Identifying a Corporate Policy or Custom From Records and Repeated Events

Scope and warning#

Counting incidents is not enough: the events must be independently assessed for a common mechanism, defendant, policy period, underlying federal-rights violation, and causal connection. Notice is especially material to failure-to-train or inaction theories; it is not a free-standing element that must be proved from repeat incidents when the injury instead flowed from an express unlawful policy or authorized final-policymaker decision. Connick v. Thompson, 563 U.S. 51 (2011); Ninth Circuit Model Instructions 9.5–9.8.

The factual situation this roadmap addresses#

The following situation is invented for illustration. A researcher collects grievances, lawsuits and public reports about medication interruptions at one jail over four years.

Governing law#

• Express policy or unwritten custom: distinguish an adopted unlawful policy from sufficiently longstanding, widespread practice; an isolated failure to follow a lawful policy is generally insufficient by itself (Trevino, 99 F.3d at 918; Gordon v. County of Orange, 6 F.4th 961, 973–74 (9th Cir. 2021)). (Grade: Ninth Circuit authority; fact dependent.)
• Final policymaker/ratification: identify function-specific final policymaking authority and, for ratification, knowing and specific approval of the subordinate’s action and its basis (Hartzell, 130 F.4th 722; Ninth Circuit Model Instructions 9.6–9.7). (Grade: published Ninth Circuit and court instructions; authority question depends on law.)
• Failure to train/prevent: deliberate indifference ordinarily requires a pattern of similar constitutional violations, subject to a narrow obvious-consequences possibility, not a presumption from one bad outcome (Connick, 563 U.S. 51; Castro v. County of Los Angeles, 833 F.3d 1060 (9th Cir. 2016) (en banc)). (Grade: Supreme Court and Ninth Circuit; theory specific.)
• Causation: entity policy/action must be the moving force behind an underlying federal-rights deprivation; breach of a contract or operational target is not itself the constitutional tort (Ninth Circuit Model Instructions 9.5–9.8). (Grade: model instructions synthesizing precedent.)

Work backward from the mechanism of harm

The most reliable way to investigate policy/custom is to start with the patient event and work outward, rather than beginning with a disliked corporation and searching for adverse material. Ask:

1. What exactly failed at the patient level?
2. Was that failure a one-person decision, a handoff failure, or a system constraint?
3. If systemic, what rule or recurring mechanism would produce it?
4. Which entity owned that mechanism?
5. What evidence shows repetition or notice before the event?
6. How did the mechanism cause this particular injury?

This method reduces confirmation bias and helps separate genuine pattern evidence from unrelated criticism.

Build an evidence ladder

Policy/custom proof often develops in levels:

Level 1 — Patient record. The chart and chronology show the mechanism: for example, referrals repeatedly expire before scheduling.

Level 2 — Operational record. Logs, policies or workflow documents show how the process was designed.

Level 3 — Similar events. Other records show whether the same mechanism recurred.

Level 4 — Notice. Audits, grievances, reports or management communications show whether the organization knew of the pattern.

Level 5 — Organizational response. Corrective action, policy revisions, staffing decisions or inaction show what the entity did after receiving notice.

Level 6 — Causation. The patient's event is tied back to the same mechanism rather than to an unrelated individual error.

Not every case reaches every level. The ladder is a research method, not a requirement that every source must exist.

Use public records without turning allegations into findings

County contracts, Board of Supervisors agendas, procurement files, inspection reports, regulatory actions and court dockets can be powerful sources. Their evidentiary meaning differs.

• A contract can establish agreed responsibilities.
• A county agenda report can establish what staff represented to the Board.
• An inspection finding can establish what the inspecting body found under its standard.
• A complaint establishes allegations, not their truth.
• A settlement can resolve litigation without an admission.
• A judicial finding carries a different status from an allegation.

The Roadmaps should label the source category every time a public record is used to support a system proposition.

Similar incidents need a similarity rule

Before counting “similar” events, define the feature that makes them similar: same referral pathway, same medication-verification rule, same staffing model, same facility, same risk, same decision level, or same type of delayed response. If the definition changes after seeing the data, disclose that. Otherwise pattern analysis can become cherry-picking.

