How to read a VA claims file (C-file): a field guide for representatives and veterans

The claims file, C-file, or eFolder is the electronic folder the Veterans Benefits Administration keeps on each claimant in the Veterans Benefits Management System (VBMS). It holds everything the veteran sent, everything VA collected under its duty to assist, every exam VA ordered, and every decision VA made. Paper C-files still exist for older claims, but what a representative opens today is a VBMS eFolder: a long list of PDFs, each tagged with a document type and a receipt date, many scanned from fax, and some scanned from paper that was itself a photocopy of a fax.

4 kinds of people read it, for 4 different reasons. The accredited representative (VSO officer, claims agent or attorney) reads it to find what the claim is missing before the next decision. The nexus-letter clinician reads it to write an opinion that is grounded in the record rather than in the veteran's account alone. The VA rating specialist reads it to decide. The veteran reads it, often for the first time after a denial, to understand what VA saw and did not see.

The representative's read is the one this guide teaches. It differs from a medical chart review in 2 ways. First, the C-file is adversarial in structure even though the process is not: the question on every page is whether it helps or hurts 1 of 3 elements (an in-service event, a current disability, a link between them). Second, the file contains VA's own prior reading of itself, in rating decisions and code sheets, and that prior reading is sometimes wrong in ways a new reader can prove from the same pages.

Vocabulary you need before page 1

C-file, claims file, eFolder

The same thing at different ages. The eFolder is the VBMS form of the C-file. "Claims folder" appears in older decisions and in 38 CFR 3.159.

STRs

Service treatment records: sick call notes, clinic visits, profiles, dental records, entrance and separation exams, the Report of Medical History the service member filled in at each exam. Older files call them SMRs.

OMPF

Official military personnel file: DD214, orders, evaluations, awards, duty stations, disciplinary actions. The place to prove where the veteran was and what they did.

C&P exam

A compensation and pension examination ordered by VA under 38 CFR 3.159(c)(4), performed by a VA clinician or a contract examiner, and documented on a DBQ.

DBQ

Disability Benefits Questionnaire: the structured form that records an exam in the order the rating schedule asks its questions. VA examiners use the full set; private clinicians can complete the 87 public ones.

Code sheet

The data table attached to a rating decision: each service-connected condition, its diagnostic code, percentage, effective date, and the combined rating. The narrative explains; the code sheet is what VA pays on.

Nexus

The link between the in-service event and the current disability. In a direct service connection claim it is usually a medical opinion stated "at least as likely as not" (50% or greater probability).

POA

Power of attorney on VA Form 21-22 (an organization) or 21-22a (an individual). The document that lets a representative see the file at all.

Duty to assist

VA's obligation under 38 USC 5103A and 38 CFR 3.159 to get the records a claimant identifies and to order an exam when the evidence meets the threshold in 3.159(c)(4).

Intent to file

A communication that holds a potential effective date for 1 year while the claim is prepared, under 38 CFR 3.155(b).

Nobody can read a file they do not have, and the route to the file depends on who is asking. An accredited representative with a POA on record reads the eFolder directly. A veteran, or a representative without electronic access, requests a copy under the Privacy Act. A third party who is neither needs the veteran's written authorization first.

The Privacy Act request on VA Form 20-10206

The form, revised August 2023, asks for the veteran's full name, date and place of birth, mailing address and a handwritten signature; a Social Security number or VA file number helps VA find the file. The instructions list the mailing address (Department of Veterans Affairs, Evidence Intake Center, PO Box 4444, Janesville, WI 53547-4444) and a toll-free fax, and the form can be submitted online from va.gov/forms/20-10206. Fees: VA may charge $0.15 per single-sided page for photocopies; the instructions say "all other requesters" pay only after the first 100 pages and 2 hours of search time. In practice most veterans receive the file on a disc or as a download at no charge, but the fee schedule is on the form if VA applies it.

2 things to say in the request. First, ask for the complete claims file, including STRs, the personnel file, all rating decisions and code sheets, all C&P examination reports, and all correspondence. A request that says "my records" without listing types can come back as the decisions only. Second, ask for it in electronic form with the document index, so you get VBMS's own document-type labels and receipt dates, which the templates in Chapter 12 are built around.

