From d4a521072a25d8afdd62021f43538d7dad7586fd Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Wed, 22 Jul 2026 13:25:15 -0500 Subject: protocol: freeze hierarchical causal-capture funnel --- experiments/analyze_oral_a_v4_calibration.py | 153 +++++++++++++++++++++++++++ 1 file changed, 153 insertions(+) create mode 100644 experiments/analyze_oral_a_v4_calibration.py (limited to 'experiments/analyze_oral_a_v4_calibration.py') diff --git a/experiments/analyze_oral_a_v4_calibration.py b/experiments/analyze_oral_a_v4_calibration.py new file mode 100644 index 0000000..b24ebda --- /dev/null +++ b/experiments/analyze_oral_a_v4_calibration.py @@ -0,0 +1,153 @@ +#!/usr/bin/env python3 +"""Audit and gate hierarchical feedback-parameter causal capture.""" +import argparse +import glob +import json +import math +import os + + +SPLIT_HASH = "8328b206a97c420e49e54e3eca4abe3274c4756b084355784ea3fb8059e4515b" +RATES = (0.1, 1.0, 10.0) + + +def load(path): + with open(path) as handle: + record = json.load(handle) + args = record["args"] + mode = args.get("mode") + if mode not in ("hfa", "lhfa"): + raise ValueError(f"{path}: unexpected mode") + expected = { + "depth": 20, "width": 16, "seed": 0, "loader_seed": 0, + "epochs": 0, "train_limit": 10000, "val_examples": 5000, + "split_seed": 2027, "eval_split": "validation", "eval_every": 0, + "normalization": "batchnorm", "a_scale": 1.0, + "pert_directions": 1, "pert_every": 4, "pert_sigma": 0.01, + "perturb_seed": 1000, "alignment_probe": 64, + } + for key, value in expected.items(): + if args.get(key) != value: + raise ValueError(f"{path}: {key} drift") + expected_warmup = 0 if mode == "hfa" else 400 + if args.get("a_warmup_steps") != expected_warmup: + raise ValueError(f"{path}: feedback warmup drift") + if mode == "lhfa" and float(args["eta_A"]) not in RATES: + raise ValueError(f"{path}: unregistered feedback rate") + if record["provenance"]["git_tracked_dirty"]: + raise ValueError(f"tracked-dirty result: {path}") + if record["split"]["validation_index_sha256"] != SPLIT_HASH: + raise ValueError(f"split drift: {path}") + if record["evaluation_protocol"]["test_evaluations"]: + raise ValueError(f"test touched: {path}") + expected_space = (None if mode == "hfa" + else "hierarchical_feedback_parameters") + if record.get("calibration_metric_space") != expected_space: + raise ValueError(f"metric-space drift: {path}") + diagnostics = record.get("diagnostics") + if diagnostics is None: + raise ValueError(f"missing diagnostics: {path}") + values = diagnostics["teaching_negative_gradient_cosine"] + early_count = max(1, len(values) // 3) + norm_ratios = diagnostics["feedback_forward_norm_ratio"] + feedback_cosines = diagnostics["feedback_forward_cosine"] + metrics = { + "early_third_alignment": sum(values[:early_count]) / early_count, + "all_layer_alignment": sum(values) / len(values), + "mean_feedback_forward_cosine": ( + sum(feedback_cosines) / len(feedback_cosines)), + "min_feedback_forward_norm_ratio": min(norm_ratios), + "max_feedback_forward_norm_ratio": max(norm_ratios), + } + warmup = record.get("apical_warmup", {}).get("mean") + if mode == "lhfa": + if warmup is None: + raise ValueError(f"missing calibration aggregate: {path}") + metrics.update({ + "mean_calibration_mse": warmup["calibration_mse"], + "mean_target_power": warmup["target_power"], + "mean_prediction_target_cosine": ( + warmup["prediction_target_cosine"]), + "mean_parameter_update_rms": warmup["parameter_update_rms"], + }) + finite = (bool(record["final"]["finite"]) + and all(math.isfinite(value) for value in metrics.values())) + return { + "path": path, "mode": mode, + "eta_A": (None if mode == "hfa" else float(args["eta_A"])), + "metrics": metrics, "finite": finite, + "logical_batch_loss_queries": int( + record["work"]["logical_batch_loss_queries"]), + "total_macs": int(record["work"]["total_macs_estimate"]), + "source_commit": record["provenance"]["git_commit"], + } + + +def main(): + parser = argparse.ArgumentParser() + parser.add_argument("--input", default="results/oral_a_v4_calibration") + parser.add_argument( + "--out", default="results/oral_a_v4_calibration_gate.json") + args = parser.parse_args() + rows = [load(path) for path in sorted( + glob.glob(os.path.join(args.input, "*.json")))] + references = [row for row in rows if row["mode"] == "hfa"] + candidates = [row for row in rows if row["mode"] == "lhfa"] + if len(references) != 1 or len(candidates) != len(RATES): + raise ValueError("incomplete V4-1 method grid") + if {row["eta_A"] for row in candidates} != set(RATES): + raise ValueError("incomplete V4-1 rate grid") + if len({row["source_commit"] for row in rows}) != 1: + raise ValueError("V4-1 source commits differ") + eligible = [row for row in candidates if row["finite"]] + eligible.sort(key=lambda row: ( + -row["metrics"]["early_third_alignment"], + -row["metrics"]["all_layer_alignment"], row["eta_A"])) + selected = eligible[0] if eligible else None + reference = references[0] + checks = { + "all_four_records_finite": all(row["finite"] for row in rows), + "candidate_selected": selected is not None, + } + if selected is None: + checks.update({ + "early_third_at_least_0.05": False, + "all_layer_at_least_0.10": False, + "early_gain_over_fixed_hfa_at_least_0.04": False, + "feedback_norm_ratios_in_0.1_to_3": False, + }) + else: + metrics = selected["metrics"] + checks.update({ + "early_third_at_least_0.05": ( + metrics["early_third_alignment"] >= 0.05), + "all_layer_at_least_0.10": ( + metrics["all_layer_alignment"] >= 0.10), + "early_gain_over_fixed_hfa_at_least_0.04": ( + metrics["early_third_alignment"] + - reference["metrics"]["early_third_alignment"] >= 0.04), + "feedback_norm_ratios_in_0.1_to_3": ( + metrics["min_feedback_forward_norm_ratio"] >= 0.1 + and metrics["max_feedback_forward_norm_ratio"] <= 3.0), + }) + output = { + "protocol": "oral_a_v4_hierarchical_causal_capture_v1", + "status": "passed" if all(checks.values()) else "failed", + "checks": checks, "rows": rows, "matched_fixed_hfa": reference, + "selected_v4": selected, "confirmation_test_seeds_touched": False, + "review_score_before": 5, "review_score_after": 5, + "score_change_rule": "causal capture alone cannot raise score", + } + os.makedirs(os.path.dirname(os.path.abspath(args.out)), exist_ok=True) + with open(args.out, "w") as handle: + json.dump(output, handle, indent=2, sort_keys=True) + handle.write("\n") + print(json.dumps({ + "status": output["status"], "checks": checks, + "reference": reference, "selected_v4": selected, + }, indent=2)) + + +if __name__ == "__main__": + main() + -- cgit v1.2.3