Also track time. A 2014 event may be poor proof of a 2026 practice if policies and management changed repeatedly in between. A later event may help explain persistence but ordinarily cannot establish pre-event notice by itself.

Negative evidence matters

If audits repeatedly found compliance, if similar high-risk cases were handled promptly, if staffing met contract levels during the relevant period, or if a disputed policy was never actually applied, record that evidence too. A credible investigation tries to falsify its theory.

“Nothing adverse found” is not necessarily proof of compliance—records can be incomplete—but contrary evidence belongs in the analysis rather than in a separate unpublished file.

Version control is essential

Policies, contracts, organizational charts and job descriptions change. For every system document record:

• title;
• exact entity name;
• effective date;
• revision date;
• superseded date if known;
• approving authority;
• facility scope;
• source location.

A later policy should not silently be used to describe an earlier period.

Elements of the claim#

For the express policy/unwritten custom route, test (1) a deprivation of an identified federal right by a state actor, (2) an adopted unlawful policy or a sufficiently widespread and settled practice attributable to the defendant entity, and (3) a direct causal link making it the moving force behind that deprivation. For the final policymaker route, identify who had final policymaking authority for the specific function and whether that official’s decision caused the violation; a single qualifying decision may suffice. Ratification requires specific approval by an actual final policymaker of both the subordinate action and its basis, not mere failure to discipline. A failure-to-train/prevent theory separately requires deficient procedures or training, requisite deliberate indifference to the risk, and causation. These are alternative theories rather than interchangeable labels (Ninth Circuit Model Civil Jury Instructions 9.5–9.8 (2026 compilation); Hartzell, 130 F.4th 722).

This research roadmap does not determine whether an underlying violation or any element is established. See 047 for entity-liability context.

Potentially relevant defendants#

Investigate each distinct legal entity (county, contracted healthcare organization, professional corporation, or other actor) against its own assigned function, policy authority, actual control, conduct, and relevant time period. A shared facility does not make county and contractor automatically responsible for one another’s practices. A private contractor performing a traditional public custodial-healthcare function may be evaluated under the § 1983 state-action and entity-policy framework, but neither contract status nor shared branding proves a particular defendant’s moving-force policy. Entity liability is not established by respondeat superior merely because an employee acted improperly (Tsao v. Desert Palace, Inc., 698 F.3d 1128 (9th Cir. 2012), as discussed in Ninth Circuit Model Instructions ch. 9).

Administrative prerequisites#

Research itself requires no grievance filing. In covered federal prison-conditions suits, the PLRA requires exhaustion of available administrative remedies by persons meeting the prisoner status test when suit is filed, and the availability and completion rules must be checked in the actual facility system (42 U.S.C. § 1997e(a)). Do not describe all civil claims or all formerly incarcerated plaintiffs as categorically subject to PLRA exhaustion. California public-entity claim presentation is a different prerequisite for qualifying state-law damages theories; it does not apply merely because a federal § 1983 defendant is a county.

Limitation periods#

Federal § 1983 civil-rights actions generally borrow California’s applicable personal-injury limitations period (Cal. Code Civ. Proc. § 335.1), while federal law governs accrual (Wallace v. Kato, 549 U.S. 384 (2007)). California imprisonment tolling (§ 352.1) is complex for pretrial and county-jail detention (Elliott v. City of Union City, 25 F.3d 800 (9th Cir. 1994); Austin v. Medicis, 21 Cal.App.5th 577 (2018)); do not assume custody tolls time. Separately, qualifying state-law death, personal-injury, and property-injury damages claims against a California public entity generally require timely six-month claim presentation, while certain other claims have a one-year period (Cal. Gov. Code § 911.2, subject to special rules and exceptions). This government-claim requirement does not govern a federal § 1983 cause of action. Medical-negligence and death/survival actions require independent analysis (Cal. Code Civ. Proc. §§ 340.5, 364, 377.30, 377.60). This roadmap does not calculate accrual, extensions, or tolling; Roadmap 008 addresses deadline screening.