Since October 1, 2019, VBA processes these requests under a Privacy Act procedure that redacts third-party personal information before release. VA's announcement of the change said the agency "is required by the Privacy Act to allow Veterans, or their representatives, the opportunity to review or make copies of claims files", and predicted no delay from the new review. The numbers below say what happened.

How long it takes, measured

VA's FOIA annual report to the Department of Justice breaks VBA's response times into buckets. The report does not separate C-file requests from other records requests, so read these as VBA's performance on the whole stream. 2 facts stand out. The median simple-track request closed in 1 day, and the average took 47.47 days. Both are true because the distribution has 2 humps: most requests close within 20 working days, and a third of them sit for 101 to 140 days.

The same report shows VBA's backlog falling from 10,893 pending requests to 1,121 during the year, and the oldest simple-track request closed in FY2024 had waited 2,911 days. For planning a claim, the practical rule is to file the records request on day 1 of representation, before the first client meeting is over, because the 1-year window to seek review of a decision under 38 CFR 3.2500(a)(1) does not wait for the file.

VBMS does not organize the eFolder into chapters. It stores documents with a type label and a date, and the reader imposes the structure. The sections below are the structure most representatives use, in the order they tend to appear once sorted. The traps column is what a first-time reader gets wrong in each section.

When the STRs are not there

On July 12, 1973, a fire at the National Personnel Records Center in St. Louis destroyed an estimated 16 to 18 million official military personnel files. The National Archives puts the loss at about 80% of Army personnel discharged between November 1, 1912 and January 1, 1960, and about 75% of Air Force personnel discharged between September 25, 1947 and January 1, 1964 with names alphabetically after Hubbard, James E. No duplicates or microfilm existed. For those veterans, and for any veteran whose records VA cannot find, the file should contain VA's requests to the service department, the responses, and a formal finding that further efforts would be futile, because 38 CFR 3.159(c)(2) requires VA to keep asking until it reaches that point. The reconstruction routes (alternate sources such as morning reports, unit records, SGO hospital admission cards, and lay evidence) start from that finding.

Everything in a compensation claim is read against 1 sentence of the regulation.

"Service connection connotes many factors but basically it means that the facts, shown by evidence, establish that a particular injury or disease resulting in disability was incurred coincident with service in the Armed Forces, or if preexisting such service, was aggravated therein."

Courts and VA break that into 3 elements: a current disability, an in-service event, injury or disease, and a nexus between them. Each element has a home section in the C-file, and most reading failures are an element looked for in the wrong section.

4 beliefs about the file that the rule does not support

The presumptions, and how they change what you look for

For a chronic disease listed in 38 CFR 3.309(a) that appeared to a compensable degree within 1 year of separation, 3.307(a)(3) presumes service incurrence and the in-service element is satisfied by the calendar. For exposures (herbicides, Gulf War, radiation, the PACT Act additions), the personnel file does the work: duty stations and dates establish the presumption, and the reader's job shifts from the STRs to the orders and evaluations. For a secondary claim under 3.310, the in-service element is already proven by the existing rating; the reading task is the chain from the service-connected condition to the new one, which lives in treatment records and the opinion.

PTSD has its own rule. 38 CFR 3.304(f) requires 3 things: a diagnosis conforming to 4.125(a), a link between current symptoms and an in-service stressor, and credible supporting evidence that the stressor occurred. The 21-0781 stressor statement, the personnel file, and for fear of hostile military or terrorist activity under 3.304(f)(3) the VA examiner's confirmation, are the pages that carry the third requirement.

A rating decision has 2 parts. The narrative is what the veteran receives: an introduction, the decision on each issue, the evidence list, the reasons for decision, and the references. The code sheet is the data table behind it: diagnostic codes, percentages, effective dates, the combined evaluation, and special notations. VA's M21-1 manual lays out both in Part V, Subpart iv, Chapter 1. Since the Appeals Modernization Act, the statute itself dictates what the notice must say.

