Constraint Sensitivity Report¶
Document — instantiates Bottleneck Capacity Shadowing
Documents, from a fixed baseline, how the objective responds as each constraint or capacity is varied across a credible range — and what each level of relief would cost.
Constraint Sensitivity Report is the written artifact that answers a question a single price cannot: not "what is one more unit worth right here," but "how does the outcome move as I loosen this limit across a whole range, and how much does each step of loosening cost?" It starts from a fixed baseline — the current configuration and workload — and varies each candidate constraint or capacity over a plausible interval, recording the objective's response and where that response curve bends or flattens. Its defining trait is that it traces a response curve with a cost axis attached, delivered as a document a mixed audience can read, rather than a lone marginal number, a picture, or an after-the-fact measurement. Crucially, it must tie response to a relief decision; a report that merely varies parameters and stops is ordinary sensitivity analysis, not this mechanism.
Example¶
An SRE team owns a checkout service whose tail latency is drifting past its budget. Rather than argue over which resource to buy, they produce a sensitivity report. From the current deployment as baseline, they sweep four levers one at a time across a range: CPU cores, memory, the database connection-pool size, and the cache. For each level they record p99 latency and sustained throughput, and they attach the monthly cost of every increment. The document shows the shape, not just a point: enlarging the connection pool cuts p99 sharply up to about 1.5× the current pool, then flattens — the range where relief is real — while adding cores barely moves the tail at all (that resource is not binding). With cost sitting beside response, the pool expansion is visibly the best buy per dollar, and the report says so. An on-call lead and a finance partner can both read the same page and reach the same conclusion.
How it works¶
- Fix a baseline. Pin the current configuration, workload, and objective; every variation is measured as a departure from it.
- Sweep each lever over a range. Vary one constraint or capacity at a time (or across a small scenario grid) across a plausible interval, not just a nudge, so nonlinearity shows.
- Record the response and its bend. Tabulate objective versus level and mark where marginal value drops off — the range over which relief actually pays.
- Attach the cost of each level. Pair every relief increment with what it costs to achieve, so response is always read against price.
- Write it for a decision. Lay response and cost side by side and end on a relief recommendation, not a bare parameter table.
Tuning parameters¶
- Sweep range width — narrow ranges stay safely linear but hide where relief stops paying; wide ranges catch nonlinearity but risk invalid extrapolation.
- Factor design — one-factor-at-a-time is cheap and legible but blind to interactions; a factorial grid catches interactions at higher cost.
- Response source — analytic model, simulation, or historical replay; each trades fidelity against speed and credibility.
- Cost granularity — a lump figure per lever or an itemized relief cost; finer costing sharpens the comparison but takes effort.
- Refresh policy — a one-off study or a versioned document re-run as the workload shifts and the curves move.
When it helps, and when it misleads¶
Its strength is that it surfaces the nonlinearity a single shadow price hides — the point where a lever's marginal value collapses — and it puts cost next to response so a relief priority is legible without a separate valuation step. It travels well to non-technical decision-makers precisely because it is a document, not a solver dump.
Its failure modes are the familiar ones for a what-if study. One-factor-at-a-time sweeps miss interactions between constraints, so a lever that looks inert alone may matter jointly. Extrapolating past the swept interval invites false precision — the response curve is only trustworthy inside the range actually explored, which is exactly the caution that right-hand-side ranging formalizes.[n1] And a report that never connects its curves to a relief choice quietly decays into diagnostics. The discipline is to bound each range to where it was really tested, flag interactions rather than pretend they are absent, and always close with cost-versus-response and a recommendation.
How it implements the components¶
baseline_solution— the fixed reference configuration and workload that every variation is measured against; without it "how outcomes respond" has no anchor.marginal_response_range— the swept intervals and the point where each response curve bends define the range over which a constraint's marginal value credibly holds.relief_cost_model— the cost attached to each relief level, so the response is always weighed against what achieving it would cost.
It does not compute the exact dual price at the optimum or certify binding status from the model (shadow_value, binding_constraint) — that's Shadow Price Analysis, its nearest twin: that method reads one point, this report maps the whole range. It does not lay out the concrete relief options visually or rank them (relaxation_option, relief_priority) — that's Bottleneck Valuation Map; nor track what actually happens after relief (constraint_migration_monitor, downstream_effect_audit) — that's Before/After Constraint Monitoring.
Related¶
- Instantiates: Bottleneck Capacity Shadowing — the report supplies the range-and-cost picture that a relief priority is argued from.
- Consumes: Shadow Price Analysis supplies the local marginal value at the baseline that anchors the top of each response curve.
- Sibling mechanisms: Shadow Price Analysis · Bottleneck Valuation Map · Before/After Constraint Monitoring · Capacity Investment Analysis · Marginal Capacity Value Review
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Documents, from a fixed baseline, how the objective responds as each constraint or capacity is varied across a credible range — and what each level of relief would cost, making its operative form a computation, comparison, model, or analytic representation used to infer, estimate, or choose.
Independent corroboration: The frozen evidence defines Constraint Sensitivity Report as 'Documents, from a fixed baseline, how the objective responds as each constraint or capacity is varied across a credible range — and what each level of relief would cost', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Operations Research
Origin pattern: Single lineage
Present-day reach: Multi-domain
Rationale: Optimization and decision analysis cohered sensitivity reports showing how an objective changes as capacities or binding constraints are varied from a baseline.
Related originating lineages:
- Economics & Finance — Shadow-price analysis interprets the marginal value of relieving a binding constraint.
- Statistics & Experimental Design — Sensitivity analysis contributes systematic parameter variation and response curves.
Review resolution: Operations research clearly cohered sensitivity reporting, with shadow pricing and statistical parameter variation as genuine supporting lineages; the report form is established rather than synthesized here.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] Right-hand-side ranging (the "allowable increase / allowable decrease" in linear-programming sensitivity analysis) is the interval over which a constraint's shadow price stays valid. It is the formal statement of exactly what this report documents empirically: a marginal value holds only across a bounded range, past which the response changes. ↩