Markdown for LLMs

nb-22 solutions

The source Markdown for this article. Copy it into your assistant or download it as a text file.

# nb-22 solutions

## 1. Optional premise, then and now

`cause = some repeat_inquirer` in `RepeatFee`: the generated rule carries
`if some cause as g { when g(subject); }`, so without a `repeat_inquirer`
fact the premise is unsatisfied and the rule cannot fire — `callback_fee`
is NEITHER, exactly what `fee needs the repeat record` asserts. `cause =
none` in `FirstNotice`: the optional premise vanishes, so `inquiry_filed`
alone suffices and `callback_notice` is TRUE_ONLY. If you predicted NEITHER
for the fee, you read the option correctly; if you predicted FALSE_ONLY,
re-read nb-01 — an unsatisfied premise is silence, not a refusal.

## 2. Surcharge explanation steps

`DeskSurcharge/applicable` (the generated defeasible rule, named by the
`self/applicable` stable identifier), the `insp` case assertion
(`inspects_at`), and the `open` case assertion (`desk_open`). The waiver
defeater `DeskSurcharge/excluded/Waiver/excludes` does not appear: no
`waived_case` fact is present, so it never fires. A miss here usually
means naming the template (`inspection`) instead of the instance rule.

## 3. Two cheapest flips

Two solutions — `...#add-i6a` and `...#add-i6b` — each at cost `1`, with
`"cutoffCost":"1"` and two candidates checked. `--verify` over the
two-candidate result reports
`{"kind":"counterfactual-verification",...,"valid":true}` for result id
`...#FlipInquiryInput02/result`. If you predicted one solution, you
assumed the solver breaks ties; it does not — every cheapest patch is
listed.

## 4. Removed spellings

The two migration probes are now rejection cases. Use
`` `relation … kind institutional` `` for relations and `` `unless` `` for
exceptions. The earlier `LDC-E1328` warning followed by `check OK` is a
historical observation from engine `law 0.1.0`; neither spelling is accepted
by the current compiler.

## 5. Which edit moves the hash

Re-indenting and the metadata note leave `programHash` at
`sha256:cfe2fcb8…` — canonicalization ignores byte layout and
`metadata` is outside the semantic preimage. Only the `50` →
`51` literal moves it (`sha256:8b4127d7…`). The stale `--verify`
then fails with `COUNTERFACTUAL_PROGRAM_HASH_MISMATCH`, exit 1,
naming both hashes; a renewed input with the old result fails
the second layer instead (`COUNTERFACTUAL_CERTIFICATE_INVALID`).
Re-obtaining the certificate finds `add-i6` at cost 1 again and
verifies `valid:true`. If you predicted that any byte change
moves the hash, re-read [§4](/tutorials/northbridge/solutions/nb-04-solutions/#4-matchedcount18--value-and-what-does-it-say) — the hash is semantic, not textual.