# Four failure modes and their DevRites controls

`rite-frame` checks four common LLM failure modes. Existing DevRites controls address
each one. Use these names for quick checks on ad hoc or express-lane work that does not
reach the normal gates.

## The map

| # | Failure mode | The tell in a diff / a reply | Cure already in the pack |
|---|---|---|---|
| 1 | **Silent assumption** | A value, contract, edge case, or interpretation was *guessed* and run with: no note, no question. The ask had two readings and one was picked silently. | [`core.md`](../../devrites-lib/reference/standards/core.md) #2 (no silent assumptions) / #3 (no guessing through confusion); the Spec Drift Guard; `devrites-doubt` for a single high-stakes call; `devrites-interview` when the ask is underspecified. |
| 2 | **Overcomplication** | An abstraction, flag, config knob, or indirection nobody asked for. 200 lines where 50 would do. A single-use factory. Defensive `try/catch` + null checks inside trusted code. A dependency added where an in-repo option exists. | [`coding-style.md`](../../devrites-lib/reference/standards/coding-style.md) (simplicity, reuse-first), [`patterns.md`](../../devrites-lib/reference/standards/patterns.md) (avoid over-engineering), `devrites-audit simplify` (deletion test, Chesterton's Fence), the "over-defensive guarding is slop" anti-pattern. |
| 3 | **Out-of-scope edit** | Touched code, comments, or formatting outside the ask. A "while I'm here" refactor. Renamed something orthogonal. Reflowed an import block. | [`core.md`](../../devrites-lib/reference/standards/core.md) #7 (feature scope only), the exact paths stated in the writer task, `git diff --name-only` plus `touched-files.md`, and the "it's only a small refactor" anti-pattern. |
| 4 | **Unverifiable goal** | "It works" with no command, no output, no test. A success criterion that can't be false ("make it better"). A tautological test that passes no matter what. | [`acceptance-criteria.md`](../../rite-spec/reference/acceptance-criteria.md) (measurable acceptance), [`core.md`](../../devrites-lib/reference/standards/core.md) #6 (evidence over confidence), [`testing.md`](../../devrites-lib/reference/standards/testing.md) (assertion strength, see it fail first), the TDD wright. |

Karpathy named the first three. The fourth supplies the falsifiable criterion needed
for an agent to work without repeated human confirmation. FRAME defines that criterion
before auditing the other three modes.

## FRAME: rewrite an imperative as a falsifiable condition

Rewrite the request as a condition that can be false. If that is impossible, the
request is ambiguous and triggers mode 1 before any diff exists.

| Imperative ask | Weak (still a wish) | Falsifiable criterion + verify |
|---|---|---|
| "Add validation" | "validate the inputs" | "`{empty, oversize, wrong-type}` → 4xx + message; a test asserts each and is red today." → `npm test path/to.spec` |
| "Fix the bug" | "make the bug go away" | "Test reproducing the report is red now, green after, nothing else changes." → the repro test |
| "Make it faster" | "improve performance" | "Endpoint X p95 `<baseline ms>` → `<target ms>` on `<named bench>`." → the benchmark cmd |
| "Refactor X" | "clean up X" | "Existing suite green before and after; behavior byte-identical." → full suite, twice |
| "Handle errors" | "add error handling" | "On `<failure F>` the system fails closed with `<observable>`; a test forces F." → that test |
| "Improve the UX" | "make it nicer" | *(no falsifiable check → ambiguous → `devrites-ux-shape` / ask which states/flows)* |

A criterion is ready when a reviewer can run the verification command and get a clear
pass or fail without asking what it means.

## AUDIT: worked finding lines

Findings read like the rest of the pack: one line, severity-tagged, cite `file:line`, route
to the cure. Severity ladder is the pack standard (`Critical / Important / Suggestion / Nit /
FYI`).

```
auth.ts:42  Important  mode 1 (assumption): assumes `role` is always present — unset on legacy rows. → validate at boundary or ask (core #2).
cart.ts:88  Suggestion mode 2 (overcomplication): single-use `StrategyFactory` for one caller. → inline; deletion test fails it (devrites-audit simplify).
utils.ts:5  FYI        mode 3 (scope): reflowed an unrelated import block. → revert to boundary; not in this ask (core #7).
sum.spec:12 Important  mode 4 (unverifiable): test asserts `mock.called`, not the result — passes if the fn is empty. → assert the value; see it fail first (testing.md).
```

Record clean modes explicitly (`1 assumption: clean`) so an empty result is not mistaken
for a skipped check.

## Related tools

- **`devrites-doubt`:** adversarial *fresh-context subagent* pre-mortem on **one** decision.
  It is independent and avoids anchoring. rite-frame is self-applied and covers all four
  modes; use doubt for one consequential claim.
- **`devrites-audit <axis>`:** *fresh-context subagent* review of an **active feature's** diff
  on one axis (security / perf / simplify). Needs a `.devrites/` workspace. rite-frame needs no
  workspace and runs in the current context, with less independence.
- **`/rite-review` · `/rite-seal`:** the **gates**: parallel reviewer fan-out, blocking
  severities, written verdict. rite-frame is the inline reflex for work that never reaches a
  gate and does not replace one.

Use rite-frame for the quickest self-check. Use doubt, audit, or the full gates when the
work needs independent or feature-level review.