Decisions: how the package reads [WU26]

Generated by tools/equation_map.py from schema.DECISIONS and the [D9] / [E13] tags in the code; edit those, not this file. "paper" is the paper's own reading, "ours" a choice it leaves open. The plans hold the reasoning behind each.

id	rule	paper or ours	plan	implemented in	tested by
D1	the attack cost counts devices (a SCADA terminal and a PMU per bus) moved beyond noise	ours	docs/plans/WU_MSFDIA_PLAN.md, Decisions	engine.attacks.minimize._Window._devices	none
D2	the fewest-tamper search runs for every generated family, At and Am	ours	docs/plans/WU_MSFDIA_PLAN.md, Decisions	none	tests.test_minimize.test_new_generation_searches_by_default
D3	PGLib-OPF v23.07 rate_a as the alternative line ratings	ours	docs/plans/WU_MSFDIA_PLAN.md, Decisions	none	none
D4	PMUs read the current phasor of every branch at their bus (eqs. 19-20)	paper	docs/plans/WU_MSFDIA_PLAN.md, Decisions	none	none
D5	the overload goal (24)-(25) holds on the noiseless reading of the false state	paper	docs/plans/WU_MSFDIA_PLAN.md, Decisions	none	tests.test_overload.test_the_goal_rides_on_the_true_flow_and_ends_at_the_rating
D6	generation makes the multi-snapshot At and Am only	ours	docs/plans/WU_MSFDIA_PLAN.md, Decisions	none	none
D7	At's stealth bound and tamper count use the meters' rated accuracy	ours	docs/plans/WU_MSFDIA_PLAN.md, Decisions	engine.attacks.minimize.MinimizeMixin.min_tamper engine.attacks.minimize._Window._current_step engine.attacks.minimize._Window._currents engine.attacks.minimize._Window._sigmas engine.attacks.minimize._Window._snapshot models.config.RampSettings models.config.TimelineKnobs models.frames.FrameKnobs timeline.generate_timeline	tests.test_overload.test_am_counts_against_the_paper_noise_and_at_against_the_rated_accuracy tests.test_overload.test_am_has_no_stealth_bound_and_at_keeps_its_own
D8	Am's tamper threshold is the paper's case-study noise, 0.03 pu SCADA, 0.01 pu PMU	paper	docs/plans/WU_MSFDIA_PLAN.md, Decisions	engine.attacks.minimize engine.attacks.minimize._Window._currents engine.attacks.minimize._Window._sigmas engine.attacks.overload formulas.noise.paper_sigma	tests.test_attack_area.test_every_attack_the_search_returns_is_valid tests.test_overload.test_am_counts_against_the_paper_noise_and_at_against_the_rated_accuracy tests.test_overload.test_am_has_no_stealth_bound_and_at_keeps_its_own
D9	the goal adds the share k/T of the gap to the rating to each snapshot's true flow	ours	docs/plans/WU_MSFDIA_PLAN.md, Decisions	engine.attacks.overload._schedule models.frames.FlowGoal	tests.test_overload.test_the_goal_rides_on_the_true_flow_and_ends_at_the_rating
D10	hybrid meters: an angle only at a PMU, and PMU branch currents	paper	docs/plans/WU_MSFDIA_PLAN.md, Decisions	dataset.base engine.attacks.minimize._Window.__init__ engine.core.FdiaGenerator.__init__ generation.emit models.choices.MeterModel models.fields.PmuCurrentFields models.grid.MeterPlan timeline.generate_timeline	tests.test_models tests.test_pmu_channels tests.test_pmu_channels.test_a_scada_voltmeter_reads_no_angle_under_the_hybrid_model tests.test_schema.test_the_fixture_carries_exactly_the_schema
D11	Am has no between-snapshot bound: noise only sets the l0 threshold	paper	docs/plans/WU_MSFDIA_PLAN.md, Decisions	engine.attacks.episodes.OverloadDesigner.design engine.attacks.minimize._Window.__init__ engine.attacks.minimize._Window._sigmas generation.write.search_attrs models.config.TimelineKnobs models.frames.FrameKnobs timeline.generate_timeline	tests.test_overload.test_am_has_no_stealth_bound_and_at_keeps_its_own tests.test_overload.test_am_reaches_the_rating_on_the_held_support
D12	what the meters measure is its own knob of the meter plan	ours	docs/plans/WU_MSFDIA_PLAN.md, Decisions	none	tests.test_pmu_channels.test_new_generation_is_hybrid_only
D13	the previous frame's PMU currents are offered for the eq. (3) pseudo-measurements	ours	docs/plans/WU_MSFDIA_PLAN.md, Decisions	none	none
D14	generators of the support move freely within (22)-(23)	paper	docs/plans/WU_MSFDIA_PLAN.md, Decisions	engine.attacks.false_state.FalseStateMixin.free_injection_buses	tests.test_overload.test_a_generator_at_its_p_limit_keeps_its_q_free
D15	a line's rating defaults to 1.25 times its peak flow over the pool	ours	docs/plans/WU_MSFDIA_PLAN.md, Decisions	engine.attacks.overload engine.attacks.overload.OverloadMixin.use_line_ratings generation.write.search_attrs models.choices.RatingSource models.config.OverloadSettings	none
