Methodology
Methodology · Red team

Find a flaw. We canonize the fix and credit you.

Every error found in this architecture has made it stronger. The list below is public on purpose.

The Accord is red-teamed continuously — by AI systems in structured adversarial review, and by anyone who reads closely. The standing offer: identify a flaw not listed below — an arithmetic error, an inverted incentive, an overclaimed citation, a mechanism that doesn't survive its own worked example — and the fix enters canon with your name (or handle) on it. The replication package gives you every parameter and the engine that produces the headline numbers.

2026-07-03 · from the author
We re-ran our own conservative case honestly — and it got worse

Our conservative scenario carried a quiet flattery: a revenue multiplier that ran above the central case, and a padded healthcare basis doing double duty as the stress. We removed both. The honest conservative case now runs the same $5.9T healthcare basis as central with revenue genuinely stressed below it — and the Year-10 balance fell from $0.78T to $0.35T. The architecture holds anyway: the Debt Sunset Governor steps the payroll rate and top rate to the corridor ceiling and the debt still retires in 2079, inside the fifty-year window. That is the design working as designed — the rates are calibrated just-barely-enough on purpose, and the governor, not optimism, is the guarantee. A build-time check now fails any deploy where the conservative case is calibrated above central.

Flaws found and fixed

2026-07-03 · commit c6e6686
Canon and fiscal engine had diverged in opposite directions — the site asserted model behavior the model didn't have (prepayment rates, VAT path, healthcare basis).
Fixed: The engine now derives every instrument from the canonical config, and a build-time reconcile script fails any deploy where they drift.
2026-07-03 · commit 4bb4415
The disclosure-window incentive inverted: a fixed 0.8% Year-1 rate sat above the 0.50% bottom bracket, making prompt disclosure worse than hiding.
Fixed: The bottom bracket returned to 0.80% and the Year-1 concession became a 10% lifetime discount on the holder's own bracket payment — a discount off a schedule can never invert against it.
2026-07-03 · commit b9e861b
The site claimed broad WTO confidence for the whole tariff architecture.
Fixed: Scoped to what the precedent supports: carbon and methane border adjustments at domestic rates. The rest is negotiated, and the page says so.
2026-07-03 · commit 0a3b202
The Chapter-5 revenue pills included an 'SS carve' the rebuilt engine no longer had, and the pill decomposition didn't sum to the unified total.
Fixed: Pills are generated from the engine at build time and asserted additive to the dollar.
2026-07-03 · commit c6e6686
The conservative scenario quietly applied an upward revenue multiplier — a stress case that wasn't stressed.
Fixed: Conservative revenue now runs at or below central by build-time assertion; the governor, not a hidden multiplier, carries the stress.
2026-07-03 · commit bb04b035
The luxury VAT tier was a cliff: +5 points on the whole price once above a threshold — the design whose bunching pathology the UK spent until 2014 unwinding, and whose 1990 US cousin was repealed within three years.
Fixed: Redesigned as a marginal-excess supplement — 5% of the amount above per-category indexed floors. A cliff replaced by a slope; the 1990 failure designed out (destination-applied, indexed floors, credit-invoice chain).
2026-07-03 · commit 755b6fa
COMPASS triage allocated proportionally, diluting every shortage below effective dose in tight years.
Fixed: A waterline allocator funds the worst gaps to minimum effective dose first, and publishes what went unfunded.

How to submit

Send the flaw, the surface it appears on, and — if you have one — the fix, via the contact page. Confirmed flaws are corrected on the affected surfaces, logged in the updates log with the prior value and the correction, and added to the list above with credit.