What Evidence Actually Wins at the Board

Every VA-claims guide says "get a nexus letter." This page says how much it has mattered — measured from the full text of the public BVA archive: for each kind of evidence, how often the Board granted the issue when the record had it, and how often when it did not.

By condition

Each condition page splits the numbers by service connection and by increased rating, because the evidence that matters is different for each.

Board-wide, all conditions

All single-issue decisions (n = 852,152)

"Granted" = share of issues decided on the merits (granted + denied); remands and dismissals are excluded from that fraction and shown separately. Fractions are the raw counts.

By benefit type

Service Connection issues (n = 362,397)

"Granted" = share of issues decided on the merits (granted + denied); remands and dismissals are excluded from that fraction and shown separately. Fractions are the raw counts.

Increased Rating issues (n = 109,671)

"Granted" = share of issues decided on the merits (granted + denied); remands and dismissals are excluded from that fraction and shown separately. Fractions are the raw counts.

TDIU issues (n = 30,954)

"Granted" = share of issues decided on the merits (granted + denied); remands and dismissals are excluded from that fraction and shown separately. Fractions are the raw counts.

Effective Date issues (n = 29,227)

"Granted" = share of issues decided on the merits (granted + denied); remands and dismissals are excluded from that fraction and shown separately. Fractions are the raw counts.

Reading these tables honestly

Association, not cause. A positive nexus opinion appears in the records of granted appeals far more often than in denied ones. That does not mean any letter wins; it means the Board grants when the evidence supports the claim, and a nexus opinion is how that support usually reaches the file.

Appeals are not claims. Every decision here is an appealed denial. Grant rates at the Board say nothing about the share of initial claims VA grants.

Findings are outcomes in disguise. Rows under "how the Board read the record" — credibility, benefit of the doubt, an inadequate exam — describe how the judge decided, not evidence to gather. They are shown because they say which arguments the Board actually reaches for.

Detection is by text pattern. Each evidence type is found by reading the decision for finding-shaped language with negation handling; precision was read on samples before publication and is stated per feature in the method notes below. Expect a few percent of misclassified rows, not zero.

Method notes, per evidence type

A medical source gave a positive nexus opinion

v2 requires a medical-source subject and a service object; 7/9 correct in the last spot check, the two misses ('less than likely', 'no further ... positive opinions') were then added to the negation list.

Private physician opinion, letter or DBQ

7/8 correct (one 'treating clinician found the knee locked').

Disability Benefits Questionnaire in the record

Literal; DBQs exist only from 2010 on.

Third-party lay statements (spouse, buddy, employer)

v1 counted the 'medical or lay evidence' standard; v2 requires a named third party. 7/8 correct.

Service treatment records document the condition or injury

8/8 correct after a 32-char post-window for 'shows no complaint'.

Service treatment records silent for the condition

9/9 correct.

Continuity of symptoms argued (38 CFR 3.303(b))

Literal.

Board hearing held or testimony taken

7/8; cancelled hearings now excluded.

Social Security records in the file

Literal.

Vocational expert or assessment

Literal.

Medical literature or treatise cited

Literal.

Functional loss / DeLuca factors argued (rating cases)

Literal on DeLuca, 4.40/4.45/4.59, flare-ups, painful motion.

Presumptive service connection discussed

Includes 3.307/3.309 citations; treat as 'theory in play'.

Secondary service connection (38 CFR 3.310) in play

Literal on 'secondary to' / 3.310.

Combat service / 1154(b) in play

Literal.

Military sexual trauma / personal assault claim

Literal.

A medical source gave a negative nexus opinion

9/9 correct in spot check.

No nexus opinion of record

v1 caught 'no opinion to the contrary'; v2 excludes it. 8/8 correct.

VA examination or VA medical opinion obtained

High-frequency literal; spot checks clean.

Board found a VA exam or opinion inadequate

9/9 correct in spot check (real inadequacy discussions, incl. remand history).

Board found the VA exam or opinion adequate

v1 matched the Barr v. Nicholson standard; v2 requires a Board finding. 8/9 correct.

Board found the veteran's statements credible

v2 requires a finding; 'no credible evidence' excluded. 8/9.

Board found the veteran's statements not credible or inconsistent

v2 requires a finding; caselaw recitations ('impeached by inconsistent statements') excluded. ~7/9 after the last fix.

PTSD stressor corroborated, verified or conceded

v2 requires a finding; 7/8 correct. Rare outside PTSD.

PTSD stressor not corroborated

7/9; 'not required to accept uncorroborated' excluded in v2.

Benefit of the doubt resolved in the veteran's favor

v2 drops the 5107(b) recitation; 7/8 with 'not in equipoise' now excluded.

Preponderance of the evidence against / doctrine not applicable

8/9 correct.

Duty-to-assist error found

v2 excludes 'no error' and the 20.802 standard; ~7/8.

Board ordered a new exam or addendum opinion

v2 imperative/future forms only; 8/8 correct. Essentially a remand marker.

AMA favorable finding acknowledged

Literal (AMA only).

Extractor version 2. Source code: bva_evidence.py (feature patterns) and bva_evidence_stats.py (aggregation); the decision corpus is public domain (CC0) at va.gov/vetapp.

See also: outcome rates by condition · what overturns the Board at the Court · judge statistics.