D16	the support's edge keeps (22)-(23), and a moved load changes by at most load_cap [YUA11]	ours	docs/plans/WU_MSFDIA_PLAN.md, Decisions	engine.attacks.false_state.FalseStateMixin._out_of_bounds engine.attacks.false_state.FalseStateMixin.solve_flow_local engine.attacks.minimize.MinimizeMixin._flow_goal_state models.config.OverloadSettings models.frames.FrameKnobs	tests.test_overload.test_an_edge_generator_stays_inside_its_limits tests.test_search_speed
D17	an overload episode drives two lines by default, as the paper's case studies	paper	docs/plans/WU_MSFDIA_PLAN.md, Decisions	engine.attacks.overload engine.attacks.overload.OverloadMixin._target_sets engine.attacks.overload.OverloadMixin.overload_goal generation.emit.overload_episode models.config.OverloadSettings models.frames.FlowGoal timeline._knobs	tests.test_overload.test_two_lines_by_default_and_the_pairs_share_one_area tests.test_search_speed.test_exhaustive_searches_return_their_pinned_answers tests.test_wu_scenarios
D18	the attacker's area: the paper's where stated, else Sec. III-A's rules (1-2 checked, 3-4 our score)	paper	docs/plans/WU_MSFDIA_PLAN.md, Decisions	engine.attacks.area.AreaMixin.rule_area	tests.test_attack_area.test_the_ieee118_area_is_fig_9_and_the_search_uses_it tests.test_attack_area.test_the_rules_check_1_and_2_and_charge_for_size
E1	trust is incremental: a trusted PMU keeps the offset its bus had before its slot (eqs. 29-30)	paper	docs/plans/WU_DEFENSE_PLAN.md, Decisions	trust.wu26	tests.test_trusted_pmus.test_trust_accumulates_from_its_slot
E2	a trusted PMU pins its own	V	and angle only (eqs. 27, 32)	paper	docs/plans/WU_DEFENSE_PLAN.md, Decisions
E3	the defense's cost is the devices tampered at any snapshot (Fig. 12); eq. (33)'s net l0 also reported	paper	docs/plans/WU_DEFENSE_PLAN.md, Decisions	models.choices.CostUnit models.config.WuDefenseConfig trust.defense trust.wu26	tests.test_wu_defense_env.test_the_reward_is_the_rise_in_the_search_cost
E4	1-minute attack snapshots interpolated from the 5-minute pool	ours	docs/plans/WU_DEFENSE_PLAN.md, Decisions	none	none
E5	IEEE-118: 10 snapshots, one trust slot each (Figs. 10-11)	paper	docs/plans/WU_DEFENSE_PLAN.md, Decisions	none	none
E6	Algorithm 1 line 8 ends an episode with no reward, as written	paper	docs/plans/WU_DEFENSE_PLAN.md, Decisions	none	tests.test_wu_defense_env.test_line_8_ends_the_episode_without_a_reward
E7	the DQN's state is Fig. 1's: the readings, the target lines' loading, the trusted set	paper	docs/plans/WU_DEFENSE_PLAN.md, Decisions	none	tests.test_wu_defense_env.test_the_state_is_what_fig_1_lists
E8	IEEE-118's 100 tests run as 10 training sessions of 10 test windows	ours	docs/plans/WU_DEFENSE_PLAN.md, Decisions	none	none
E9	the Q network is ours (two hidden layers of 128); its hyperparameters are the paper's	ours	docs/plans/WU_DEFENSE_PLAN.md, Decisions	trust.dqn.q_network	tests.test_trusted_pmus_dqn.test_the_stated_hyperparameters_are_the_defaults
E10	detection with PMU support is the residual test with trusted PMUs reading truth	ours	docs/plans/WU_DEFENSE_PLAN.md, Decisions	none	none
E11	the single-snapshot trusted-meter classes stay as the linear analogue	ours	docs/plans/WU_DEFENSE_PLAN.md, Decisions	none	none
E12	IEEE-1354 is not reproduced	ours	docs/plans/WU_DEFENSE_PLAN.md, Decisions	none	none
E13	one support is held for the window (a support per trust slot never changed an answer, removed)	ours	docs/plans/WU_DEFENSE_PLAN.md, Decisions	none	none
E14	reproduction ratings: each target's last flow plus 0.10 pu (Fig. 4's scale); k times the peak as a sensitivity	ours	docs/plans/WU_DEFENSE_PLAN.md, Decisions	none	tests.test_attack_area.test_delta_ratings_generate_am_episodes
E15	generation keeps the linear ramp of D9	ours	docs/plans/WU_DEFENSE_PLAN.md, Decisions	none	none
E16	the attack minimizes the l1 of the window's summed attack, eq. (12) read literally ([37]'s l1)	paper	docs/plans/WU_DEFENSE_PLAN.md, Decisions	trust.wu26	none
E17	the reproduction meters every branch at both ends, each node's SCADA its incident flows ([29])	paper	docs/plans/WU_DEFENSE_PLAN.md, Decisions	trust.wu26	none
E19	the faithful attack's goal is (24)-(25) as written: the end at S_max, the flow never falling; D9's ramp stays for generation	paper	docs/plans/WU_DEFENSE_PLAN.md, Decisions	trust.wu26 trust.wu26._WindowProblem.constraints	tests.test_wu26_faithful.test_the_goal_is_eqs_24_25_as_written
E18	the reproduction's overload rho, per-snapshot l1 weight tau and trust region are ours	ours	docs/plans/WU_DEFENSE_PLAN.md, Decisions	trust.wu26	none