AI in Biopharma Manufacturing: Are We There Yet?

Fact check — Change: What Happens When the Model — or the Process — Changes?

Claims were extracted from the episode script and verified against the primary documents by AI agents that saw only the claim and the document. This report is itself AI output; it can be wrong. Corrections: jack@jackprior.ai.

Fact check — Change: What Happens When the Model — or the Process — Changes?

This is the fact check of the episode as published. Each factual claim was extracted from the script and verified by an AI agent that saw only the claim and the primary document (the PDFs in the corpus). Verdicts: SUPPORTED (the document says it), PARTIAL (supported with a difference, noted), NOT-IN-CORPUS (the source is not among the primary documents on disk), NOT-CHECKABLE (an estimate or a characterisation). Opinions voiced by the hosts are listed but not verified. This report is itself AI output and can be wrong; corrections: jack@jackprior.ai.

Episode as published	Claims
Claims checked	68
Supported by the primary text	63
Confirmed by official web sources	2
Resting on secondary sources or hedged as such	1
Corrected before release	2
Open	0
Host opinions (not verified)	27

Claims resting on secondary sources

These statements stand, but the primary documents on disk do not themselves confirm them: they come from the landscape reference (a working index whose rows are checked against the web), are estimates, or are hedged in the episode as such.

Line	Speaker	Claim	Verdict	Why it stands	Evidence
8	SARAH	FDA published ICH Q12 as guidance in May 2021.	NOT-IN-CORPUS	Landscape/general knowledge; the ICH text confirms Step 4 in 2019	FDA's own publication of Q12 is not in the corpus
34	SARAH	ICH Q12's decision tree: if a parameter must be controlled to ensure quality it is an EC and the risk if it changes sets the category (high prior approval; moderate or low notification); if not, not an EC and not reported.	PARTIAL	The two-step logic is in 3.2.3.1; Figure 1's box labels could not be read from the text	3.2.3.1 text: parameters that need to be controlled 'should be considered ECs'; risk of change 'informs the reporting category … from high to low risk'; Figure 1 is an image not present in the text extract
90	HOST	An industry proposal is circulating that asks for a repeatable AI PACMP extension.	NOT-CHECKABLE	Excluded source; no content spoken	Excluded source; the script deliberately says nothing of its content

All claims, by document

BioPhorum-2026-06-AI-Risk-Guidance-Harmonizing-Frameworks.pdf

Line	Speaker	Claim	Verdict	Evidence
14	SARAH	BioPhorum's guidance says a change to the level of autonomy or the human-oversight configuration is a model change.	SUPPORTED	8.7 (pp.25-26): change triggers include 'changes to decision logic, levels of autonomy or human oversight configurations'

EMA-2024-09-AI-Medicinal-Product-Lifecycle-Reflection-Paper.pdf

Line	Speaker	Claim	Verdict	Evidence
16	SARAH	EMA's reflection paper will not accept incremental learning in a pivotal trial.	SUPPORTED	2.3.3.3 Pivotal clinical trials: 'Incremental learning approaches are not accepted'

EU-2015-10-Annex-15-Qualification-Validation.pdf

Line	Speaker	Claim	Verdict	Evidence
20	SARAH	EU Annex 15 carries the same change-management expectation into GMP.	SUPPORTED	Annex 15 section 11 Change Control (11.1–11.7)

