55-Decision & Change Log Template v1.0 | Chapter 6 Impact Assessment

Home ／ Docs-Technical WhitePaper ／ 55-Decision & Change Log Template v1.0

Chapter 6 Impact Assessment

I. Chapter Goals & Scope (Mandatory)

• Establish a standardized Object × Dimension impact framework that yields quantifiable, comparable, and replayable conclusions, aligned with state-machine gates and rollback triggers.
• Coverage: model/caliber/process/data-contract/implementation binding; excludes org/HR/budget approvals.

II. Assessment Objects & Boundaries (Mandatory)

1. Object layers:
   • Theory & caliber (S/P/M/I anchors);
   • Data & pipeline (data contracts, processing flows, replay suites);
   • Implementation binding (API/CLI/Schema, compatibility assertions);
   • Operations & release (SLO/SLA, rollback strategy).
2. Boundary: assess technical and operational impact only; business/legal to be recorded elsewhere.

III. Dimension System & Metrics (Mandatory)

1. Default dimensions & measures:
   • Performance: gate_accuracy, AUC, RMSE, etc.
   • Cost: unit compute/storage/personnel cost.
   • Time: gate_latency (batch/stream), time-to-release.
   • Risk: incident rate/severity (P×S), incident_level.
   • Dependency: number/coupling of affected subsystems.
   • Reproducibility: replay success, cross-env consistency.
   • Compatibility: prod replay pass rate, count of breaking API/Schema changes.
2. Gate naming: gate_<metric><comparator><threshold>@<window> (e.g., gate_accuracy>=0.99@7d).

IV. Normalization & Uncertainty (Mandatory)

• Normalize each raw metric to [0,1] to obtain s_i (monotone mapping for higher-is-better or lower-is-better).
• Uncertainty discount: if std. dev. is σ_i, use
  [
  s_i' = \mathrm{clamp}\bigl(s_i - k·σ_i,\ 0,\ 1\bigr),\quad k=1
  ]
• Weights: w_i≥0, ∑ w_i = 1; total score S = ∑ w_i · s_i'.
• Confidence reporting: provide CI_{95%} (or equivalent) and the statistical method (e.g., bootstrap).

V. Impact Matrix Template (copy-ready)

VI. Path/Formula Consistency (Mandatory)

1. If assessment involves arrival-time criteria, use the unified forms:
   • Constant factored: T_arr = ( 1 / c_ref ) * ( ∫ n_eff d ell )
   • General form: T_arr = ( ∫ ( n_eff / c_ref ) d ell )
2. Whenever T_arr appears, in the same or adjacent paragraph explicitly declare path gamma(ell) and measure d ell; all dimensional checks must pass check_dim.
3. No mixing: T_fil ≠ T_trans, n ≠ n_eff, c ≠ c_ref; no Chinese in formulas/symbols/definitions.

VII. Risk Grading & Triggers (Mandatory)

• Levels: L1 (minor) / L2 (moderate) / L3 (severe) / L4 (critical) based on impact surface and recovery cost.
• Example triggers: gate_accuracy<0.98@7d, gate_latency>2h@7d, incident_level≥2, data_drift>0.03@14d.
• Handling: upon trigger, enter the rollback branch; produce RollbackReport and re-deployment conditions (see Chapter 4).

VIII. Dependencies, Compatibility & Coverage (Mandatory)

• Dependency mapping: list impacted S/P/M/I anchors and system components; mark breaking changes (breaking=true/false).
• Compatibility statement: provide API/Schema version ranges and fallback paths; provide replay coverage (coverage@dataset) for replay suites.

IX. Machine-Readable Schema (YAML; JSON equivalent, copy-ready)

impact:

dimensions:

performance:

metric: "gate_accuracy"

target: ">=0.99@7d"

baseline: 0.97

after: 0.992

normalize: { mode: "higher_is_better", to: [0,1] }

sigma: 0.010

weight: 0.30

cost:

metric: "unit_cost"

target: "<=1.0x@base"

baseline: 1.00

after: 0.90

normalize: { mode: "lower_is_better", to: [0,1] }

sigma: 0.030

weight: 0.15

latency:

metric: "gate_latency"

target: "<=2h@7d"

baseline: "2.5h"

after: "1.8h"

normalize: { mode: "lower_is_better", to: [0,1] }

sigma: 0.050

weight: 0.10

risk:

metric: "incident_ps"

target: "P99@window"

baseline: "P98"

after: "P99.2"

normalize: { mode: "higher_is_better", to: [0,1] }

sigma: 0.040

weight: 0.15

dependency:

metric: "affected_subsystems"

target: "<=2"

baseline: 1

after: 2

normalize: { mode: "lower_is_better", to: [0,1] }

sigma: 0.000

weight: 0.10

reproducibility:

metric: "replay_success"

target: ">=0.98"

baseline: 0.95

after: 0.99

normalize: { mode: "higher_is_better", to: [0,1] }

sigma: 0.010

weight: 0.10

compatibility:

metric: "prod_replay_pass"

target: ">=0.99"

baseline: 0.98

after: 0.995

normalize: { mode: "higher_is_better", to: [0,1] }

sigma: 0.010

weight: 0.10

scoring:

k_sigma: 1.0

formula: "S = sum_i(w_i * clamp(s_i - k*sigma_i, 0, 1))"

dependencies:

anchors:

- "EFT.WP.Core.Equations v1.1:S20-1"

- "EFT.WP.Core.Metrology v1.0:check_dim"

- "EFT.WP.Core.DataSpec v1.0:I30-2"

systems:

- { name: "pipeline.arrival_time", breaking: false }

- { name: "api.metrics.v2", breaking: true }

compatibility:

api_schema:

version_range: "[2.0,3.0)"

fallback: "v1 adapter enabled"

replay:

coverage:

datasets: { "cmb_set_v3": "0.995", "lens_v1": "0.992" }

triggers:

- "gate_accuracy<0.98@7d"

- "gate_latency>2h@7d"

- "incident_level>=2"

- "data_drift>0.03@14d"

X. Human × Machine Alignment (Mandatory)

XI. Minimal Filled Example (copy-ready)

impact:

summary:

hard_gates_passed: true

S: 0.91

ci95: [0.88, 0.93]

key_findings:

- "Performance ≥0.99@7d; reproducibility and compatibility improved"

- "Cost ~10% below baseline; dependency surface +1 subsystem"

recommendation: "Proceed to Approved; keep API v1 adapter for 30 days as rollback buffer"

XII. Conclusion Expression (Mandatory)

• Hard gates first: if any hard gate fails, return to Review or enter rollback evaluation; only then proceed to scoring/ranking.
• Score-driven: report S and CI, provide ranking and a non-do list; record exception clauses and exit conditions (aligned with Chapter 3 fields).
• Rollback readiness: output monitoring metrics, thresholds, and restoration entry steps to ensure one-click rollback operability.