 8.6e-11 also inside the NA62 error), not a unique TFPT-vs-SM discriminator. [N] 2. THEORY ALIVE: NO decidable prediction is in KILL (|n_sigma| >= 3); the compiler survives current data. [N] 3. POWER: the pipeline is not trivially green -- at least one prediction sits at WATCH/TENSION (|n_sigma| >= 1.5), and the most-tensioned CORE prediction is sin^2 theta13 (~2 sigma), the honest pressure point the external analysis under-weighted (it is already documented in v62). [N] 4. JUNO #1 KILL TEST: the prediction-of-record sin^2 theta12 = 0.306747 is currently compatible (|n_sigma| < 1); JUNO will sharpen it (the fastest falsifier). [N] 5. INFLATION/DARK-ENERGY FRONTS: the R^2 band r in [0.0033,0.0048] is below the BK18 limit (0.036) and the kill threshold (0.01) -- allowed, not excluded; w=-1 is flagged WATCHDOG (DESI dynamical-DE hints = the most dangerous negative front). HONEST SCOPE: the measured central+sigma values are the repo-documented current bests (CODATA 2022, NuFIT 6.0, ACT DR6, Planck 2018, PDG 2024, BK18 -- the same sources as v62/v100), NOT freshly imported. Ratio observables with a known source-vs-pole / scheme caveat are judged against the conservative v18/v100 window, not a knife-edge sigma. A decision harness, not a new claim. Python-only. """ import json import math import os from tfpt_constants import check, summary, reset # ---- current best measurements (repo-documented; provenance in the comment) ---- # id -> (central, sigma, source). Clean experimental sigma -> n_sigma decision. MEAS = { "ALPHA_INV": (137.035999177, 0.000000021, "CODATA 2022"), "SIN2_THETA12_SEED": (0.307, 0.012, "NuFIT 6.0"), "SIN2_THETA13": (0.02195, 0.00058, "NuFIT 6.0"), "BETA_BIREFRINGENCE_DEG": (0.215, 0.074, "ACT DR6"), "OMEGA_B": (0.0493, 0.0006, "Planck 2018"), "LAMBDA_C": (0.2245, 0.0005, "PDG 2024"), "S23_CKM": (0.0411, 0.0013, "PDG 2024 |Vcb|"), "S13_CKM": (0.00382, 0.00024, "PDG 2024 |Vub|"), "DELTA_CKM_RAD": (1.1467, 0.05236, "PDG 2024 gamma=65.7+-3.0 deg"), } # ratio observables with a documented source-vs-pole / scheme caveat -> # judged against a conservative relative window (v18 5% / v100), not a knife-edge. SCHEME = { "MMU_OVER_MTAU": (0.05946, 0.03, "PDG 2024 (source vs pole)"), "ME_OVER_MMU": (0.004836, 0.03, "PDG 2024 (source vs pole)"), "MU_OVER_MD": (0.47, 0.05, "FLAG 2024 (scheme spread)"), "MC_OVER_MS": (13.6, 0.05, "PDG 2024 MSbar (scheme spread)"), "MT_OVER_MB": (41.0, 0.05, "PDG 2024 (scheme spread)"), } # F_transfer BRIDGE readouts: NOT frozen compiler predictions (external short-distance # input), tracked so the live experimental confrontation is visible, typed [C]. # id -> (prediction, measured_central, sigma, source) BRIDGE = { "BR_KP_PINUNU": (9.45e-11, 9.6e-11, 1.8e-11, "NA62 2016-2024 (La Thuile 2026)"), } BRIDGE_HIST = { # the earlier observation each bridge readout was previously judged against "BR_KP_PINUNU": (13.0e-11, 3.0e-11, "NA62 2016-2022"), } # no current decisive datum (downstream / assigned / not-yet-measured) NO_DECISION = {"MSCAL_OVER_MBAR", "DELTA_CP_NU_DEG", "SIN2_THETA23"} # the dark-energy / vacuum front: w=-1 readout, the most dangerous negative front WATCHDOG = {"RHOL_OVER_MBAR4"} # data provenance: every measured/bridge source carries a release date, so the watchdog # board cannot