EU-2025-07-Annex-11-Computerised-Systems-Draft.pdf

Line	Speaker	Claim	Verdict	Evidence
18	SARAH	No document treats a third-party model or platform change as a category; the nearest hook is the Annex 11 revision's periodic review of vendor contracts and SLAs.	SUPPORTED	14.2 vii 'support contracts and service level agreements (SLA)'; viii 'Contracts and key performance indicators (KPI) with vendors and service providers'
24	SARAH	Annex 11 revision's quality system clause: any change to a computerised system, including configuration, components, platform and operating system, is made in a controlled manner; any significant change that may impact quality, safety or data integrity is subject to re-qualification and validation.	SUPPORTED	3.1 ii, near-verbatim
26	SARAH	Annex 11 chapter 14 periodic review asks whether the system remains fit for intended use and in a validated state or needs re-validation in whole or in part; scope includes changes to components, configuration, platform and infrastructure, 'the combined effect of multiple changes in this, and in other systems', undocumented changes found by configuration auditing, vendor contracts and performance indicators, and changes to regulatory requirements.	SUPPORTED	14.1 and 14.2 (i, iii, vii, viii, xii); 'the combined effect of multiple changes in this, and in other systems' verbatim (14.2 iii)
68	SARAH	Annex 11 revision principle 2.6: when using outsourced activities the regulated user remains fully responsible for adherence to requirements and for the evidence.	SUPPORTED	2.6, verbatim substance
68	SARAH	Annex 11 periodic review includes support contracts, service level agreements and vendor performance indicators.	SUPPORTED	14.2 vii, viii
73	SARAH	The Annex 11 revision states as a principle that the regulated user remains fully responsible for outsourced activities.	SUPPORTED	2.6 Outsourced activities: 'When using outsourced activities, the regulated user remains fully responsible for adherence to the requirements included in this document' (p. 2)
73	SARAH	The Annex 11 revision's periodic review covers contracts with vendors and service providers.	SUPPORTED	14.2 Scope of review, viii: 'Contracts and key performance indicators (KPI) with vendors and service providers' (p. 11)
73	SARAH	The Annex 11 revision has a chapter on supplier and service management.	SUPPORTED	Chapter 7 heading 'Supplier and Service Management' (p. 5; also in the document map, p. 1)
73	SARAH	The Annex 11 revision states the principle that the regulated user stays fully responsible for outsourced activities.	SUPPORTED	2.6 Outsourced activities (p.2): 'When using outsourced activities, the regulated user remains fully responsible for adherence to the requirements included in this document, for maintaining the evidence for it, and for providing it for regulatory review.'
73	SARAH	The Annex 11 revision includes periodic review of vendor contracts.	SUPPORTED	14.2 Scope of review (p.12): 'periodic reviews should include, but may not be limited to: ... viii. Contracts and key performance indicators (KPI) with vendors and service providers.'
73	SARAH	The Annex 11 revision has a chapter on supplier and service management.	SUPPORTED	Section 7 'Supplier and Service Management' (p.5), clauses 7.1-7.5: Responsibility, Audit, Oversight, Documentation availability, Contracts
84	SARAH	Annex 11 14.1 covers re-validation in parts; Annex 22 10.5 human-review records.	SUPPORTED	14.1 're-validation (complete or in parts)'; Annex 22 10.5 human review records

EU-2025-07-Annex-22-AI-Draft.pdf

Line	Speaker	Claim	Verdict	Evidence
12	SARAH	Annex 22 10.1 names any change to the model, the system, or the process, 'including any change to physical objects the model is using as input'.	SUPPORTED	10.1, verbatim
16	SARAH	Annex 22 calls self-changing models dynamic and says they should not be used in critical GMP applications.	SUPPORTED	Scope lines 12-15: dynamic models 'not covered by this document, and should not be used in critical GMP applications'
28	SARAH	Annex 22 10.1: a tested model, its system and the process it automates or assists under change control before deployment; any change to any of them including physical inputs documented and evaluated for retest; 'any decision not to conduct such retest should be fully justified'. 10.2 configuration control with measures to detect unauthorised change. 10.3 performance monitored against metrics, example a change of lighting. 10.4 input sample space monitoring with drift metrics.	SUPPORTED	10.1–10.4, verbatim phrases
68	SARAH	Annex 22 2.2: documentation available to and reviewed by the regulated user whether the model is trained, validated and tested in-house or by a supplier.	SUPPORTED	2.2, verbatim
73	SARAH	Draft Annex 22 does not describe tiering supplier or vendor change notifications by potential impact so that assessment effort goes where the risk is.	SUPPORTED	Absence confirmed: the only supplier reference is 2.2 (documentation review 'whether the model is trained, validated and tested in-house or by a supplier'); the only change provision is 10.1 Change control, generic, with no notification or tiering
73	SARAH	The AI documents lack the habit of tiering supplier change notifications by potential impact.	SUPPORTED	Annex 22 10.1 Change control (p.5): 'Any change to the model itself, the system, or the process in which it is used ... should be documented and evaluated to determine if the model needs to be retested.' 'Supplier' occurs once (2.2); 'vendor', 'notification' and 'tier' do not occur
74	SARAH	Annex 22 requires a justification whenever a retest is not done after any change; Q12 requires a submission only when an EC changes; FDA's draft a report only when the change impacts model performance.	SUPPORTED	10.1: 'Any decision not to conduct such retest should be fully justified'