1. The issues adjudicated. 38 USC 5104(b)(1). Check them against the 21-526EZ. An issue the veteran claimed and VA did not list was not decided, and that is a different problem from a denial.
2. A summary of the evidence considered. (b)(2). Compare the list with your index. A private MRI you know is in the file but absent from this list is your first argument for Higher-Level Review.
3. A summary of the applicable laws and regulations. (b)(3). Usually boilerplate, but the diagnostic code VA chose is a decision in itself and sometimes the wrong one.
4. Favorable findings. (b)(4). Under 38 CFR 3.104(c), a favorable finding binds every later adjudicator unless evidence identifies a clear and unmistakable error in it. A denial that concedes the in-service event has narrowed the next claim to the nexus.
5. The elements not satisfied. (b)(5). The sentence that tells you what evidence to go get. "The evidence does not show a link" means a nexus opinion; "no current diagnosis" means a treatment record or exam.
6. How to get the evidence VA used. (b)(6). The route back to Chapter 2 of this guide.
7. What it would take for the next higher level. (b)(7). For an increase claim, this is the rating criterion the file failed to meet, in VA's own words.

What a decision opens

Under 38 CFR 3.2500, a claimant has 1 year from the decision notice to request Higher-Level Review (20-0996, no new evidence, a senior reviewer looks for error) or a Board appeal (10182), and may file a supplemental claim (20-0995) at any time with evidence that is new and relevant under 3.2501. Which one fits depends on what the reading found: a decision that ignored a page already in the file points to Higher-Level Review; a missing nexus points to a supplemental claim with the opinion attached. For an increase, 38 CFR 4.16 is the page to keep open: total disability based on individual unemployability needs 1 disability rated 60% or more, or 2 or more with 1 at 40% and a combined 70%, plus evidence the veteran cannot hold substantially gainful employment because of them.

The DBQ asks the rating schedule's questions in the rating schedule's order, so the way to read one is with the diagnostic code open beside it. 3 examples follow, chosen because they cover the 3 shapes a rating criterion takes: a measurement in degrees, a symptom formula, and a test threshold.

Example 1: a knee under DC 5260 and 5261

The Knee and Lower Leg Conditions DBQ walks the examiner through diagnosis (section 1), history including flare-ups and instability (2), initial range of motion (3A), range after at least 3 repetitions (3B), repeated use over time (3C), flare-ups with an estimate in degrees (3D), other contributing factors (3E), muscle atrophy (4), ankylosis (5), instability and subluxation (6), surgery (9), assistive devices (11), imaging (13), functional impact (14A) and the examiner's signature (16). Section 3A carries a note that the joint should be tested for pain on active and passive motion, in weight-bearing and non-weight-bearing, and compared with the opposite joint. That note is 38 CFR 4.59's last sentence, which the Veterans Court in Correia v. McDonald, 28 Vet. App. 158 (2016), read as a requirement for an adequate exam.

Source: 38 CFR 4.71a. 3 things to check on every knee DBQ. First, the number in 3A against the table: flexion to 90 degrees is noncompensable under 5260 on its own, but 4.59 says painful motion of a joint with arthritis shown on imaging warrants at least the minimum compensable rating, and DC 5003 says the same for degenerative arthritis with noncompensable limitation of motion. So a 90-degree knee with pain at 80 and an X-ray in section 13 showing arthritis supports 10%, and an examiner who wrote "no additional loss after repetition" without recording where pain began has left the question open. Second, section 3D: 38 CFR 4.40 and 4.45 require the rater to consider functional loss during flare-ups, and the DBQ asks the examiner to estimate range in degrees during a flare. "Unable to estimate without speculation" is a common entry; Sharp v. Shulkin, 29 Vet. App. 26 (2017), held the examiner must explain why, after eliciting the veteran's own description. Third, section 6: DC 5257 rates instability separately from motion, with 10%, 20% and 30% tiers that turn on the ligament finding (sprain, incomplete tear, or unrepaired or failed-repair complete tear) and whether a medical provider prescribed a brace, an assistive device, or both. A DBQ that says "stability testing not performed" or "unable to test" next to a history of giving way is an incomplete exam under 38 CFR 4.2, which tells the rater to return a report that does not contain sufficient detail.

