This is the verbatim audit record behind 13 answers small businesses should know about SBIR — every claim checked, every verdict, and the predictions that turned out to be wrong. It is published because a verification badge that cannot be inspected is just a logo.
Run: 2026-07-30 · Scope: Q7–Q13 (Part 2 "Foundations", never human-checked) plus a targeted risk check on the date-sensitive Part 1 claims that carry the ✓. Method: every claim checked against live primary/official sources fetched during the pass. Nothing adjudicated from model memory — the document leans on April 2026 legislation that post-dates this model's training cutoff.
Verdicts: SUPPORTED · IMPRECISE · WRONG · UNVERIFIED · MISSING
All findings in §1, §2 and §3 below have been applied to sbir-13-answers.html, and the PDF was re-rendered from the corrected HTML. The findings text is kept verbatim as the audit record.
Applied: NIH inversion corrected (Q12) · Phase III "sole-source" replaced with "without further competition" + 15 U.S.C. 638(r)(4) (Q7) · all six non-resolving citation chips replaced with real instruments (Q4, Q6, Q7, Q8, Q11, Q12, Q13) · award figures and the $30M Strategic Breakthrough Award added (Q11) · SAM annual-renewal trap and 6–8 week guidance added (Q10) · the six-month lapse added to the "it's dead" myth (Q6) · the unsourced "large share of awardees won on a resubmission" statistic removed (Q12) · Part 2 divider now states these answers are cite-checked but not agent-adjudicated or human-signed, so they carry no ✓.
Read directly from Public Law 119-83, § 3 (amending 15 U.S.C. 638(ff)(3)), fetched from govinfo:
"A Federal agency may award from a strategic breakthrough allocation not more than $30,000,000 to a small business concern, including its affiliates, in a single award or series of awards based on reaching production or development milestones, if the total period of performance of the project with respect to which such funds are awarded is not more than 48 months."
The figure and the single-award-or-milestone-series framing were correct. But the statute carries gates the draft omitted, and reading it changed the answer materially — it is a Phase II mechanism, capped at 48 months, and requires at least one prior Phase II award plus 100% matching funds from new private capital or non-SBIR government funding. Agencies fund it from an allocation capped at 0.50% of extramural R&D budget. Q11 now says all of that; "rare and mission-driven" was true but far too vague to act on.
All four S. 3971 citations were also repointed from congress.gov (which bot-walls automated checks) to the govinfo Public Law text, which is both primary and fetchable. Link sweep improved to 32 OK / 1 BLOCKED.
Q6 gained new prose during this pass (the six-month-lapse sentence), which invalidated its prior signature. Frank Kurka reviewed the amended text and signed it on 2026-07-30. The ✓ on Q6 is current.
Q6's as_of was moved 2026-07-23 → 2026-07-30 to match. Leaving it at the 23rd would have dated the signature to before the text existed — the badge would have been asserting a review that never happened on that wording.
Signed text:
Myth — "it's dead": it genuinely did lapse — both programs expired 2025-09-30 and ran dark for six months — but S. 3971 was signed 2026-04-13 and reauthorizes them through 2031-09-30, with new anti-mill proposal caps from FY2027.
Q4 and Q5 had only citations changed — per the rule in PUBLISHING.md, citation fixes strengthen a ✓ answer and do not require re-signing. Their signatures and as_of dates stand.
No open items on this read.
The read is promotable with edits, not as-is. Nothing in it is dangerous, but there is one factual inversion, one terminology error a contracting officer would catch, three material omissions, and a citation-integrity problem that contradicts the document's own stated promise.
The citation problem is the one that actually threatens the ✓, because it is visible without any subject-matter knowledge.
The how-to-read panel promises:
"Every claim is tied to a primary source — a statute, a regulation, or an official agency page."
Several citation chips are topic labels styled to look like sources. They resolve to nothing:
| Chip | Where | Problem |
|---|---|---|
your solicitation | Q11 | Not a source. It is an instruction. |
Phase I win rates by agency (2026) | Q6, Q12 | No publisher, no URL, no document. |
2026 reauthorization / (anti-mill caps) | Q6, Q13 | Real law, but uncited. It is S. 3971, the Small Business Innovation and Economic Security Act of 2026, signed 2026-04-13. |
SBIR data rights | Q4 | Topic label. |
DFARS 20-year rule | Q4 | Real rule, uncited. It is DFARS 252.227-7018, final rule published 2024-12-17, effective 2025-01-17. |
bare sbir.gov | Q6, Q7, Q8, Q12, Q13 | Domain, not a page. |
Three of these sit on ✓ VERIFIED answers (Q4, Q6). That is the worst placement: the badge asserts a cite-check that the chip visibly fails to show.