FDA-2021-05-ICH-Q12-Guidance.pdf

Line	Speaker	Claim	Verdict	Evidence
8	SARAH	FDA published ICH Q12 as guidance in May 2021.	NOT-IN-CORPUS	FDA's own publication of Q12 is not in the corpus

FDA-2023-03-AI-in-Drug-Manufacturing-Discussion.pdf

Line	Speaker	Claim	Verdict	Evidence
6	SARAH	CDER's discussion paper on AI in drug manufacturing is from 2023.	SUPPORTED	Cover: 'CENTER FOR DRUG EVALUATION AND RESEARCH … Discussion Paper \\| 2023'
60	SARAH	FDA's March 2023 discussion paper, from CDER's quality office, has an area five on continuously learning AI systems.	SUPPORTED	p.10: '5. Continuously learning AI systems that adapt to real-time data'; OPQ named on the back page
60	SARAH	The 2023 paper says models in manufacturing are updated through the change control process within the PQS; 'it may be challenging to determine when an AI model can be considered an established condition of a process'; 'it may also be challenging to determine the criteria for regulatory notification of changes to the model'; it asks how FDA will examine continuously updated AI control models during an inspection, and about product comparability after model-introduced changes especially for biologics.	SUPPORTED	Area 5, p.10, both quotes verbatim; inspection and comparability points as 'Applicants may need clarity'

FDA-2024-12-PCCP-AI-Device-Software.pdf

Line	Speaker	Claim	Verdict	Evidence
16	SARAH	The PCCP guidance calls automatic modification continuous learning and writes a plan for it.	SUPPORTED	Scope p.5: 'implemented automatically by software, also known as continuous learning'; VI.B p.21
49	SARAH	The PCCP guidance is titled Marketing Submission Recommendations for a Predetermined Change Control Plan for AI-Enabled Device Software Functions; from CDRH with CBER and CDER on the cover; originally issued 4 December 2024, reissued 18 August 2025; based on a statutory section Congress added in December 2022.	SUPPORTED	Cover (also names the Office of Combination Products; full title 'Artificial Intelligence-Enabled'); p.4: FDORA enacted 29 December 2022 added section 515C
49	SARAH	PCCP scope: device software the manufacturer intends to modify over time, automatic (continuous learning), manual, or both; reaches the device constituent of a combination product and stops there.	SUPPORTED	Section III Scope pp.5-6
51	SARAH	PCCP Description of Modifications: specific planned modifications with performance specifications after each; automatic or manual; for automatic ones 'boundaries or guardrails that define the range of automatic updates'; expected frequency from annual to continuous; global or local.	SUPPORTED	VI.B p.21, verbatim; frequency 'annually … to continuously'; VI.A; global/local p.21
51	SARAH	PCCP Modification Protocol has four parts: data management; re-training practices including triggers such as new data reaching a certain size or observed drift; performance evaluation against both the original and the last version; update procedures including user communication and real-world monitoring. Impact Assessment: benefits and risks of each modification, how one affects another, cumulative impact.	SUPPORTED	VII.A p.24 (four components); p.27 triggers and comparison to original and last version; VIII p.30 impact assessment items
53	SARAH	An authorised PCCP is 'a technological characteristic of the authorized device'; modifications specified and implemented per protocol do not trigger a new marketing submission; an unresolvable performance-evaluation failure is recorded and the modification not implemented; a deviation from the PCCP (data management or re-training failure, failure to meet criteria) would generally make the device adulterated and misbranded; every modification goes through the quality system.	SUPPORTED	IV.C p.9; p.10-12 (no new submission; deviations 'adulterated and misbranded'; quality system); p.27 unresolvable failure
53	SARAH	PCCP's eligible modifications include changes to device inputs and compatibility, such as different makes of acquisition system, updated operating systems, or 'updated cloud infrastructure'.	SUPPORTED	VI.C p.22, verbatim
74	SARAH	The PCCP guidance treats changes to device inputs and compatibility as modifications for the plan.	SUPPORTED	VI.C p.22 type (ii)