silently drift onto stale numbers (review point: provenance + date). SOURCE_DATES = { "CODATA 2022": "2022", "NuFIT 6.0": "2024", "ACT DR6": "2025", "Planck 2018": "2018", "PDG 2024": "2024", "PDG 2024 |Vcb|": "2024", "PDG 2024 |Vub|": "2024", "PDG 2024 gamma=65.7+-3.0 deg": "2024", "PDG 2024 (source vs pole)": "2024", "FLAG 2024 (scheme spread)": "2024", "PDG 2024 MSbar (scheme spread)": "2024", "PDG 2024 (scheme spread)": "2024", "NA62 2016-2024 (La Thuile 2026)": "2026", "NA62 2016-2022": "2025", } BK18_R_LIMIT = 0.036 # BICEP/Keck 2018 95% upper limit on r R_KILL = 0.01 # TFPT kill: robust r > 0.01 kills the R^2 branch def classify(n): a = round(abs(n), 1) if a < 1.0: return "PASS" if a < 2.0: return "WATCH" if a < 3.0: return "TENSION" return "KILL" def run(): reset() print("v307 data watchdog: decision pipeline over the frozen registry") here = os.path.dirname(os.path.abspath(__file__)) with open(os.path.join(here, "predictions_frozen.json")) as fh: reg = json.load(fh) preds = reg["predictions"] assigned = reg["assigned_texture_values"] bands = reg["conditional_bands"] board = [] # (id, layer, verdict, detail) core_clean = {} # id -> n_sigma for CORE clean-sigma predictions for p in preds: pid, val, layer = p["id"], float(p["frozen_value"]), p["layer"] if pid in MEAS: m, s, src = MEAS[pid] n = (val - m) / s verdict = classify(n) board.append((pid, layer, verdict, f"{n:+.2f} sigma vs {src}")) if layer == "core": core_clean[pid] = n elif pid in SCHEME: m, win, src = SCHEME[pid] dev = abs(val - m) / m verdict = "PASS" if dev <= win else "FLAG" board.append((pid, layer, verdict, f"{dev*100:.2f}% vs {win*100:.0f}% window ({src})")) elif pid in NO_DECISION or pid in WATCHDOG: tag = "WATCHDOG" if pid in WATCHDOG else "NO-DECISION" board.append((pid, layer, tag, "no decisive datum yet")) else: board.append((pid, layer, "NO-DECISION", "unmapped")) for a in assigned: board.append((a["id"], "assigned", "NO-DECISION", "assigned texture (octant/phase not predicted)")) # F_transfer bridge readouts (external SD input; not frozen compiler predictions) bridge_board = [] for bid, (pred, m, s, src) in BRIDGE.items(): n = (pred - m) / s bridge_board.append((bid, "bridge", classify(n), f"{n:+.2f} sigma vs {src}")) board.extend(bridge_board) # conditional bands rband = next(b for b in bands if b["id"] == "R_TENSOR") rmax = float(rband["band_max"]) r_allowed = rmax < BK18_R_LIMIT and rmax < R_KILL board.append(("R_TENSOR", "band", "PASS" if r_allowed else "TENSION", f"band<= {rmax:.4f} < BK18 {BK18_R_LIMIT} and kill {R_KILL}")) nsb = next(b for b in bands if b["id"] == "N_S") ns_lo, ns_hi = float(nsb["band_min"]), float(nsb["band_max"]) ns_c, ns_s = 0.9649, 0.0042 # Planck 2018 ns_ok = (ns_lo <= ns_c + 2 * ns_s) and (ns_hi >= ns_c - 2 * ns_s) board.append(("N_S", "band", "PASS" if ns_ok else "TENSION", f"band [{ns_lo:.4f},{ns_hi:.4f}] vs Planck {ns_c}+-{ns_s}")) # ---- print the decision board ---- print(" --- decision board ---") for pid, layer, verdict, detail in board: print(f" {verdict:<11} {pid:<24} [{layer}] {detail}") from collections import Counter tally = Counter(v for _i, _l, v, _d in board) print(" tally:", dict(tally)) # ---- 1. coverage ---- expected = len(preds) + len(assigned) + len(bands) + len(BRIDGE) check("COVERAGE [N]: every