Example 2: PTSD under DC 9411 and the General Rating Formula

Every mental disorder in 38 CFR 4.130 is rated on the same formula, and the DBQ for mental disorders records the symptom checklist, occupational and social history, and the examiner's summary level in the formula's own words. 4.125(a) requires a diagnosis that conforms to DSM-5, and tells the rater to return a report that does not support its diagnosis. 4.126(a) tells the rater to assign the level based on all the evidence of occupational and social impairment, not the examiner's checkbox alone.

Source: 38 CFR 4.130, General Rating Formula for Mental Disorders, paraphrased. The formula says "such symptoms as", so the listed symptoms are examples of a level, and 4.126(a) ties the level to the occupational and social impairment the symptoms cause. On the DBQ, that means reading the symptom checklist and the narrative history together. A checklist with suicidal ideation ticked and a summary level of 30% is an internal conflict for the rater to resolve under 4.126(a), and a page to cite. The VA treatment records then supply what the DBQ cannot: frequency over time, missed work, medication changes, hospitalizations, each dated.

Example 3: hearing loss under 38 CFR 3.385

Hearing is a threshold, which makes it the easiest DBQ to read and the easiest to misread. For VA purposes, impaired hearing is a disability only when the auditory threshold in any of the frequencies 500, 1000, 2000, 3000 or 4000 Hertz is 40 decibels or greater; or the thresholds for at least 3 of those frequencies are 26 decibels or greater; or speech recognition using the Maryland CNC Test is less than 94%. A private audiogram that meets the threshold still needs to show which word list was used; many private clinics do not use Maryland CNC, and VA's examiner will repeat the test. The in-service element lives in the entrance and separation audiograms in the STRs, read side by side for a shift, and in the personnel file for the military occupational specialty and its noise exposure.

These are the errors I see most in files that come back from a denial, ranked by how much each one costs when it is missed. None of them is exotic. All of them come from reading a 1,200-page file the way people read: front to back, faster as it goes, and fastest through the handwritten pages.

What the read produces

3 things, each cited. First, the in-service event is established twice over, at STR p. 188 and OMPF p. 377, and the 2024 decision should have recorded it as a favorable finding under 38 USC 5104(b)(4); the 2016 decision's "STRs silent" was wrong on the file. Second, the private DBQ's extension limited to 10 degrees and lateral instability with a prescribed brace are findings that DC 5261 and DC 5257 rate separately from flexion, and the 2024 code sheet has neither. Third, the 2024 VA exam is incomplete on its face: no stability test, no flare-up estimate with an explanation, and no pain-on-motion findings in passive or non-weight-bearing testing. The reading points to a Higher-Level Review on the codes VA did not apply, or a supplemental claim with a corrected exam. Which one is the representative's call, and so is the effective date argument under 3.400(o)(2) for the 2021 MRI.

A C-file read splits into page handling and judgment. Page handling is sorting, deduplicating, dating, finding every mention of the knee, and typing out what each page says with its number. Judgment is deciding which diagnostic code applies, whether the exam is adequate, what the next filing should be, and what to ask the clinician. In the hypothetical above, page handling was about 15 of 21.5 hours. That ratio is what AI medical record review changes, and it changes nothing else.

The last row is the same in both columns on purpose. Software that organizes a C-file produces an index, a cited chronology and a list of flags; it does not know which diagnostic code applies or what to file next, and a tool that claims to should be treated the way the Veterans Court treats an opinion without a rationale. The hours it removes come from the first 6 rows, and only if the reader then spends the saved time on the pages the tool pointed at.

Veterans law practices and VSOs are being sold legal AI tools and AI document review platforms that promise to read the file. Some of that promise is real and some is not, and a C-file is a harder test than most medical records because so much of it is handwritten, faxed, scanned from carbon copies, and 15 years old.