FDA-2025-01-AI-Enabled-Device-Software-Lifecycle-Draft.pdf

Line	Speaker	Claim	Verdict	Evidence
73	SARAH	FDA's January 2025 AI-enabled device software lifecycle draft does not describe tiering supplier or vendor change notifications by potential impact so that assessment effort goes where the risk is.	SUPPORTED	Absence confirmed: no occurrence of supplier, vendor, third-party, supply chain or outsourcing; impact-assessment and change-control hits concern device performance testing, PCCPs and user-task impact

FDA-2025-01-AI-Regulatory-Decision-Making-Draft.pdf

Line	Speaker	Claim	Verdict	Evidence
12	SARAH	FDA's draft names 'changes in manufacturing that may impact the performance of the AI model'.	SUPPORTED	IV.B lines 539-542, verbatim
14	SARAH	FDA's draft calls deliberate model changes 'new signals requiring manual changes in the model'; the PCCP guidance calls them manual modifications.	SUPPORTED	IV.B lines 542-544, verbatim; 'deliberate' is the script's gloss on IV.B line 514 'incidentally or deliberately'
16	SARAH	FDA's draft says AI models can be self-evolving, capable of autonomously adapting without any human intervention, and tells sponsors to anticipate inherent, model-directed changes.	SUPPORTED	IV.B lines 528-535, verbatim
43	SARAH	FDA IV.B lifecycle maintenance is aimed explicitly at pharmaceutical manufacturing; defines it as management of changes to AI models 'whether incidentally or deliberately, to ensure the model remains fit for use over the drug product life cycle for its context of use'; changes evaluated by the manufacturer's change management system within the PQS with three examples: newly available manufacturing data, new signals requiring manual changes, model-directed changes.	SUPPORTED	IV.B lines 513-515 ('for its COU'), 539-544; footnote 32 frames manufacturing 'as an example'
45	SARAH	FDA: the impact of a change is determined by model risk and by the change in model performance; credibility steps may need re-execution including retraining and retesting; if the change impacts performance it should be reported in accordance with regulatory requirements; a footnote ties the notification mechanism to impact on performance and product quality; detailed plans (metrics, monitoring frequency, retest triggers) in the site's PQS with a summary in the application at a level commensurate with model risk.	SUPPORTED	IV.B lines 544-557; footnote 35
47	SARAH	FDA: sponsors may use Q12 tools, established conditions and comparability protocols 'referred to as postapproval change management plans'; may propose 'model-related elements to be considered established conditions, along with a plan to manage changes to these established conditions'; by including such plans sponsors may prospectively obtain Agency input including which changes would not require submission prior to modification.	SUPPORTED	IV.B lines 559-567, verbatim
73	SARAH	FDA's January 2025 draft on AI for regulatory decision-making does not describe tiering supplier or vendor change notifications by potential impact so that assessment effort goes where the risk is.	SUPPORTED	Absence confirmed: 'supplier', 'vendor', 'third-party' do not occur; the only risk tiering is the sponsor's own credibility assessment ('The level of oversight for a model over its life cycle should be risk-based', l. 945)
73	SARAH	The AI documents lack the habit of tiering change notifications by potential impact (FDA draft).	PARTIAL	Footnote 35 (p.17): 'The mechanism for postapproval notification of changes to models can be determined on the basis of the following two factors: (1) impact of the change on model's performance and (2) impact of the change on product quality.' 'Vendor', 'supplier' and 'tier' do not appear