frozen prediction + assigned texture + band + " "F_transfer bridge is classified (%d entries match the registry)" % expected, len(board) == expected) # ---- 2. theory alive: no KILL among decidable predictions ---- decidable = [n for n in core_clean.values()] + \ [(float(p["frozen_value"]) - MEAS[p["id"]][0]) / MEAS[p["id"]][1] for p in preds if p["id"] in MEAS and p["layer"] != "core"] no_kill = all(abs(n) < 3.0 for n in decidable) check("THEORY ALIVE [N]: no decidable prediction is in KILL (|n_sigma|<3); " "the compiler survives current data", no_kill) # ---- 3. power + the documented pressure point ---- watch_items = [pid for pid, _l, v, _d in board if v in ("WATCH", "TENSION")] check("POWER [N]: the pipeline is not trivially green -- >=1 prediction at " "WATCH/TENSION (|n_sigma|>=1.5): %s" % ", ".join(watch_items), len(watch_items) >= 1) top_core = max(core_clean, key=lambda k: abs(core_clean[k])) check("PRESSURE POINT [N]: the most-tensioned CORE prediction is sin^2 " "theta13 (~2 sigma) -- the honest current pressure point (v62), which " "the external analysis under-weighted; |n|=%.2f" % abs(core_clean[top_core]), top_core == "SIN2_THETA13" and 1.9 <= round(abs(core_clean[top_core]), 1) < 3.0) # ---- 4. JUNO #1 kill test currently compatible ---- th12 = next(p for p in preds if p["id"] == "SIN2_THETA12_SEED") n12 = (float(th12["frozen_value"]) - MEAS["SIN2_THETA12_SEED"][0]) \ / MEAS["SIN2_THETA12_SEED"][1] check("JUNO #1 KILL TEST [N]: the prediction-of-record sin^2 theta12 = " "0.306747 is currently compatible (|n_sigma|=%.2f<1); JUNO is the " "fastest falsifier" % abs(n12), abs(n12) < 1.0) # ---- 5. inflation / dark-energy fronts ---- check("INFLATION FRONT [N]: the R^2 band r in [0.0033,0.0048] is below the " "BK18 limit (0.036) and the kill threshold (0.01) -- allowed, not " "excluded", r_allowed) check("DARK-ENERGY FRONT [N]: w=-1 (rho_Lambda readout) flagged WATCHDOG -- " "DESI dynamical-DE hints are the most dangerous negative front", any(v == "WATCHDOG" for _i, _l, v, _d in board)) # ---- 6. rare-kaon bridge: the NA62 2016-2024 update ---- pred_k, m_k, s_k, _src = BRIDGE["BR_KP_PINUNU"] n_k = (pred_k - m_k) / s_k hm, hs, _hsrc = BRIDGE_HIST["BR_KP_PINUNU"] n_k_hist = (pred_k - hm) / hs check("RARE-KAON BRIDGE [N]: BR(K+)=9.45e-11 now lands ON NA62 2016-2024 " "(9.6 +1.9/-1.8)e-11 -> %+.2f sigma (was %+.2f sigma vs the 2016-2022 " "13.0e-11) -- a strong consistency hit, but an F_transfer bridge (the SM " "8.6e-11 is also inside the NA62 error), not a unique TFPT-vs-SM " "discriminator" % (n_k, n_k_hist), classify(n_k) == "PASS" and abs(n_k_hist) > 1.0) # ---- 7. data provenance: every measured/bridge source is dated ---- used_sources = ([src for (_c, _s, src) in MEAS.values()] + [src for (_c, _w, src) in SCHEME.values()] + [src for (_p, _c, _s, src) in BRIDGE.values()] + [src for (_c, _s, src) in BRIDGE_HIST.values()]) undated = sorted({s for s in used_sources if s not in SOURCE_DATES}) check("DATA PROVENANCE [N]: every measured/bridge source on the board carries a " "release date (no silent drift onto stale numbers); %d sources, undated=%s" % (len(set(used_sources)), undated), undated == []) return summary("v307 data watchdog (decision pipeline)") if __name__ == "__main__": raise SystemExit(1 if run() else 0)