What a large language model does well on a C-file

• Indexing at volume. Document type, provider, date of service and page range for each PDF in the eFolder, which is the first 3 hours of every manual read.
• Finding every mention. Every page that mentions the knee, the back, the stressor, or the word "jump", listed with its page, including the sick call note between the dental pages.
• Deduplication. Grouping the 3 copies of the 2019 orthopedics note and showing the 2 copies of the exam that differ.
• A cited chronology. An AI medical chronology that lists each event with the page it came from, so the reader can open the citation instead of trusting the summary.
• Flags. Pages whose name or date of birth does not match the file; pages with no date; gaps between dated records; ordered tests with no result in the file.
• Questions over the record. "Where does any clinician state the cause of the knee condition?" answered with page citations, which is faster than reading 555 treatment pages.

What it gets wrong, and why military records make it worse

• OCR on faxed and photocopied STRs. Optical character recognition reads a faxed "110" as "118" and a smudged "R" as "B". Older STRs were photocopied before they were scanned. A misread measurement is worse than a missing one because it looks like evidence.
• Handwriting. Sick call entries, profiles and chaplain notes are handwritten in military shorthand. Handwriting recognition gets some of it; a tool that routes the unreadable pages to a person is usable, and one that guesses silently is not.
• Dates. A model that uses the VBMS receipt date as the event date builds the wrong chronology. It has to date pages by content and say when it could not.
• Hallucination. Ask generative AI "does this veteran have a nexus?" and it can produce a confident paragraph the file does not support. In Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), lawyers were sanctioned under Rule 11 for filing case citations a chatbot invented. A fabricated page citation in a Board brief is the same error.
• Templated VA notes. A problem list that says "PTSD" on every visit for 6 years looks like 6 years of evidence to a model; it is 1 entry copied forward. Clinical NLP that counts mentions mistakes copy-forward for severity.
• Judgment. No model knows whether 5257 should be rated with 5260, whether the exam is adequate, or what to file. Those stay with the representative, and a tool that outputs a win probability has crossed a line it should not.

Privacy, the Privacy Act and the BAA

A C-file is a system of records under the Privacy Act, and it holds protected health information once it is in a clinician's hands. The veteran's records stay the veteran's: a tool that uploads them to a vendor that trains on them, or stores them without a business associate agreement, puts the practice on the wrong side of HIPAA and of the veteran's trust. The checklist below is the floor for any HIPAA compliant AI vendor a VSO, law firm or nexus clinic allows near an eFolder.

• A signed business associate agreement before the first uploadNo BAA, no C-file. Consumer chatbots do not offer one.
• SOC 2 report available on requestAn independent audit of the controls, not a badge on a website.
• Written commitment that your files do not train the modelIn the contract, not the FAQ.
• A page citation on every output lineChronology, summary, answer. A line without a page is a line you cannot verify.
• Handwriting and low-quality scans are marked, not guessedAsk to see what the tool does with a faxed 2009 sick call page.
• Flags, never verdicts"Name on page differs from file" is a flag. "Claim likely to succeed" is a verdict and a warning sign.
• Retention you control and deletion on requestThe veteran can ask you to delete; you need to be able to.
• Access by named user, so you know who opened whatThe eFolder has that discipline; your tools should match it.

0 of 8 checked

Human-in-the-loop is the phrase vendors use; the practice behind it is the one in Chapter 11. An AI medical records summary is usable only as a set of page-level citations a person opens, and the person who files the brief owns every one of them.

A cited summary is a list of claims about pages. Verification means opening the pages. How many you open depends on what the summary is for, and the mistake is to apply 1 tier to everything: tier 1 on a 1,200-page intake wastes a day, and tier 3 on a Board brief invites a Mata problem.

2 rules hold across tiers. A line that cannot be traced to a page is deleted, not softened. And the tier is recorded on the summary, so the next person who picks it up knows what it has been through.

Copy these into your own documents. Each has bracketed fields and a rule at the bottom. The first 2 are what the hypothetical in Chapter 8 was built on; the third is how to ask a tool for the same thing without inviting it to decide the claim.

