Bound the object
One named workflow with a start, end, and decision someone may later have to defend.
Public synthetic · method walkthrough
How I approach a workflow evidence review: synthetic OOS investigation.
This page is not a client case study. It shows the method behind the Workflow Evidence Hardening Design Sprint on an invented quality-control scenario: can a later reviewer reconstruct the decision from retained records, or only from a clean summary?
How the engagement would run
Same shell as the paid design sprint
On a real workflow, free diagnostic first (short go / no-go only). If we proceed, the fixed-fee sprint produces the map, gap register, controls, and plan, not free workshops.
Trace one representative case
Follow one output backward through inputs, versions, human checks, and retained evidence.
Separate source from summary
Authoritative records must survive; derived narratives must link to them, not replace them.
Register gaps and failure modes
Name what a later reviewer cannot reconstruct, and what that blocks.
Design minimum records and gates
Specify the smallest defensible record and the review gates that protect it.
Prioritize hardening
Order fixes; leave production build and validation as separately scoped work.
Synthetic scenario
Sterile potency assay goes OOS, but the real failure may be document control.
A potency assay (HPLC) on a sterile compounded allergen extract batch reports below specification. The governing SOP is Rev 2.3. The twist: the analyst worked from a local Rev 2.2 printout; the acceptance criterion changed in 2.3. The investigation will only surface that if the record chain survives reconstruction.
The reconstruction question is not “did the report read well on the day of release?” It is: can I identify the authoritative source, the SOP version in effect, each investigation step, each corrective action, who approved each disposition, and the retained evidence for the final written conclusion?
Scenario card (invented)
Assessment disposition
Needs hardening
The investigation narrative can be written cleanly while still failing reconstruction: obsolete SOP printout, weak links from summary to raw data, and a conclusion that can outrun Phase I/II evidence if the retained package is only a PDF story.
Conditions before proceeding
- Bind the executed SOP revision and training status to the test date before disposition.
- Preserve first failing result, raw data, and LIMS identity separately from summary narrative.
- Require each written claim (lab error ruled out, no impact, CAPA closed) to point to named evidence.
- Treat retained evidence package completeness as a release gate, not an afterthought.
Evidence map
Authoritative record vs derived output at each step
This is the map step of the design sprint: for each stage, what must survive, what may be derived, and how the chain usually breaks.
StepAuthoritative recordDerived outputTypical failure
0 · Test executionLIMS row, instrument raw data, batch ref, effective SOP revisionWorksheet summary; pass/fail flagWork under obsolete printout; report cites current SOP
1 · OOS detectedFirst chromatogram/peak table; OOS flag timestamp; specification referenceAlert email or summary noticeFailing result overwritten before Phase I
2 · Deviation openedDeviation ID, open date, owner, linked lotSummary paragraph in QA packRetroactive deviation after pressure rises
3 · Phase INotebook, standards, calibration/suitability, repeats if performedPhase I narrative“Analyst error” with no attached checks
4 · Phase IIBatch records, sampling, materials, relevant monitoringRoot-cause write-upPlausible story without record support
5 · ImpactLinked lots, quarantine, distribution check“No impact” statementNarrow claim without genealogy
6 · CAPACAPA + document-control change order + training assignments“Distribution corrected” lineVerbal fix instead of controlled change
7 · ApprovalsNamed signatures / e-sign events in orderConclusion pageApproval before investigation complete
8 · Retained packagePointers to raw data, SOP history, deviation/CAPA, approval logPDF investigation reportOrphaned narrative with broken source links
Minimum record specification
Reconstruction checklist (R1-R8)
Whether the domain is pharma QC or an AI-assisted workflow, these are the same family of questions I use in a design sprint.
R1 · Authoritative source
Can I identify the OOS result source, or only a summary?
R2 · Version in effect
Can I recover the SOP revision that governed the test date?
R3 · Contemporaneous deviation
Is the deviation owned, dated, and linked to the lot?
R4 · Investigation path
Is Phase I before Phase II (when used), with sequence visible?
R5 · Claims ↔ evidence
Does each written claim point to named attachments?
R6 · Controlled CAPA
Is the fix a controlled change, not a verbal update?
R7 · Approval chain
Are approvers attributable and ordered before release?
R8 · Retained package
Can a later reviewer rebuild the decision without a re-narration?
If R8 fails, the workflow failed the reconstruction test even if the commercial batch decision “worked.”
Gap register (illustrative)
What the synthetic case is designed to surface
In a paid sprint these become versioned findings with owners. Here they show the analytical output of the method.
Executed method version is not bound to the test event.
If only the “current” SOP is cited after the fact, a later reviewer cannot prove which acceptance criterion governed the assay.
Consequence:Document-control failure can be misclassified as product failure, or the reverse.
Derived investigation PDF can outrun Phase I/II evidence.
A fluent narrative without attached raw data, calibration, and sequence records is not a reconstructable package.
Consequence:The conclusion may be right by luck and still indefensible.
Impact and CAPA claims lack mandatory evidence classes.
“No impact” and “distribution corrected” without genealogy, change control, and training completion records fail the minimum record test.
Consequence:Release and closure look complete while the retained package does not support them.
Hardening plan (design only)
What the design sprint would recommend next
These are control designs and verification criteria, not production implementation inside the fixed fee.
Bind effective SOP revision to the test event
Require document-control effective revision and training status to be recorded with the LIMS result before disposition proceeds.
- Verification
- An independent reviewer can answer “which SOP governed this test date?” from retained objects alone.
Separate source classes from summary narrative
Define mandatory attachments for OOS open, Phase I, Phase II, impact, and CAPA; block closure when classes are missing.
- Verification
- Test cases show that a PDF-only package cannot pass the release gate.
Make R8 a release criterion
Treat retained-package completeness (pointers to raw data, versions, deviation/CAPA, approvals) as a formal release check.
- Verification
- A cold reconstruction exercise succeeds without interviewing the original author.
Portability
Same reconstruction instinct outside pure lab QC.
AI-assisted and document-heavy workflows fail the same way: confident output, weak version binding, retrieval mistaken for support, and human approval without dispositioning exceptions. The design sprint applies this reconstruction test to one named workflow. OOS investigations are one instance of the shape.
Reliance boundary: this walkthrough demonstrates method structure only. It does not certify any organization, product, system, or intended use. Employer names, peer contacts, and real engagements are intentionally omitted.
Have one workflow with a similar reconstruction problem?
Send a short non-confidential description. Free outcome is proceed / not a fit / insufficient access, not a free map or hardening plan.