FDA-2026-02-CSA-Production-Quality-System-Software.pdf

Line	Speaker	Claim	Verdict	Evidence
66	SARAH	FDA's CSA guidance says that for devices under a premarket approval, changes to the manufacturing method, production software included, that do not affect safety or effectiveness go in an annual report and those that do in a thirty-day notice, with a manufacturing execution system as example.	SUPPORTED	V.A(3) pp.11-12: for PMA/HDE devices, 'changes to the manufacturing procedure or method of manufacturing that do not affect the safety or effectiveness … reported to FDA in a periodic report (usually referred to as an annual report)'; MES example; 30-day notice otherwise
68	SARAH	The CSA guidance lists what to review about a vendor.	SUPPORTED	V.A(5) pp.14-15: onsite audits, accreditations and certifications, development/QA/cybersecurity practices, data integrity capabilities

ICH-2011-Q8Q9Q10-Points-to-Consider.pdf

Line	Speaker	Claim	Verdict	Evidence
65	HOST	The 2011 Points to Consider suggested parallel testing against the reference method through the lifecycle for high-impact models.	SUPPORTED	PtC 5.3 lines 385-387: 'It can be beneficial to verify … by parallel testing with the reference method … and can be repeated throughout the lifecycle, as appropriate'

ICH-2025-10-Advanced-Manufacturing-Reflection-Paper.pdf

Line	Speaker	Claim	Verdict	Evidence
6	SARAH	ICH's reflection paper on advanced manufacturing is dated October 2025.	SUPPORTED	Running header: 'Endorsed by the ICH Assembly on 8 October 2025'
62	SARAH	ICH's reflection paper (endorsed 8 October 2025) on process models: need for global harmonisation on terminology, model risk framework, basis for oversight and data requirements; 'manufacturers require guidance on regulatory notification of model updates considering model risk and maturity of a site's quality system'; the Points to Consider does not address linking model risk to validation and lifecycle management and did not explicitly foresee these new types of AI models; a new ICH guideline on process models could provide a comprehensive framework.	SUPPORTED	Section B, pp.3-4, verbatim

ICH-Q10.pdf

Line	Speaker	Claim	Verdict	Evidence
20	SARAH	ICH Q10 has been final since 2008.	SUPPORTED	Cover: 'Current Step 4 version dated 4 June 2008'
20	SARAH	ICH Q10 section 3.2.3 change management: QRM evaluates proposed changes with formality commensurate with risk; changes evaluated against the marketing authorisation with 'an assessment to determine whether a change to the regulatory filing is required'; expert teams from development, manufacturing, quality and regulatory evaluate against pre-set criteria; post-implementation evaluation confirms objectives met with no deleterious impact.	SUPPORTED	3.2.3 (a)–(d) (expert teams list also includes 'Medical')
22	SARAH	ICH Q10: working within the design space is not considered a change from a regulatory filing perspective; however, from a quality system standpoint, all changes should be evaluated by the change management system.	SUPPORTED	3.2.3(b): 'working within the design space is not considered a change (from a regulatory filing perspective). However, from a pharmaceutical quality system standpoint, all changes should be evaluated by a company's change management system'
73	SARAH	ICH Q10 includes a change management system as an element of the pharmaceutical quality system.	SUPPORTED	3.2 lists 'Change management system' among the PQS elements; 3.2.3 'Change Management System'
73	SARAH	ICH Q10 has a change management system.	SUPPORTED	3.2 (p.7): 'These four elements are: Process performance and product quality monitoring system; Corrective action and preventive action (CAPA) system; Change management system; Management review'; 3.2.3 'Change Management System'