Fix: either cite the actual instrument (all six have a real one — they are listed above) or drop the chip. A verification brand cannot ship a decorative citation.
Reads: "NIH is the most competitive and the highest-volume."
Highest-volume is correct. Most competitive is backwards. NIH Phase I success has averaged 23.9% since 1994, against roughly 17% overall (FY2025) and ~15–20% at most other agencies. By success rate NIH is one of the better odds, not the worst.
This matters more than a stray statistic: Q8 and Q12 exist to help someone choose which agency door to walk through. This sentence pushes a reader away from the agency with the highest historical hit rate.
Fix: "NIH is the highest-volume agency and historically posts above-average Phase I success rates (~23.9% since 1994); its dual review (scientific merit + commercialization) is a different bar, not a longer shot."
Reads: "an agency can award Phase III follow-on work ... sole-source, without re-competition."
"Without re-competition" is right. "Sole-source" is technically wrong and it is a known precision point in this community. Statutorily a Phase III is an award without further competition under 15 U.S.C. 638(r)(4) — it is expressly not made under the FAR Part 6 sole-source provisions. Agency guidance says this explicitly.
Using the colloquialism on a page selling precision is the exact failure mode a knowledgeable reader punishes.
Fix: drop "sole-source"; say "without further competition (15 U.S.C. 638(r)(4)) — statutorily distinct from a FAR Part 6 sole-source award."
SBIR/STTR expired 2025-09-30 and was dark until 2026-04-13. A document dated July 2026 presents the program as continuous and never says this happened.
Anyone who tried to apply in that window lived through it. Its absence is most conspicuous to precisely the audience being targeted, and it undercuts Q6's "Myth — it's dead" bullet, which answers the myth without acknowledging the six months when it was, functionally, true.
The question is literally "How much money is it?" The answer discusses guideline amounts, inflation adjustment, and fee treatment, and never gives a number. sbir.gov publishes them plainly:
"As of April 2026, agencies may issue a Phase I award (including modifications) up to $323,090 and a Phase II award (including modifications) up to $2,153,927 without seeking SBA approval."
The "confirm your solicitation" caveat is right and should stay — but it is a caveat, not a substitute for the number.
The 2026 reauthorization created strategic breakthrough awards up to $30 million, as a single award or a milestone-tied series. This is the largest change to SBIR award economics in the law the document already cites twice, and Q11 — the money question — omits it entirely.
Q10 correctly warns that registration takes weeks. Official guidance says 6–8 weeks, and adds that SAM requires annual renewal, and renewal can take as long as the initial registration. The renewal trap catches second-time applicants — exactly the readers who think they are already registered.
Q9 (SBIR vs STTR) is the strongest answer in Part 2 — every claim checks out:
Also confirmed:
Part 1 risk check — all four date-sensitive ✓ claims survived:
| Claim | Verdict |
|---|---|
| Q5 TABA $6,500 Phase I / $50,000 Phase II | SUPPORTED |
| Q5 April 2026 vendor freedom, no approved-vendor list, + cyber & foreign-risk screening | SUPPORTED |
| Q6 reauthorized through 2031 | SUPPORTED — signed 2026-04-13, runs to 2031-09-30 |
| Q4 DFARS 20-year data rights, effective Jan 2025, then Government Purpose Rights | SUPPORTED — effective 2025-01-17; post-expiration license is perpetual GPR |
Before running, the pass predicted: structural claims mostly hold; the failure zone will be numbers and recent legislation; at least 4 citation chips will not resolve.
Worth keeping: the AI-drafted content was most trustworthy exactly where it was predicted to hallucinate, and failed on judgment and completeness instead.
Do not promote until §2 (both errors) and §1 (citation chips) are fixed. Those are cheap — an hour of edits — and they are the two categories a hostile expert reader checks first.
§3 omissions are a content decision, not a correctness blocker, but the six-month lapse and the $30M award are the two most newsworthy SBIR facts of 2026 and their absence dates the document.
On the ✓ framing: the four Part 1 claims spot-checked here all held, so the badge is not currently making a false factual claim. But the badge's stated meaning includes "cite-checked," and Q4 and Q6 carry chips that cite nothing. Fix the chips or narrow the badge language.
Concept · rexxit · content is real, verified analysis · not affiliated with Reddit