Loading simulation checkpoint…
Constraint focus
━ Execution buffer━ All unfinished work
Detected exceptions
Recovery policy comparison
Same workload and fault schedule. Pending work stays visible.
Execution log
Latest 24-hour arrival cohort + all active workThese are simulated service actions. Review is an unresolved terminal exception in this demo; no real human decision is claimed.
Workflow stress replays
Examine the trade-off, not just the successful requests.
Total unfinished work
Includes upstream release queue━ Basic retry━ Controlled recovery
Paired outcomes
Identical arrivals and faultsPolicy settings
Fixed and versionedInjected conditions
Deterministic daily scheduleA smaller execution buffer alone is not an improvement. Compare total unfinished work, completed requests, unresolved exceptions and SLA together. These scenarios do not change the stored checkpoint.
RESEARCH / INCIDENT-INSPIRED EXPERIMENTS
Find the constraint. Test the intervention.
Documented context. Synthetic inputs. Calculated outcomes.
Loading offline research artifacts…
01 / DOCUMENTED EVENT
02 / SYNTHETIC ASSUMPTIONS
03 / CALCULATED RESULTS
Paired differences across 32 seeds. All policies receive identical requests for each seed. No result is a measured corporate outcome.
Work still waiting or executing
Fixed seed 0 · all three policies━ Basic━ Flow━ Flow + capacity
Outcomes and resource cost
Median [P10–P90] across seedsRanges describe simulated workload variability, not confidence intervals for the real company. SLA denominator: all unique arrivals. P95: completed requests only.
04 / DECISION LOGIC & LIMITATIONS
A&D / METHOD & EVIDENCE
Identify constraints. Test decisions. Improve flow.
Inspired by Altshuller & M. Goldratt
TOC / WHERE TO ACT
- Identify — locate the limiting service stage and its accumulating work.
- Exploit — protect available capacity from futile attempts.
- Subordinate — pace release; keep all held work visible.
- Elevate — test an extra worker and count its resource cost.
- Reassess — compare deadlines, queue time and completed work after the intervention.
TRIZ / HOW TO FRAME ALTERNATIVES
State the contradiction before selecting a solution: more recovery attempts can reduce reliability; faster release can worsen queues; more capacity requires resources.
Test separation in time, segmentation and temporary resource configurations. These are analyst-authored adaptations inspired by TRIZ, not an automated invention engine or complete implementation of its tools.
MATRIZ methodological reference ↗EXPERIMENT CONTRACT
3 cases × 3 demand levels × 32 paired seeds × 3 policies = 864 policy-runs. Six-hour arrival window, twelve-hour observation horizon. The numeric assumptions are synthetic; corporate events only motivate the failure mechanisms.
Basic: six attempts, two-minute retry delays, first-eligible ordering. Flow: six attempts, 5/10/20/40/60-minute delays, deadline priority, buffer of 12 and a ten-minute dispatch pause after three consecutive failures. Flow + capacity: the same controls plus one extra worker during the stated window. All policies deduplicate deliveries.
Reported queue time includes upstream holds and retry waits for all arrivals until completion, review or the horizon. Early failures can make this metric look artificially low. Failed and unfinished work remain in the SLA denominator. No numeric improvement is guaranteed.
These offline experiments use AD-CRISIS-1.0. The separate durable workflow retains AUDRATT-1.0 for compatibility with its saved state. Its three-attempt policy and live simulation assumptions differ from the six-attempt research experiment.
Independent portfolio research by Eddington Roman A. No affiliation, endorsement or client engagement with the named authors, organizations or companies is claimed. This is a simulation platform; it does not prevent cyberattacks or alter real operations.