C-FILE INDEX
Veteran: [NAME]   File no.: [NUMBER]   Rep: [NAME]   POA on file: [21-22 / 21-22a, DATE]
Source of file: [VBMS eFolder / 20-10206 copy received DATE]   Total pages: [N]
Unique pages after duplicates: [N]   Wrong-veteran pages excluded: [PAGES]   Undated pages: [PAGES]

A. SECTION MAP (page ranges)
   STRs [X to Y] | OMPF [X to Y] | Claim forms [X to Y] | Rating decisions + code sheets [X to Y]
   C&P exams/DBQs [X to Y] | VA treatment [X to Y] | Private treatment [X to Y]
   Lay statements [X to Y] | Correspondence/notices [X to Y] | Appeals [X to Y]

B. CLAIMS HISTORY
   [CONDITION] | claimed [DATE, FORM, p. X] | decided [DATE, p. X] | result [GRANT %/DENIED] | element failed [p. X]
   Review window open until: [DATE + 1 YEAR]   Intent to file on record: [DATE, p. X]

C. PAGES BY ELEMENT (1 row per page, content date not receipt date)
   Element | Content date | Page | Document | What it says (quote) | For / against / neutral
   In-service event  | [DATE] | p. [X] | [STR sick call] | "[EXACT WORDS]" | [FOR]
   Current disability| [DATE] | p. [X] | [DBQ sec. 1 / VA note] | "[DIAGNOSIS]" | [FOR]
   Nexus             | [DATE] | p. [X] | [opinion] | "[PROBABILITY PHRASE + RATIONALE]" | [FOR/AGAINST]

D. EXAMS AND DBQs (1 row each)
   Date | Examiner / type | Pages | Measurements (quote) | Sections blank or "unable" | Opinion + pages cited | File review claimed?

E. CODE SHEET CHECK
   DC | % | Effective date | DBQ finding that supports a different DC or % | Page

F. EVIDENCE LIST VS FILE
   Document in file not on the decision's evidence list | Page | Why it would have mattered

Prepared by: [NAME]   Date: [DATE]   Verification tier: [1 / 2 / 3]
Facts only. Every row has a page. No row states merit.

RECORDS-GAP LIST
Veteran: [NAME]   File no.: [NUMBER]   Claim: [CONDITION(S)]   Date: [DATE]

1. FEDERAL RECORDS NOT IN THE FILE (38 CFR 3.159(c)(2): VA requests until further efforts are futile)
   Record: [STRs for DATES / OMPF / unit records / VAMC records from FACILITY, DATES]
   Evidence it exists: p. [X] ("[QUOTE]")
   VA request in file? [YES p. X / NO]   Response in file? [YES p. X / NO]   Futility finding? [YES p. X / NO]

2. PRIVATE RECORDS NOT IN THE FILE (3.159(c)(1): initial request plus at least 1 follow-up)
   Provider: [NAME, CITY, DATES]   Named at: p. [X]
   21-4142 on file? [YES p. X / NO, attached]   VA request? [p. X]   Follow-up? [p. X]   3.159(e) notice? [p. X, DATE]
   Status: [WE HAVE REQUESTED DIRECTLY ON DATE / ATTACHED / VA TO REQUEST]

3. EXAM ADEQUACY (3.159(c)(4); 4.2; 4.59; 4.40 and 4.45)
   Exam dated [DATE], p. [X]: [section blank / "unable to test" / no flare estimate / no passive or non-weight-bearing testing / opinion does not address p. X]
   Request: [new examination / addendum opinion addressing p. X / accept private DBQ at p. X under 3.326(b)]

4. LAY EVIDENCE TO BE SUBMITTED
   [21-10210 from NAME, covering EVENT/DATES]   [21-0781 stressor statement]

5. FAVORABLE FINDINGS TO CITE (38 USC 5104(b)(4); 38 CFR 3.104(c))
   Finding: "[QUOTE]"   Decision dated [DATE], p. [X]

Prepared by: [NAME]   Reviewed against the file on: [DATE]
Every gap cites the page that shows the record exists or was named.