Evidence#

Records to obtain#

Potential sources include county-held healthcare contracts, amendments, staffing exhibits, Board reports, and responsive performance records, requested under the California Public Records Act (Cal. Gov. Code § 7920.000 et seq.), subject to custody, applicable exemptions, redactions, and any withholding justification. Private-contractor-only files are not automatically CPRA records solely because of an agreement with a county. A person’s own medical records may be requested using Cal. Health & Saf. Code § 123110 and, where applicable, 45 C.F.R. § 164.524, subject to their exceptions. Grievances, authenticated court dockets, policies, versioned audits, and internal reviews may require different requests or counsel-controlled discovery. Do not collect other patients’ records without lawful authority. See Roadmap 006 and Roadmap 007.

Research matrix

Authentication and source integrity

Public webpages can change. Save the source URL, access date, title, issuing body and, where permitted, a stable PDF or archive copy. For litigation records, record court, case number, docket number and filing date. For spreadsheets, preserve the original export and document how any calculations were performed. Do not convert a filtered spreadsheet into “the data” without retaining the underlying file.

Discovery sequencing can save cost

Begin with the sources most likely to prove whether the proposed mechanism exists: policy versions, workflow descriptions, limited operational data and key witnesses. If those sources show no pattern, broader comparator discovery may not be justified. If they show a recurring bottleneck, targeted expansion becomes more defensible.

Build a proposition ledger

For each proposed policy/custom statement, maintain: proposition, source, source type, date, entity, facility, contrary evidence, and confidence. This prevents a later article or pleading from repeating a proposition whose source actually belongs to another facility or entity.

Distinguish notice from authority

A quality manager may receive every incident report yet lack power to change staffing. A corporate officer may have authority but never receive the local reports. For a failure-to-act theory, identify what the responsible policymaker knew or reasonably should have understood and what alternatives existed. For a final-policymaker theory, seniority alone is insufficient: final authority is function-specific and ultimately assessed from governing law and actual delegation (City of St. Louis v. Praprotnik, 485 U.S. 112 (1988); Ninth Circuit Model Instruction 9.6). Preserve organizational charts, written delegations, applicable contracts and escalation records, but do not equate receipt of a report with legal authority or deliberate indifference.

Use testimony to explain data, not replace it

A witness can explain how a queue or policy worked, but memory can be incomplete. Pair testimony with system records. Likewise, an unexplained spreadsheet should not be given a legal meaning until a knowledgeable witness or documentation explains its fields.

Public-source investigations should preserve correction mechanisms

If a public article identifies a pattern and later records contradict it, update the analysis and Corrections page. The Roadmaps' “what would change the analysis?” convention is designed to make correction possible without pretending early research was infallible.

Avoid causal overreach from corporate structure

Shared executives, office addresses, branding or management services can help identify where to investigate policy authority. They do not by themselves establish that a parent or manager issued the specific policy that caused the injury. The policy/custom ledger should identify the actual issuer and mechanism.

Discovery should test the theory, not merely collect volume#

Once counsel is involved, discovery can target the proposed mechanism: policy versions, training curricula, staffing data, incident logs, audits, communications, organizational authority and selected comparator events. An overbroad request for “all complaints ever” can produce expense and noise without proving similarity or notice.

Favorable and adverse facts#

Favorable facts

• Similar events, same facility, same period, documented in records.

Adverse facts

• Events spread across facilities, periods or entities.

Likely defenses#

Anticipate defenses that comparators concern different facilities, policies, clinical risks or time periods; a single injury cannot establish a widespread custom; the alleged institutional practice was lawful; supervisors lacked final policymaking authority; no qualified official specifically ratified the act and rationale; an audit was received only after the event; corrective measures changed the policy; or the decisive delay arose from an outside specialist, transport agency, independent clinician, refusal, or one employee’s departure from policy. Ask for the records that could verify or disprove each alternative. In Gordon v. County of Orange, 6 F.4th 961, 973–74 (9th Cir. 2021), an alleged single incident plus post-event procedural changes did not establish a county custom; that holding should be tested against the particular evidence rather than generalized into blanket entity immunity.

