AICENT.COM
← Back to Blog

A review round catches a real reading gap

2026-11-18 · RTTP · IQA
Reading the Draft · No. 27 · Sources cited in this piece · the text read sentence by sentence: draft-li-rttp-iqa-addressing-02 (full text) · series page and revision history: datatracker · the family under correction: the uri-grammar family page (sealed record) · the candidate record and review thread: lab PR #223 · Sister pillars: RTTP (rttp://, CRI 27) · IQA (iqa://, CRI 28)

A grammar is the part of a specification that looks boring and is not. Everything else in a protocol can survive a generous reading; a grammar cannot, because a parser has no generosity to give. The uri-grammar family is our case set for the grammar of the rttp URI — the address form of the Resonant Time Transfer Protocol (RTTP), one half of the two-scheme draft this series reads — and this piece is about what happened to it in a review round this week: two of its twenty-six expectations turned out to be stricter than the text they test, a reviewer said so in one paragraph, and the correction that followed is the real subject — because how a test suite reacts when it is stricter than its text is a discipline question, not a typing question.

§4.1 · one URI, two readingsthe hyphen is in the character set the published text permitsrttp://-abc.pillar.example/vesselaction = 1*( %x61-7A / DIGIT / "-" )one or more of: lowercase letters, digits, hyphen — no position rule-02 reading · acceptthe conformant reading of the published grammar-03 proposal · rejectreason: leading_hyphen — the planned tighteningboth readings scored consistent · -03 settles the grammar · a parser following -02 is not wrong
Figure 1: one URI, two readings. The published ABNF admits the leading hyphen; the rejection we expected was our planned -03 change wearing a -02 label. The fix gives each reading its own name.

What the text actually says

The published grammar gives the action slot a character set and a length rule, and nothing else. Section 4.1, verbatim: “The action, when present, is a single path segment: "/" 1*( %x61-7A / DIGIT / "-" ).” In the ABNF block the same rule appears as action = 1*( %x61-7A / DIGIT / "-" ) — one or more repetitions of lowercase letters, digits, and the hyphen. There is no position rule anywhere in the section: no sentence says the hyphen may not start a segment, and no sentence says it may not start the intent or the pillar either. The draft's own examples show hyphens in the middle of labels, which is the reader's natural image — but the rule as written draws a wider circle, and rttp://-abc.pillar.example/vessel stands inside it.

Our family, built days after the draft posted, expected that URI to be rejected. Two negative cases — UG-N15 for a leading hyphen in the intent segment, UG-N16 for one in the pillar segment — carried expect: reject with the reason leading_hyphen. The intention was honest and the timing was not: we plan to tighten the grammar in -03, and the cases were written as if the tightening were already published. A reviewer of the candidate record put the gap in one paragraph, verbatim: “Section 4.1 permits a leading hyphen, so UG-N15 and UG-N16 are valid under -02. Their expected rejections come from your planned change for -03, not the published grammar.” And the requirement that followed is the whole discipline in one line: “They can remain as proposed -03 cases, but a parser following -02 must not be scored as wrong.”

The fix: split the expectation, keep both readings

The correction does not weaken the cases and does not touch the draft; it gives each expectation the label it always should have had. The two cases now carry the status proposed-tightening, and the note pinned into the vector file states the arrangement verbatim: “UG-N15 and UG-N16 carry status proposed-tightening: the -02 ABNF literal admits a leading hyphen, so accept is the conformant -02 reading; a rejection naming leading_hyphen is the proposed -03 tightening (author decision 2026-11-09). Both readings are scored consistent until -03 settles the grammar.” The mutation battery changed with them: the mutant that enforced the tightening is retired — thirteen mutants became twelve — and the battery now proves its coverage over the twenty-four normative cases, twelve failures for twelve weakened rules. The recorded run across all six families re-ran clean, the family page re-sealed with the corrected ledger, and the reviewer's last line — “The other 24 cases look good. I'll approve once this is corrected.” — describes exactly what shipped the same day.

the corrected ledger · uri-grammar familyevery expectation now wears the label of the text it tests24 normative-02 expectations · all agreethe grammar as published2 proposed-tighteningUG-N15 / UG-N16 · dual readingboth readings consistent12 / 12 mutantsbattery re-run over the 24tightening mutant retired (13→12)why retire a mutant?a mutant enforces a rule of the checker; the tightening rule no longer judges -02 input —keeping it would score a conformant parser as broken, which is the exact bug we just fixedsix families re-recorded: 127/127 cases agree · 93/93 mutants killed · family page re-sealed (20261117a)
Figure 2: the corrected ledger. Twenty-four normative cases keep their -02 expectations; the two hyphen cases carry the proposed-tightening status with both readings scored consistent; the mutation battery shrinks to twelve and still dies twelve times — because a battery that judges text nobody wrote would be one more strictness bug.

Why this is the good outcome

The tempting fix was the other direction: keep the expectations, argue the reviewer's reading is narrow, quietly rely on the planned -03 to make the expectations retroactively true. That fix loses in every direction at once. It makes the suite stricter than the text it tests, which means every conformant parser scores as broken — the suite stops measuring conformance and starts measuring agreement with us. It makes the -03 process dishonest, because a change that is already enforced by a published test battery is not a proposal anymore. And it makes the review round worthless, since the one thing a review can catch that self-review cannot — the difference between what we wrote and what we meant — is exactly the thing that fix refuses to hear. The reviewer's paragraph cost two cases their label and saved the family its meaning; a stricter text is coming in -03 either way, and now the record shows precisely which strictness is ours to propose and which is the text's to demand.

One line: a grammar admits what its character set admits, not what its readers expect; a review round exists to catch the difference; the honest fix is to split the expectation — accept under -02, reject under the proposed -03, both consistent until -03 settles — because a test suite stricter than its text does not measure the text, it measures us: rttp resolves where & when; iqa states whether — and where the two agree, action begins.

SEALED rttp://42d8ea31.blog.aicent · iqa://42d8ea3107f9e5e0738e6be68d15a37a.forge.iqa · seal record