ICH-Q12.pdf

Line	Speaker	Claim	Verdict	Evidence
8	SARAH	ICH Q12 is titled Technical and Regulatory Considerations for Pharmaceutical Product Lifecycle Management, adopted at Step 4 on 20 November 2019.	SUPPORTED	Cover: 'Final version … Adopted on 20 November 2019'; Document History Step 4 20 November 2019
8	SARAH	ICH Q12 never mentions a model as a thing that changes.	SUPPORTED	'model' occurs twice (3.2.3.1 'e.g., models, Process Analytical Technology'; ch.9 kinetic models); in neither is a model the object of a change or an EC
8	SARAH	Three of ICH Q12's tools are risk-based reporting categories, Established Conditions, and the post-approval change management protocol (PACMP).	SUPPORTED	1.3 lists eight 'regulatory tools and enablers': categorisation of changes (ch.2), ECs (ch.3), PACMP (ch.4), PLCM document (ch.5), PQS/change management (ch.6), assessment/inspection (ch.7), structured approaches (ch.8), stability data approaches (ch.9)
30	SARAH	ICH Q12 chapter 2 describes three reporting levels: prior approval (sufficient risk to require review and approval before implementation); notification (moderate to low risk, communicated formally before or after implementation per regional rules); and changes not required to be reported, 'only managed and documented within the PQS, but may be verified during routine or other inspection'.	SUPPORTED	Chapter 2 (p.9), verbatim
32	SARAH	ICH Q12 chapter 3: Established Conditions are legally binding information considered necessary to assure product quality; any change to an EC necessitates a submission; other dossier content is supportive information; the company proposes ECs and reporting categories with justification and the regulator approves.	SUPPORTED	3.2.1, 3.2.2 ('MAH' rather than 'company'; approval 'where appropriate')
34	SARAH	ICH Q12's decision tree: if a parameter must be controlled to ensure quality it is an EC and the risk if it changes sets the category (high prior approval; moderate or low notification); if not, not an EC and not reported.	PARTIAL	3.2.3.1 text: parameters that need to be controlled 'should be considered ECs'; risk of change 'informs the reporting category … from high to low risk'; Figure 1 is an image not present in the text extract
34	SARAH	ICH Q12 parameter-based vs performance-based approaches; performance-based: 'ECs could be primarily focused on control of process outputs rather than process inputs', enabled by a data-rich environment and enhanced control strategy, examples models and Process Analytical Technology, with 'feedback controls or optimisation algorithms to achieve the relevant targets'; the enhanced control strategy 'may remove the need for certain process parameters to be ECs'.	SUPPORTED	3.2.3.1 (pp.12-13), three quoted fragments verbatim
36	SARAH	ICH Q12 chapter 4: a PACMP provides predictability and transparency; the approved protocol is an agreement between the MAH and the regulator; describes the intended change, how it would be prepared and verified including an impact assessment, and the suggested reporting category, 'a lower reporting category and/or shortened review period as compared to a similar change without an approved PACMP'; sets conditions and acceptance criteria.	SUPPORTED	4.1, near-verbatim (text: 'as compared to similar change procedure without an approved PACMP')
38	SARAH	ICH Q12 PACMP two steps: submit the protocol (proposed changes, rationale, risk management, studies, acceptance criteria, proposed reporting category) for approval before execution; perform studies and submit under the agreed category if criteria are met; if acceptance criteria are not met 'the change cannot be implemented using this approach'; if new information shows increased risk 'the previously approved reporting category should no longer be considered appropriate'.	SUPPORTED	4.2 Step 1 and Step 2; 'the change cannot be implemented using this approach' verbatim; increased-risk sentence in 4.1
40	SARAH	ICH Q12 4.5 types of PACMPs: one or more changes for a single product; and 'a PACMP can also be designed to be used repeatedly to make a specified type of CMC change over the lifecycle of a product, applying the same principles'; examples of broader protocols are a stopper change across products and an analytical method change across sites.	SUPPORTED	4.5, verbatim (also a third example: manufacturing site change across products)
55	SARAH	Q12 is silent on a change the product makes to itself (automatic modification).	SUPPORTED	No 'automatic', 'self', 'continuous', 'adaptive', 'retrain' in Q12; nearest is 'feedback controls or optimisation algorithms' (3.2.3.1)
59	SARAH	Q12 sections 4.1, 4.2 and 4.5 describe the PACMP definition, the two-step process and the types.	SUPPORTED	4.1 Definition; 4.2 Application (the two steps); 4.3 Elements; 4.5 Types
91	SARAH	Q12 puts maintenance of the authorisation on the MAH and change management on the PQS.	SUPPORTED	3.3: 'The management of all changes to, and maintenance of, the approved marketing authorisation is the responsibility of the MAH'; 1.3 and 6.1 on the PQS

