Assess STATE against INTENT: seven gaps, one sharp failure

The 'a regulation that applies was found before it was needed' test is
not met and cannot be retro-fitted: the policy set found two obligations
already live, the e-invoicing one by about nineteen months. The catalogue
makes the test passable going forward; it does not make that instance a
pass.

Also: remediation tracking does not track — fix_tracking is a string
nobody reads, so an owner who goes quiet and a fix that goes quiet look
identical; there is no intake path for incidents or external reports, so
the register is a self-assessment aggregator rather than what INTENT
claims; and no coverage model, so a system with zero findings is
indistinguishable from one nobody assessed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

docs/assessments/2026-08-21-state-vs-intent.md

@@ -0,0 +1,152 @@
+---
+id: RISK-ASSESS-0001
+type: assessment
+title: "STATE.md against INTENT.md — what is missing"
+date: "2026-08-21"
+owner: risk-nexus
+---
+
+# What this repo promised, and what it has
+
+`INTENT.md` was written before the register existed. Four days in, this is what
+it claimed, what is true, and what is not.
+
+## What it owns — six claims
+
+| Claim | State | Honest reading |
+| --- | --- | --- |
+| Finding intake, from any source: research, review, **incident, external report** | Partial | Nine findings, all from research or self-review. **No incident path and no external-report channel exist at all.** |
+| Severity and triage | Yes | Scale adopted, amended twice under use, applied to nine findings with no `unset` left. |
+| Remediation tracking — owner, state, whether it stalled | **Nominal** | `fix_tracking` holds strings nobody reads. See gap 2. |
+| Disclosure timing | Yes | Three states, exercised, and it has bitten across two findings. |
+| Regulatory intake | Yes, and beyond | One determination, thirteen-entry policy set, a written route. |
+| The escalation rule | Yes | Adopted 2026-08-20 with the operator's thresholds. |
+
+## What it does not own — held
+
+Fixing, publication, blocking delivery, legal advice, amending canon. Nothing
+in four days crossed those. The nearest approach was running `kubectl` against
+another repo's cluster, and that is bounded in `docs/method/verification.md`:
+read-only, recorded, routed as a question, never a conclusion on the owner's
+behalf.
+
+## What good looks like — four tests from INTENT
+
+**"A gap that was known is recorded with a severity and an owner."** Met.
+Nine, including the estate-wide one that had lived as open question 3 of an
+unratified draft.
+
+**"A regulation that applies was found before it was needed."** **Not met, and
+this is the sharpest failure in the assessment.** The policy set found two
+obligations that are *already live*: commercial and tax retention, and the
+e-invoicing receiving duty in force since January 2025. Neither was found
+before it was needed. They were found roughly nineteen months late, by writing
+a catalogue that should have existed first. The catalogue now makes the test
+passable going forward; it does not make this instance a pass.
+
+**"A fix that stalled was visible while it stalled."** **Untested, and weaker
+than it looks.** No fix has stalled yet. The mechanism — fourteen days at the
+bottom rung — measures whether the *register* is checking, not whether the
+*fix* is moving. See gap 2.
+
+**"The things that genuinely needed the operator's judgement reached them."**
+Met, four times, three answered. One was withdrawn before sending because the
+fix landed first, which the register recorded as luck rather than a success.
+
+## The gaps, in the order they would hurt
+
+### 1. Nothing can report to this register from outside
+
+Every finding arrived from a repo reading its own code, or from this register
+reading a cluster. `INTENT.md` claims intake from **incident and external
+report**; neither has a path.
+
+There is no address an outsider could use, no `security.txt` anywhere in the
+estate, and no defined route from "something is happening right now" to a
+record. The severity scale assumes a latent defect; an active incident has a
+different tempo, and `RISK-POL-0005` (breach notification, 72 hours) is dormant
+with nothing wired to it.
+
+The register is currently a **self-assessment aggregator**. That is a
+respectable thing to be and it is not what the intent says it is.
+
+### 2. Remediation tracking does not track
+
+`fix_tracking: FLEX-WP-0015-T02` is a string this repo writes down and never
+reads. The hub knows whether that task moved; the register asks a human
+instead, and "has the fix moved?" is question three of every check.
+
+That decays the moment nobody is looking, which is precisely the condition the
+whole cadence design assumes. **The fix state should be read from the hub, and
+a fix that has not moved should reset nothing and escalate on its own timer** —
+independent of whether the register happened to check.
+
+Today, an owner who goes quiet and a fix that goes quiet look identical.
+
+### 3. There is no coverage model
+
+The register knows what was reported. It has no view of what was never looked
+at.
+
+A system with zero findings is indistinguishable from a system nobody has
+assessed, and `RISK-N-0003` records the uncomfortable evidence: every repo that
+has examined its own boundary this month found a defect — two for two.
+`INTENT.md`'s "nobody is surprised" cannot hold while the denominator is
+unknown.
+
+### 4. The production transition has no owner and no definition
+
+Five findings owe a re-score at "the production transition", and
+`RISK-F-0007`'s acceptance **expires** there. Nothing outside this repo knows
+that, no one has defined what the event is, and no one has been named to
+declare it.
+
+An acceptance that expires on an undefined event expires on nobody's calendar.
+
+### 5. The repo claims a surface it does not have
+
+`README.md` and `INTENT.md` both say `risk-nexus` serves `risk.coulomb.social`.
+It serves nothing. Two documents wait on `policy-nexus` for an address, and the
+site does not exist.
+
+Small, but it is exactly the class of claim this register grades other repos
+down for: a stated control that is not there.
+
+### 6. Escalation has no delivery guarantee
+
+Escalations go to an inbox and are said aloud in whatever session is running.
+If the operator does not read the inbox, the escalation is indistinguishable
+from one never sent — which is the failure this register committed on
+2026-08-19 and fixed *for itself* with `hourly-register-inbox-watch`.
+
+The same fix has not been applied to the escalation path. Nothing acknowledges,
+and nothing re-raises.
+
+### 7. Everything runs from one checkout
+
+The activity definitions instruct a session that may not exist. If no agent
+picks up the daily sweep, `make due` still reports and nobody reads it. The
+ladder degrades to decoration silently — and a `1q` rung would then mean "no
+one checked for a quarter", which reads identically to "stable for a quarter".
+
+`clean_streak` makes a suspiciously smooth climb visible, but only to someone
+looking.
+
+## What is deliberately not a gap
+
+- **No monitoring.** `RISK-N-0003` is a note, ruled so on purpose: no repo owns
+  estate-wide detection and a register that grows probes becomes a second
+  engineering team.
+- **No OpenBao access.** Bounded in `verification.md`. Acquiring production
+  secret-store credentials for a risk register would trade a verification
+  problem for a worse one.
+- **No blocking.** Build mode: a finding is information. Unchanged.
+
+## Ranking
+
+If one thing is done next it is **gap 2**, because it is the only one where the
+register currently reports something it cannot actually see, and INTENT names
+it explicitly. Gaps 1 and 4 are the next two, and both are cheap.
+
+Gap 3 is the largest and the most expensive, and it is the one that decides
+whether "nobody is surprised" is a claim this repo can ever make.