Causation#

A challenged entity action must be the moving force behind an underlying federal-rights deprivation, not merely a background contract deviation. Test cause in fact and proximate cause: whether the same injury would likely have occurred without the challenged practice and whether independent clinical, transport or third-party events explain it. The Ninth Circuit’s 2026 model instructions discuss both components with Tsao v. Desert Palace, Inc., 698 F.3d 1128, 1146 (9th Cir. 2012); this is a litigation element requiring fact-specific proof, not an inference from multiple bad outcomes.

Use source weighting

Not all pattern sources carry the same weight. A useful internal ranking is:

1. final judicial findings or undisputed official records;
2. contemporaneous operational data/audits;
3. sworn testimony/declarations;
4. regulatory findings within their scope;
5. grievances/complaints as notice evidence;
6. pleaded allegations not adjudicated;
7. press/social-media reports requiring independent verification.

The ranking is not a rule of evidence. It prevents an investigative database from treating an allegation and a finding as equivalent.

Control for organizational and policy change

If the contractor, facility leadership, EHR, referral vendor, staffing model or policy changed during the period, divide the data. A pattern crossing materially different systems can obscure rather than reveal custom. Conversely, persistence across a formal change can be significant if the same mechanism remains.

Ask who received the notice

An audit sent to county contract administrators may establish county notice while leaving contractor executive notice uncertain. A grievance routed to contractor quality staff may do the reverse. Map notice to the defendant whose deliberate-indifference theory relies on it.

Avoid the denominator trap

Whenever possible, pair adverse-event counts with exposure: number of patient encounters, referrals, medication administrations, grievances, occupied beds or relevant high-risk episodes. Denominators do not eliminate qualitative proof, but they prevent a large institution from looking worse merely because it handles more events.

Look for repeated workarounds

Staff workarounds can reveal that the written process does not function as designed. Examples include unofficial spreadsheets to track referrals, personal reminder lists, routine supervisor overrides or recurrent use of emergency transport because ordinary scheduling fails. Workarounds can also show employees successfully mitigating a problem, so their legal meaning depends on frequency, management awareness and outcome.

Preserve contrary cases and positive performance

A serious investigative record should include comparable instances where the system worked. If ten urgent referrals were timely and one failed, that context matters. If the same failure occurred repeatedly despite corrective plans, that context matters too. The goal is to understand the mechanism, not to build a one-sided archive.

Damages and remedies#

As for 047.

Discovery#

Potential discovery includes appropriately scoped incident databases, grievances, policies, decisionmaker delegations, and relevant operational measures. Do not assume all quality-assurance material is discoverable or all of it is privileged. California Evidence Code § 1157 addresses proceedings and records of specified qualifying peer-review/medical committees, with defined limits and exceptions; its reach to a specific jail contractor or document requires factual and statutory analysis. HIPAA, state medical confidentiality, protective orders, work-product protections, and other-patient privacy require independent treatment. Record whether a source was unavailable because it was not requested, was withheld, or was held legally protected; do not treat absence as negative evidence.

The CFMG/Wellpath dossier can generate hypotheses, not automatic proof

The existing documentary investigation contains contracts, public filings, litigation records and entity material that can point researchers toward policy questions. For a particular case, those materials should be treated as leads until tied to the event date, facility, entity and mechanism. A statewide article about another jail cannot by itself prove Fresno practice.

Cross-facility patterns require a common policy source

Events at several California facilities can be powerful if the same policy, platform, training or decision authority governed them. Without that link, cross-facility similarity may reflect common correctional problems rather than one defendant's custom. Identify the common source before aggregating.

Corporate statements are company positions

Public statements describing affiliates, services or compliance programs can help identify potential structure and attributed positions; they are not judicial findings. Compare the exact speaker, legal entity, effective period, and operational scope with signed contracts, corporate filings and operational records before labeling a statement as any particular entity’s policy.

