Skip to content

Commit 4ecafc7

Browse files
Re-derive the tenant-audit write-call-site census as an in-tree artifact (page + generator + drift gate) (#13584)
* Re-derive the tenant-audit write-call-site census as an in-tree artifact The measurement that decides this repair family's severity lived in a comment on issue #13178. That issue became unreachable — 404 on unauthenticated REST, on the rendered page and on authenticated MCP alike, while its neighbours answer 200 — and took the census with it while three open cards still named it as their input. What survived did so by luck: a changeset author had quoted two figures in prose. Adds the isSystem census triple's shape for the tenant-audit control: scripts/tenant-audit-census.mjs the instrument content/docs/permissions/...mdx the page (prose + generated figures) docs/audits/...counts.md the per-site ledger, regenerated whole scripts/check-tenant-audit-census.mjs the gate It is a DRIFT gate, deliberately not an anchor gate: rows aggregate by (file, verb, object, tenancy, context posture) and carry no line numbers, so a pure displacement cannot move them. That skips the sibling's open defect class — anchors that rot on displacement and a repair arm that then misreads displacement as a population change. Measured: 214 write call sites, 143 statically decidable, 41 tenancy-enabled carrying no tenant context. The deviations from all four carried figures are recorded on the page untrimmed, including that the 135/77% "silenced by the isSystem guard" figure has no surviving corroboration and is not reproduced. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC * Regenerate the tenant-audit census against the merged tree `packages/plugins/plugin-security/src/platform-admin-service.ts` landed on main between this branch's census run and CI's, so the committed artifacts named 533 scanned sources where the tree holds 534 — and the drift gate caught its own artifact within the hour of it being written. Population unchanged: the new source contains no engine write call site (no insert/update/delete call syntax at all; the census places 0 sites and 0 unresolved receivers in it). 214 / 143 / 41 all hold, so the page's "cite 41 / 214" redirection needs no edit. Only `Sources scanned` moves, in both artifacts, which is exactly the shape "+1 scanned source, +0 population" should take. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC * List the census page in the permissions module index `check-section-landing-index` holds each section's "What's in this module" block to its meta.json in both directions, and the merge brought that gate into this card's derived family. Adding the page to meta.json without a row here left the reader's map short by one. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC * wip: elevation classifier fix + self-test * Correct two over-claims in the census's own classifiers The drift gate fired on this page's artifact twice in an hour, and tracing the second one found the numbers themselves were wrong in two places. Both are the shape this artifact exists to stop — a thing the walker could not read, scored as a thing it had read — wearing opposite hats. ELEVATION read a spread as a denial. It walked an object literal's named properties for `isSystem` and returned "not elevated" when the loop ended, so `{ ...SYSTEM_CTX }` scored decidably NOT elevated while every SYSTEM_CTX in the tree is `{ isSystem: true }`. It never unwrapped `as const` either, so a bare `context: SYSTEM_CTX` read as undecidable. 51 verdicts move: 45 undecidable -> elevated, 6 inverted outright. Decidably-not-elevated is now 0. TENANT CONTEXT counted sites it had not read. An unreadable options argument was folded in with one read and found empty, publishing "84 sites carrying no tenant context at all" when 17 said so and 67 were unread — an over-claim in the alarming direction, on the figure the page tells other cards to cite. `carries` is now three-valued; the provable count is 17 (9 tenancy-enabled), with 67 (32) unreadable. The generator gains its own `--self-test` (18 cases, 6 red against the old reading) and CI runs it: these classifiers are published FIGURES, not findings, so the gate's verdict never covered them and a clean tree could not tell a working classifier from a weakened one. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC * Route the census step's two self-tests through a collector `check-step-collectors` caught it: two independent self-tests as a bare sequence in one `run:` block means the second never runs under `bash -e` once the first fails — neither green nor red (#10814). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC * wip: unpromote generator * Keep the census generator a followable module, not a gate file `dispatch-gates --self-test` went red on this branch, by name: ✗ promoting 11 module(s) to gate files subtracts no inherited hint from any other family — LOST: scripts/check-tenant-audit-census.mjs <- scripts/tenant-audit-census.mjs (packages/services, packages/plugins, …) Wiring `node scripts/tenant-audit-census.mjs --self-test` into lint.yml made the generator a GATE FILE, and `discoverFamilies` refuses to follow one. The gate spelled ZERO path literals of its own and inherited all 22 from that module, so the promotion cut every one — a PR touching `packages/services/**`, the exact population this census counts, would silently stop being told this gate reads its diff. A lead that stops appearing is indistinguishable from one never earned. Fixed at the wiring rather than by weakening the assertion. The generator drops its `--self-test` FLAG and keeps its `selfTest()`, which the gate's own `--self-test` now drives — so the 18 classifier cases still run on every CI pass while the module stays followable. lint.yml goes back to two commands, so the collector this step briefly needed is no longer one. Belt and braces: the gate now carries a `ROOT_DIR_WATCH_HINTS` provenance declaration naming the two corpus subtrees and both artefacts, so the next attempt to invoke the generator directly cannot silently cut discoverability again. Family hints: 0 -> 24. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC * Split the tenant-audit census gate into enforced population and unenforced corpus scale Maintainer ruling 2026-08-31 (option A): the census gate adopts the enforced/unenforced split the sibling `check-system-context-census.mjs` already proved. Population -- every write call site and its tenancy/context verdict -- stays enforced byte-for-byte. Corpus scale (sources scanned and its kin) becomes present, dated, and not compared. The sibling's mechanism does NOT transplant: it splits a list of regexes over a hand-written page, while this gate's enforced half is a byte comparison of generated text with no per-number comparison to drop. So the scale values are masked out of both sides of that comparison, and required to be present and dated instead. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC --------- Co-authored-by: Claude <noreply@anthropic.com>
1 parent 4cc99bc commit 4ecafc7

