Editorial policy
What decides a number here, what only describes one, and what has to be true before either ships.
What decides, and what only explains?
The engine decides; the prose explains. Every figure you see is produced by deterministic code with zero dependencies, no network access and no AI in the calculation path — money is held in integer cents and interest rates in basis points, so 360 months of arithmetic cannot drift. The writing around it can describe a result, warn about it or link to the rule behind it, but it can never adjust it. When prose and engine disagree, the engine is what ran.
Where do the rules come from?
From the regulation itself, read in full. The governing document for every plan on this site is the RISE final rule, 91 Fed. Reg. 23768, published 1 May 2026 as FR Doc. 2026-08556 and effective 1 Jul 2026, amending 34 C.F.R. parts 674, 682 and 685. The poverty guidelines come from the HHS table published each January; the tax treatment of forgiveness from 26 U.S.C. § 108(f).
Where a citation points somewhere unexpected, that is deliberate. Both eCFR and the Federal Register's own site refuse automated requests and redirect every path to a block page — which means a response from either host proves nothing, since a real URL and a mistyped one behave identically. The rule text was therefore retrieved from govinfo.gov, which serves the identical GPO text, and that is the URL the rule files cite. Citing the document that was actually read matters more than citing the prettier link.
What closes a verification gap?
A primary source, and nothing else. A secondary source — a law-school mirror, a research summary, a commercial code site — may cross-check a value, but it may never close a gap; any value resting on one is labelled as such. Where a source could not be reached, the engine keeps whatever it already had and the gap is recorded openly. No row is filled with a plausible-looking substitute, a recalled figure, or an estimate.
Closing a gap means editing the code and the register entry together. Deleting an open item without a fetched URL and a date is not closing it, and a citation that has never been opened is treated as an open item even when the figure it would support is verified elsewhere. Three citations on this site are in exactly that state, and the about page names them.
One further rule comes from experience rather than principle: figures are read from extracted statute or PDF text, never from a summarising layer. During the same verification programme, a summarising fetch tool misnumbered a revenue procedure and returned an entirely fabricated table for another engine on this site. The encoded values were already correct; trusting the summary would have replaced a right answer with a wrong one.
What review gate has not been met?
Credentialed review. Before public launch, the encoded rules and the guidance on these pages must be reviewed by a named, credentialed student loan professional — a CSLP or an attorney — and their name and credentials must appear on the about page and on every guide they signed off. No reviewer, no launch.
No reviewer has been engaged, and no review has taken place.
Nobody is named on this page because there is nobody to name. Verification so far has confirmed that encoded values match their cited sources — a narrower question than whether the output is legally or financially sufficient for any particular borrower.
How are errors corrected, and how fast?
Within 48 hours of confirming one. The sequence is fixed: correct the rule file, re-run the full test suite, then publish a dated changelog entry stating what changed, which primary source it was changed against, and which results moved. Because every page renders from the same rule files, a single edit updates the tool, the methodology page and the sources table together — there is no second copy of a number to forget.
Report an error to info@bracketsight.com. Include the inputs you used if you can; a scenario link carries them. We will credit the report unless you ask otherwise. A report we cannot reproduce still gets an answer.
How are rule changes versioned?
Every rule set is a separate dated file with its own version string, its own effectiveFrom and effectiveTo, and at least one citation carrying a URL and the date it was last verified. The engine resolves which file applies from the simulation date, so an old scenario keeps computing under the rules that were in force when it ran.
A rule set that changes gets a new dated file rather than an edit in place. That is why the annual poverty guidelines and the plan terms carry different date shapes: the guidelines are republished every January and the plan terms took effect with the RISE rule on 1 Jul 2026. The changelog lists the five current version strings, and the sources page renders straight from the files, so it cannot drift from what the engine computed.
Where AI is and is not used
No AI computes, estimates, or adjusts any number on this site — the calculation engine is deterministic, dependency-free code with its purity enforced in continuous integration. Planned AI features (reading an uploaded statement into the form, explaining your results in plain language) always place a human review screen between the AI and the engine, and never publish a number the engine did not produce.
Who pays, and can they reach a number?
No lender, servicer, or debt-relief company pays us, and none can influence the ranking, because the ranking is arithmetic. The structural guarantee is stronger than the promise: the engine has no dependencies and makes no network calls, so there is no place inside a calculation for a partner's code or data to sit.
No advertising runs on this site today and no ad network script is loaded on any page. Planned revenue is advertising on content pages — never inside the tool, where an ad could be mistaken for part of the result — an optional paid PDF report, and clearly labelled affiliate links shown only when the engine determines they serve you, always with any forfeited benefit disclosed at equal visual weight. The site privacy policy states what changes on the day advertising is switched on.
Rules verified · 34 C.F.R. § 685.209 · How every plan is simulated →
This is an independent decision engine, not your loan servicer and not financial or tax advice. Estimates run under current federal rules — confirm any plan change with your servicer before acting. Your loan data never leaves your browser.
Pre-launch build. Three things here are not fully settled and are worth knowing before you act on a ranking. The tax on a forgiven balance is estimated at a flat assumed marginal rate, which is a modelling choice and not a regulatory figure — it is labelled as an assumption wherever it is used. The eligibility test for PAYE models only part of the two-part new-borrower rule, so the engine can show PAYE to a borrower a servicer would turn down. And the Tiered Standard step interval is a servicer convention rather than a term the regulation prescribes. Each is set out on methodology.