Three days until CreditRefresh launches publicly. Yesterday's post covered the one-tap pull that loads all three bureau reports into the app as structured data. Today's covers what happens next: the AI scan that reads every line of those reports and flags items that look inaccurate, outdated, or unverifiable under specific FCRA subsections.
This is the step that, done manually, requires per-item legal research. A consumer reading their own reports has to know which FCRA subsection applies to which kind of error, and even careful consumers usually miss subtle inconsistencies that an automated scan catches. Here is what the scan actually checks for, how the results get categorized, and why the legal accuracy at this step is what makes the downstream dispute letters effective.
What the Scan Is Looking For
The scan runs every account and every personal information field on the three bureau reports through a classification model trained on the structure of the Fair Credit Reporting Act. The categories the model uses correspond directly to specific FCRA subsections, because the subsection determines the legal argument that will eventually be made in the dispute letter.
The major categories the scan checks for are: inaccuracies under § 1681i(a)(1), outdated items past the seven-year reporting limit under § 1681c(a), duplicate reporting of the same debt from multiple furnishers, cross-bureau data inconsistencies, items that already received a prior dispute response and are candidates for Method of Verification follow-up under § 1681i(a)(6)(B), and items with patterns suggesting mixed-file or identity theft errors under § 1681c-2. Each category triggers a different downstream dispute strategy.
The scan also identifies items that are not disputable. Accurate debts that you owe, current payment history that matches your records, recent late payments that you actually missed — these are visible in the scan but tagged as non-disputable. Marking them clearly serves two purposes. It is honest about what the FCRA cannot do for you, and it focuses the dispute campaign on items where the legal argument has actual merit.
The Outdated-Item Check
Under 15 U.S.C. § 1681c(a), most negative items must be removed from a credit report seven years after the date of first delinquency. The scan calculates the seven-year mark for every derogatory account and flags items that should already be off the report. It also flags items that are within six months of the seven-year mark, which become high-priority candidates for upcoming disputes.
Re-aging is a specific pattern the scan looks for. When a debt has been sold from the original creditor to a debt buyer or collection agency, the original date of first delinquency does not reset. If the new furnisher reports the account with a more recent date than the original delinquency, that is a clear FCRA violation. The scan compares dates across the three bureaus and flags accounts where the date appears to have moved forward in time — a typical re-aging signature.
Chapter 7 bankruptcies follow a separate ten-year limit under § 1681c(a)(1). The scan accounts for the different timeline. Public records, judgments, and tax liens have their own reporting rules, some of which the bureaus have voluntarily updated in recent years — most tax liens, for example, no longer appear on credit reports at all following the National Consumer Assistance Plan changes. The scan reflects the current rules.
The Cross-Bureau Reconciliation
Each account that appears on more than one bureau gets compared field-by-field. The scan checks balance consistency, status consistency (open/closed/charged-off), payment history alignment, date of first delinquency consistency, date opened consistency, and original creditor consistency. Any field where the bureaus disagree gets flagged as a candidate for an inaccuracy dispute under § 1681i(a)(1).
The reasoning is doctrinal. Federal courts have held that the FCRA requires accuracy at the level of the individual data field, not just the existence of an account. A balance reported as $1,800 on Equifax and $1,950 on Experian is, by definition, not accurate on both bureaus simultaneously — one of them is wrong, even if the underlying debt is valid. The cross-bureau inconsistency is itself a basis for a dispute, and the dispute language can lean on the bureau's obligation to maintain accurate information across its reporting.
The Mixed-File Detector
Mixed files — where a credit bureau has confused two consumers and merged their data — are rare in absolute terms but devastating when they happen. The scan looks for patterns that suggest mixed-file errors: accounts that do not match the user's known credit history, addresses on file that the user does not recognize, employers the user has never worked for, variant spellings of the user's name attached to specific accounts, and Social Security numbers reported with single-digit variations.
Mixed-file disputes are some of the highest-impact disputes possible because removing a single misattributed account can move a credit score by 80 to 100 points. They are also some of the highest-stakes from a documentation standpoint — the bureaus typically require a driver's license, proof of address, and sometimes a sworn affidavit. The scan flags candidates so the user can collect the required documentation before drafting the dispute.
The Unverifiable Pipeline
If you have previously disputed an item and the bureau came back with a "verified" response, the scan flags it for a Method of Verification follow-up under § 1681i(a)(6)(B). This is the FCRA subsection that forces the bureau to disclose specifically how the verification was conducted — what documents were reviewed, who was contacted at the furnisher, what data points were confirmed.
Most consumers stop after the first "verified" response. The Method of Verification request is where many contested items actually move. Federal courts have repeatedly held that simply forwarding a dispute back to the furnisher and accepting their response does not meet the bureau's statutory investigation obligation. When the Method of Verification response comes back generic or incomplete, the item becomes vulnerable to a follow-up dispute citing the bureau's failure to substantiate the verification.
The scan tracks the dispute history per item across past report pulls, so a user who has been disputing for a while sees the Method of Verification candidates surfaced automatically rather than having to manually remember which items are at which stage of the process.
Confidence Scores
Every flagged item carries a confidence score reflecting how likely the AI thinks the item is a genuine error worth disputing. A re-aged debt where the date moved forward by three years between bureaus gets a very high confidence score because the pattern is unambiguous. A balance that differs by $20 between bureaus gets a much lower confidence score, because the difference could easily be the result of normal timing differences in how each bureau processes the same furnisher data.
The confidence score is visible to the user during the review step covered in Saturday's post. High-confidence items get drafted into dispute letters by default. Lower-confidence items are surfaced as options the user can choose to dispute or skip. The system does not auto-file disputes on items where the legal basis is weak — partly because that is bad strategy and partly because filing disputes against accurate, properly documented information would violate CROA.
What the Scan Does Not Do
The scan does not assess whether you actually owe a debt. It cannot tell from your credit report whether the underlying transaction occurred. What it can tell is whether the data being reported about the debt matches the data on file at the other bureaus, whether the date of first delinquency is past the reporting limit, whether the bureau has substantiated the verification when challenged, and whether the patterns match what an FCRA violation typically looks like.
The scan does not provide legal advice for unusual fact patterns. If you have a complex situation — identity theft with police reports, a bankruptcy that was discharged but is being reported incorrectly, a debt that is the subject of pending litigation — the scan flags the item but recommends review by an FCRA attorney before filing. There are cases where the right next step is not a dispute letter but a federal lawsuit under § 1681n or § 1681o.
Coming Tomorrow
Tomorrow's post covers Day 3 inside CreditRefresh: the custom dispute letter generation with real legal citations. Once the scan has identified and categorized the disputable items, the next step is drafting individual letters for each one — with the correct FCRA subsection cited, the specific factual basis explained, and the requested correction stated explicitly. The letters are what reaches the bureaus, and the quality of the writing at this step is what determines whether the dispute gets dismissed as generic or processed as a substantive challenge.
Three days until launch. Join the waitlist at creditrefresh.ai.
Results may vary. No specific outcome is guaranteed. CreditRefresh disputes inaccurate, unverifiable, or improperly reported information — not accurate items. This article is for informational purposes only and is not legal advice. For legal questions, consult an attorney.