You are organizing a VA claims file. Do not assess merit, probability of success, or what to file.
Do not rate any condition or name a diagnostic code. Cite the page number of this file for every
line; a line you cannot cite must be omitted. Date each page by the content on the page, not by
the receipt date; mark "undated" if no date appears on the page.

Claimed condition(s): [CONDITION(S)]

1. INDEX: list every document with type, provider or author, content date range, and page range.
2. DUPLICATES: list groups of identical or near-identical pages. For near-identical groups, quote
   the text that differs and give both page numbers.
3. HEADER CHECK: list every page where the name, date of birth or file number differs from the
   veteran's. Quote the differing text.
4. EVERY MENTION: list every page that mentions [CONDITION / BODY PART / STRESSOR TERMS], with the
   content date and a verbatim quote of the relevant line. Include service treatment records,
   personnel records, exams, treatment notes, lay statements and correspondence.
5. EXAMS: for each C&P exam or DBQ, list the date, the examiner's stated credentials, every
   measurement as written, every section marked blank, "unable to test" or "without speculation",
   the opinion's probability phrase verbatim, and each page the examiner cites.
6. DECISIONS: for each rating decision, list the issues, the evidence list verbatim, each favorable
   finding verbatim, each element stated as not satisfied verbatim, and the code sheet entries
   (diagnostic code, percentage, effective date).
7. GAPS: list every provider or record named anywhere in the file for which no records appear,
   with the page where it is named, and every VA notice that says a record could not be obtained.
8. CHRONOLOGY: in content date order, 1 line per event, each with a page citation.

Output as plain text tables. Mark anything you could not read as [ILLEGIBLE p. X].

Medrecords AI is medical record review software. You upload the C-file you already have, from VBMS or from a 20-10206 request. It does not request the file from VA, retrieve records from providers, or contact an examiner.

Its flags are signals, not verdicts. A flag that says p. 917 carries a different date of birth tells you where to look; whether to exclude the page is your decision. It runs at 100 pages a minute, so a 1,200-page eFolder is indexed in about 12 minutes. Security is SOC 2 and HIPAA with a BAA (security, HIPAA); retention is configurable and deletion is on request. Self-Service starts at 10 cents a deduplicated page, down to 5 cents at volume. Duplicate pages are free. Enterprise On-Prem is an annual license, no per-seat fees. See VA C-file review, medical record review for VA-accredited representatives and nexus letter record review.

How do I request my VA C-file?

File VA Form 20-10206, the FOIA or Privacy Act request, online at va.gov or by mail or fax to the Evidence Intake Center in Janesville, Wisconsin. Tick "Claims File (C-File)" and any other record types you want, sign it by hand, and ask for the complete file in electronic form with the document index.

How long does a C-file request take?

VA's FY2024 FOIA report shows VBA's median simple-track response at 1 day and the average at 47.47 days, because about 60% of requests close within 20 days and about a third take 101 to 140 days. The report does not separate C-file requests from other records requests. File on day 1 of representation.

What is the difference between a C-file and an eFolder?

The eFolder is the electronic C-file inside VBMS. Older claims had paper C-files; most have been scanned. Accredited representatives with a POA on record read the eFolder directly; veterans receive a copy of it through a Privacy Act request.

What is a code sheet in a VA rating decision?

The data table behind the narrative: each service-connected condition with its diagnostic code, percentage and effective date, the combined evaluation under 38 CFR 4.25, and special notations. It is what VA pays on, and it is where a missing diagnostic code shows up.

Can a private DBQ be used instead of a VA exam?

38 CFR 3.326(b) says VA accepts a private examination report if it is adequate for rating purposes without a further exam. 87 DBQs are public and can be completed by a private clinician; the initial PTSD DBQ is not among them, so a private PTSD opinion comes on the mental disorders DBQ or as a letter.

What if the STRs are missing or were destroyed in the 1973 fire?