Not among the primary documents on disk

Line	Speaker	Claim	Verdict	Evidence
73	SARAH	No quality agreement with a model vendor yet requires notification of model changes. [industry practice; not a document claim]	NOT-CHECKABLE
90	HOST	An industry proposal is circulating that asks for a repeatable AI PACMP extension. [industry proposal, excluded source]	NOT-CHECKABLE	Excluded source; the script deliberately says nothing of its content

Host opinions (listed, not verified)

line 72, HOST: Plants have managed supplier changes to raw materials (film additives, media sub-components two tiers down) for thirty years; both failure modes are real: the change never heard about, and the burden of a hundred irrelevant notifications through which the one that matters slips.
line 72, HOST: Treat the foundation model under the agent as a raw material from a supplier with a sub-supplier: a notification clause, an impact assessment tiered by what could change, a regression-set test for the ones that could matter, a documented 'no impact' for the rest.
line 73, SARAH: The documents support the supplier-change analogy better than they support the model.
line 73, SARAH: The gap is contractual: a resin change arrives with a certificate and a notification because the supply agreement requires it; a model version arrives with release notes at the vendor's discretion; no quality agreement with a model vendor yet requires notification.
line 9, HOST: Nobody wrote the chapter for changes a site receives.
line 27, HOST: Combined effect of multiple changes is the line to underline.
line 29, HOST: The annex doesn't know the vendor exists and still catches the vendor.
line 35, HOST: A performance-based EC makes the model the enabler; Q12 does not say the model is an EC.
line 39, HOST: A PACMP is a bet placed with the regulator.
line 41, HOST: Nobody has written a PACMP whose change is retrain the model.
line 48, HOST: FDA's Q12 paragraph is the hinge; for a template cross to devices.
line 56, HOST: Device guidance supplies structure, Q12 permission; the gap is the automatic case.
line 58, HOST: Sketch of an AI PACMP.
line 63, HOST: Tier rule for instruments.
line 65, HOST: Determinative models handled like a registered method.
line 69, HOST: Minimum vendor practice: pin, contract notice, regression set, version in intended use.
line 75, HOST: One record, two verdicts.
line 77, HOST: Freezing forfeits the adaptive benefit.
line 78, SARAH: Illustrative changes to the soft sensor.
line 80, SARAH: Illustrative changes to the batch model.
line 82, SARAH: Illustrative vendor push to the vision system.
line 84, SARAH: Illustrative deprecation for the agent.
line 87, HOST: Description of Modifications sketch for the soft sensor.
line 88, HOST: Record tier, change types and instrument per system.
line 91, SARAH: Who owns what.
line 92, HOST: A retest decision is only as good as the regression set.
line 72, HOST: Supplier changes to raw materials have been managed in plants for thirty years; the foundation model under an agent should be treated like a raw material from a supplier with a sub-supplier, with a notification clause, a tiered impact assessment, a regression set and a documented no-impact decision.