County records can show the public side of the relationship

Board reports, amendments, invoices, performance documents and public-record responses can identify what the county understood the contractor to provide. They can also reveal gaps between formal contract language and operating reality. Label any inference from that comparison.

Publication itself requires correction discipline

If later records show that an article attributed a policy to the wrong entity or treated an allegation as a finding, the public Corrections process should identify and correct the proposition. Entity precision is not only a litigation concern; it is an editorial standard for this project.

CFMG/Wellpath-specific relevance#

Bankruptcy can alter a particular debtor’s litigation posture and recovery sources (11 U.S.C. §§ 362(a), 524(e), 1141(d)); nothing in this general evidence roadmap establishes that CFMG, a named Wellpath entity, or a particular employer is covered by a stay, discharge, injunction or confirmed plan. Test the precise debtor identity, docket, filing dates, plan, confirmation and injunction orders, and nondebtor provisions before stating any case-specific procedural effect. These are research leads, not findings about liability.

Primary authorities#

Related roadmaps#

Monell Liability Against Private Correctional Healthcare Contractors; Policy, Custom, Practice and Failure-to-Train Claims; Staffing Models and Whether Staffing Evidence Can Support a Claim.

Companion worksheet (print-ready)#

Print-ready companion worksheets are in development.

Worked records investigation

Hypothetical only: Suppose a patient experiences a six-week specialty-referral delay. The chart alone shows delay but not policy. A public contract shows the contractor is responsible for specialty scheduling. A policy describes an electronic referral queue. Queue data show several urgent referrals aging beyond the target. An audit predating the patient's event identifies the same backlog. A corrective plan is issued but remains incomplete. Testimony identifies which entity owned the queue. That chain can support investigation of custom/notice/causation far more effectively than collecting unrelated complaints about the contractor.

Now change one fact: the queue data show the plaintiff's referral was timely processed, but the outside specialist canceled twice. The proposed internal custom weakens. Good policy/custom research must be capable of reaching that conclusion.

Publish pattern claims only at the strength the evidence supports

“Records show repeated delayed referrals in the reviewed sample” is different from “the company maintained a statewide policy of denying specialty care.” Scope, sample, facility and entity must match the evidence. Overbroad public claims can undermine later legal credibility and create avoidable factual disputes.

Separate research completeness from legal sufficiency

A comprehensive records review can still yield insufficient evidence for Monell. Conversely, a smaller set of highly probative documents can establish a focused policy issue. The purpose of the research method is not to maximize the archive; it is to reach a supportable proposition with known limitations.

Final policy/custom dossier

A mature dossier should contain the proposition ledger, source index, policy versions, operational data, comparator-event table, notice chronology, authority/decisionmaker map, contrary evidence, and a one-page causal diagram. That package can then be evaluated by counsel without requiring the public Roadmap to decide liability.

Research protocol — Proving a repeatable practice rather than compiling allegations#

Define the proposed mechanism before counting examples

A claim that an institution maintains a recurring practice has to identify the behavior that repeats. Merely identifying several complaints about the same organization is not the same thing. The incidents might concern intake, withdrawal screening, urgent imaging, specialist referral, medication access or discharge transition, each governed by different personnel and rules. A trustworthy research protocol starts with a mechanism statement—for example, a recurring delay between a documented urgent request and the required clinical escalation—and then defines which source events satisfy that statement. If the definition is not fixed before examples are selected, the resulting count will tend to reflect what the investigator happened to find.

A useful incident table records the facility, date, clinical or operational issue, applicable policy version, source type, outcome, whether the asserted problem was verified, and why the incident was included or excluded. Multiple reports of one event should not be counted as multiple incidents. An individual's later lawsuit may include allegations derived from a preexisting government audit; these are two sources about the same underlying event, not independent confirmation of two separate events. The source register should identify dependency among records where possible.

Distinguish a comparable event from a superficially similar event

Two referral failures may share the word “delay” without sharing the same causal mechanism. One may concern the absence of a specialist appointment, another a clinical authorization denial, another a county transport cancellation, and another the patient's refusal. An entity-policy theory depends on who controlled the mechanism. Counting all four together can suggest a larger pattern while obscuring that no single relevant defendant controlled all four. The table should therefore classify decision authority and operational route separately from the observed outcome.

