AMD-0003 — decision rationale

OUTCOME: VETOED, 2026-08-03, by the ratifying authority. Reason given: ledger fidelity. XII.1 prevents editing the ledger, and an approval state recording an amendment before enactment would create a historical claim that is not true. Preserve the record first; repair the mechanism through a compliant amendment path.

Permanently abandoned per Article X. The analysis below is preserved as written — it is the reasoning the decision was taken against, and editing it to match the outcome would destroy exactly the auditability the veto was chosen to protect.

This is a design argument, not a proof. The uniqueness theorem that previously supported this recommendation was refuted (veto-necessity.md §6). Both APPROVE and VETO admit progressing executions. What follows is an engineering preference with its criteria stated so it can be disagreed with.

---

Evaluation criteria

1. Ledger fidelity — does the record accurately represent what governs?
2. Operational cost — how many decision cycles are required?
3. Reversibility — how hard is the choice to undo if it proves wrong?
4. Third-party auditability — can someone who was not present reconstruct what happened and what is in force?
5. Precedent — what does this establish for future amendments?
6. Residual defect risk — how likely is a known defect to survive into the enacted state?

The two paths

APPROVE — approve AMD-0003; propose AMD-0004 supplying determinate Article XIII text; decide; enact. AMD-0003 remains approved and never enacted.

VETO — veto AMD-0003; write the two missing probes; re-propose with determinate text, probe coverage, and the en-bloc clause; decide; enact.

Assessment

Criterion	APPROVE	VETO
Ledger fidelity	poor — an entry marked approved that never took effect	good — vetoed is unambiguous under X.1.3
Operational cost	2 decision cycles	2 decision cycles
Reversibility	poor — approval is permanent under XII.1; only supersedable	poor — X.1.3 makes abandonment permanent
Auditability	poor — determining what governs requires knowing AMD-0003 was never enacted, a fact recorded nowhere	good — exactly one entry governs
Precedent	approving a known-defective amendment is acceptable if patched later	defective amendments are withdrawn, not patched
Residual defect risk	moderate — the VII.1 probe gap survives unless AMD-0004 addresses it	low — closing VII.1 is a precondition of re-proposal

Two corrections to earlier claims

Operational cost is a wash. I previously said VETO "costs one re-proposal" against APPROVE's "extra amendment cycle". Both require exactly two decisions: APPROVE → decide AMD-0004, or VETO → decide the replacement. The cost argument for VETO was wrong and is withdrawn.

Reversibility favours neither. X.1.3 makes a vetoed proposal permanently abandoned; XII.1 makes an approval permanent and supersedable only by a new entry. Both directions are one-way.

The strongest argument against VETO

X.1.3: "Re-proposing the same amendment requires materially new runtime evidence, and must say what changed." A replacement for AMD-0003 is substantially the same amendment, so it must clear that bar — and if the bar were judged unmet, Article XIII would be permanently blocked.

This is dischargeable, and discharging it is itself valuable. The two absent probes for SUBJECT_KINDS and FACT_KINDS do not exist yet; writing them produces genuinely new runtime evidence and closes the VII.1 gap that made AMD-0003 defective. Precedent supports this reading of the bar: AMD-0002's evidenceProbe cites GET /api/stress alongside a grep of db/schema.sql, so inspection-derived evidence has previously counted.

Sequenced correctly, the constraint stops being a risk and becomes a forcing function: VETO cannot be executed sloppily, because the re-proposal is gated on work that fixes a known defect.

---

Recommendation

VETO CONSTITUTION CHANGE, on two criteria only — ledger fidelity and third-party auditability.

Everything else is a tie. Cost is identical, reversibility is symmetric, and precedent is a soft consideration. The decisive point is narrow:

Article XII exists so the constitution is "auditable rather than merely asserted." An entry marked approved for an amendment that never took effect makes the ledger assert something untrue about what governs — and it does so permanently, since XII.1 forbids editing it away.

That is a small consideration, honestly. It is not a theorem and does not pretend to be. Anyone weighting "avoid re-litigating a decision already made" above ledger cleanliness could reasonably choose APPROVE, and the resulting system would be correct — just harder to read six months from now.

What is not in dispute

Whichever path is taken, the repair queue is unchanged and independent of it: the X/XII deadlock, the XII.3 operative-text requirement, XII.1 ledger repair for the two in-place edits, SUBJECT_KINDS/FACT_KINDS probe coverage, and attestation.