8 files changed

Lines changed: 2699 additions & 0 deletions

File tree

‎.claude/workflows/docs-accuracy-audit.js‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -166,6 +166,7 @@ const ALL_HANDWRITTEN = [
166166
"content/docs/permissions/sharing-rules.mdx",
167167
"content/docs/permissions/sso.mdx",
168168
"content/docs/permissions/system-context.mdx",
169+
"content/docs/permissions/tenant-audit-census.mdx",
169170
"content/docs/plugins/adding-a-metadata-type.mdx",
170171
"content/docs/plugins/anatomy.mdx",
171172
"content/docs/plugins/development.mdx",

‎.github/workflows/lint.yml‎

Lines changed: 52 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1402,6 +1402,58 @@ jobs:
14021402
node scripts/check-system-context-census.mjs --self-test
14031403
node scripts/check-system-context-census.mjs
14041404
1405+
# #13548 tenant-audit census. `content/docs/permissions/tenant-audit-census.mdx`
1406+
# carries the population `SqlDriver.auditMissingTenant` acts on: every
1407+
# application-surface write call site against a tenancy-enabled object, and
1408+
# what execution context each one threads.
1409+
#
1410+
# ⭐ The reason it is an in-tree artefact at all is the failure it replaces.
1411+
# That census used to live in a COMMENT on an issue; the issue became
1412+
# unreachable (404 on three independent channels while its neighbours
1413+
# answer 200) and took the list with it, while three open cards still named
1414+
# it as their input and nothing noticed. What survived did so because a
1415+
# changeset author had happened to quote two of the figures in prose. A
1416+
# measurement that decides a repair family's severity and a ruling's scope
1417+
# is not a comment.
1418+
#
1419+
# ⚠️ It is a DRIFT gate, deliberately NOT an anchor gate like its `isSystem`
1420+
# sibling above. The generated region aggregates by (file, verb, object,
1421+
# tenancy, context posture) and carries no line numbers, so a pure
1422+
# displacement — an import inserted above a site — cannot move it. That
1423+
# skips the sibling's whole open defect class: anchors that rot on
1424+
# displacement, and a repair arm that then misreads displacement as a
1425+
# population change and asserts it on a security-relevant surface. There is
1426+
# exactly one repair path here (`node scripts/tenant-audit-census.mjs
1427+
# --write`), it is mechanical, and it has no case where it must guess.
1428+
#
1429+
# It lives in THIS job, next to its sibling, for the same reason that one
1430+
# does: the job has no paths filter, so the census runs on code-only PRs —
1431+
# the ones that ADD a write call site. The `--self-test` invocation is what
1432+
# `check:self-test-wired` requires, and it is the only instrument on this
1433+
# gate's matching rules; a clean tree cannot tell a working rule from a
1434+
# weakened one.
1435+
#
1436+
# ⭐ THREE commands, not two. The GENERATOR carries a self-test of its own
1437+
# because its classifiers are published FIGURES rather than findings: the
1438+
# gate's verdict is "no unplaceable receiver", and the elevation and
1439+
# tenant-context verdicts are not part of it. Nothing else reads them, so a
1440+