Where an investigator compares facilities, differences in custody classification, population acuity, staffing schedules, contracted services and local policies may matter. A national enterprise policy can sometimes support cross-site comparison, but its applicability to each facility should be confirmed. A local practice may be confined to one facility and period. Neither statewide branding nor a shared software platform proves that all hospitals, jails and professional corporations used an identical clinical protocol.

Document the denominator and the source universe

If the research identifies five delays, the number of similar situations in which timely care was provided may be just as important for interpreting the proposed practice. A denominator could consist of all urgent referrals during a defined quarter or all requests meeting a documented escalation threshold. An arbitrary count of web-visible lawsuits is rarely a reliable rate because filing behavior, public access and reporting differ by region and time. Where no denominator can be determined, the manuscript should state “five located complaints” rather than implying a proven incidence rate.

Source discovery also requires an exclusion log. Reasons may include duplicate filings, allegations later withdrawn, wrong defendant, unrelated legal time period, incomplete records, unverified identity, a different policy version or inability to establish a relevant event. Publishing the exclusion criteria may be more valuable than publishing an impressive raw count. It demonstrates that the researcher tested the apparent pattern rather than selectively assembled it.

Translate the apparent pattern into a legally bounded question

Even a real recurrent practice does not automatically establish a constitutional policy claim. The legal question also requires the underlying deprivation, entity attribution and causation. A series of negligent individual mistakes might lack the required institutional link. A training problem may be relevant but requires a showing that the deficiency, notice and causation satisfy the applicable standard. A high-ranking manager's awareness of general performance problems is not the same as an approved policy authorizing unconstitutional behavior. A court's procedural ruling on an allegation should never be described as a final finding that a company maintained a pattern.

A final research memo should therefore divide outcomes into “documented incidents,” “company or government position,” “judicially adjudicated facts,” “potential operational commonality,” and “questions requiring records not yet available.” The record may support only the first two categories. That can still justify public discussion of institutional accountability without alleging a violation that the sources do not prove.

Test the negative case

The strongest counterexample may be a facility with identical policy language but consistently timely escalation, or a period after a corrective action when the alleged problem stopped. Such evidence can weaken a theory that the written policy itself caused the failures and point instead to implementation, staffing, local workflow or individual circumstances. A new reporting system may produce an apparent increase in incidents because more events become visible, rather than because care deteriorated. A workforce expansion may change the absolute number of errors while reducing the rate. These alternatives should be considered before drawing a trend line.

For CFMG/Wellpath-specific public material, local county agreements, published audits and court opinions can help identify policy sources and date ranges. They cannot supply private patient histories or employer-confidential documents to a public Roadmap. The article investigation can document what the public contracts and filings say; the Roadmap explains how to test a policy theory in general. It should not reframe an unresolved corporate governance question as a proved constitutional wrong.

A falsification-ready research conclusion

A publishable conclusion must say what would change it. A newly located adopted policy might prove that an apparently local practice was actually enterprise-wide. A version archive might show that the cited policy was not in effect during the alleged events. A complete referral dataset might show the exceptions were far less frequent than the public litigation sample suggests. Signed board minutes might clarify who possessed final decision authority. The investigator should state these potential disconfirming records plainly and retain the uncertainty grade until they are examined. This makes the analysis more valuable than an accusation built around selected incidents.

Version changes and corrective measures can disprove a single continuous pattern

An investigator should mark policy changes, contract amendments, staffing changes and new oversight requirements directly on the incident timeline. A pattern apparently spanning five years may actually consist of several distinct eras. Evidence of correction may undermine a claim of uninterrupted practice, while repeated failures under different versions may suggest a more persistent operational mechanism; neither conclusion follows automatically. The strongest account reports the periods separately and describes precisely what was changed, who approved the change, and whether reliable measures of performance existed before and after implementation.

What this page cannot do#

It cannot say whether any custom existed.