38 CFR 3.159(c)(2) requires VA to keep requesting Federal records until further efforts are futile, and the file should show those requests and a finding. The claim then rests on alternate sources: unit records, morning reports, hospital admission cards, and lay statements, plus the benefit of the doubt under 38 USC 5107(b).

Can AI read a VA C-file?

For the page handling, yes. AI medical record review tools index documents, group duplicates, flag wrong-veteran and undated pages, and build a chronology with a page citation on each line. OCR on faxed and handwritten STRs is the weak point, and every cited line still has to be opened by a person before it is filed.

Is it HIPAA compliant to upload a C-file to an AI tool?

Only with a signed business associate agreement, SOC 2 controls, and a written commitment that your files do not train the model. A consumer chatbot without a BAA is not a place for a veteran's records. The Privacy Act governs VA's side; HIPAA governs the clinician's and, through the BAA, the vendor's.

Can ChatGPT write a nexus letter from the C-file?

A chatbot can draft prose, but a nexus opinion's weight comes from a clinician's reasoning on accurate facts, and a general model without page-level citations can state facts the file does not contain. The clinician reads the cited pages and writes the opinion; software can hand them the index and the chronology.

Does Medrecords AI decide claims or choose ratings?

No. It organizes, cites and flags what is in the file you upload. Choosing a diagnostic code, judging whether an exam is adequate, and deciding what to file stay with the accredited representative; the opinion stays with the clinician.

Regulations and statutes were read at law.cornell.edu and quoted text is verbatim. The 20-10206 facts come from the form's August 2023 PDF and its instructions. The response-time figures come from VA's FOIA Annual Report for fiscal year 2024, section VII; the report covers all VBA requests, not C-file requests alone. The knee DBQ section list comes from the public form. The M21-1 rating decision structure is cited from VA's published manual and is VA's internal guidance, subject to change. Case holdings are summarized, not quoted, except where marked. The worked example, the code sheet mock and the hours ledger are hypothetical. Product facts come from this site's product pages. Nothing here is legal or medical advice.

• 38 CFR 3.303, principles of service connection; 3.304, direct service connection and PTSD; 3.307 and 3.309, presumptions; 3.310, secondary; 3.385, hearing loss.
• 38 CFR 3.159, duty to assist: (c)(1), (c)(2), (c)(4), (e); 3.155, intent to file; 3.156, new evidence and service department records; 3.400, effective dates; 3.2500 and 3.2501, review options; 3.102, reasonable doubt; 3.326, examinations.
• 38 CFR 4.71a, DC 5003, 5257, 5260, 5261; 4.130, General Rating Formula for Mental Disorders; 4.2, 4.3, 4.16, 4.25, 4.40, 4.45, 4.59, 4.125, 4.126.
• 38 USC 5103A, duty to assist; 38 USC 5104(b), decision notice elements; 38 USC 5107(b), benefit of the doubt; 5 USC 552(a)(6)(A)(i), the 20-working-day FOIA clock.
• VA Form 20-10206 and the form PDF with instructions (August 2023): record types, required fields, mailing address, third-party and POA signature rules, fees.
• VA FOIA Annual Report, fiscal year 2024: VBA requests received, processed, backlog, median and average response times, and the response-time buckets in section VII.C.
• VA news release, October 21, 2019, on VBA's Privacy Act process for claims file requests.
• VA Office of General Counsel, accreditation: 21-22 and 21-22a, and read-only remote access to VBMS claims records for accredited attorneys and claims agents.
• VA public DBQ list, 87 forms, including the Knee and Lower Leg Conditions DBQ whose sections Chapter 6 follows.
• National Archives, the 1973 fire: records lost by branch, date range and estimated percentage.
• Nieves-Rodriguez v. Peake, 22 Vet. App. 295 (2008); Correia v. McDonald, 28 Vet. App. 158 (2016); Sharp v. Shulkin, 29 Vet. App. 26 (2017); Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023).

Related guides: the AI-native VA claims agent, a verified medical chronology from scanned records, how to get medical records for a lawsuit, medical abbreviations for legal professionals, medical source statements and RFC, and VA C-file review.