# clean tree cannot tell a working classifier from a weakened one. Both of
1441+
# the shapes it pins were measured WRONG on the first edition — a spread
1442+
# read as a denial of the flag it might carry, and an unreadable options
1443+
# argument folded into "carries no tenant context" — and six of its
1444+
# eighteen cases red against that reading.
1445+
#
1446+
# Invoked as `node` rather than through a `pnpm check:*` alias: see the
1447+
# GATE INVOCATION IDIOM note at the top of this file. AST walk over every
1448+
# tracked non-test source under the two roots plus the object registry, no
1449+
# spawns; ~8s. ⛔ The source COUNT is deliberately not quoted here: it is
1450+
# corpus scale, it moves on any file added under those roots, and a comment
1451+
# nothing checks is exactly where a stale number hides.
1452+
- name: Tenant-audit census matches the tree
1453+
run: |
1454+
node scripts/check-tenant-audit-census.mjs --self-test
1455+
node scripts/check-tenant-audit-census.mjs
1456+
14051457
# #12236 one <h1> per doc page. `DocsTitle` renders the frontmatter `title`
14061458
# as the page's <h1> unconditionally (apps/docs/app/[lang]/docs/[[...slug]]/
14071459
# page.tsx), so a body-level `# ` heading compiles to a SECOND one inside

‎content/docs/permissions/index.mdx‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -67,6 +67,7 @@ alike — is [Declaring Capabilities](/docs/permissions/capabilities).
6767
- [Social & Enterprise SSO](/docs/permissions/sso)
6868
- [Authorization Architecture](/docs/permissions/authorization)
6969
- [System Context (`isSystem`)](/docs/permissions/system-context) — what the platform's one elevation flag short-circuits, and what it deliberately does not
70+
- [Tenant-Audit Census](/docs/permissions/tenant-audit-census) — every application-surface write call site against a tenancy-enabled object, re-derived from the tree rather than quoted
7071
- [Profiles (removed)](/docs/permissions/profiles)
7172
- [Permission Sets](/docs/permissions/permission-sets)
7273
- [Positions](/docs/permissions/positions)

‎content/docs/permissions/meta.json‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -6,6 +6,7 @@
66
"sso",
77
"authorization",
88
"system-context",
9+
"tenant-audit-census",
910
"profiles",
1011
"permission-sets",
1112
"positions",
Lines changed: 236 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,236 @@
1+
---
2+
title: Tenant-Audit Census
3+
description: The authoritative enumeration of every application-surface write call site against a tenancy-enabled object — how many thread an execution context, how many thread none, and how much of the population a static instrument can decide at all. Built by census over the whole repo, not by recall.
4+
---
5+
6+
`SqlDriver.auditMissingTenant()` is the platform's one *diagnostic* on write-side
7+
tenancy. It warns when a write lands on a tenancy-enabled object without
8+
`options.tenantId` — the shape of a seed, migration or background path that
9+
forgot to thread the acting session.
10+
11+
This page is **the authority** for the population that control acts on: every
12+
write call site on the application surface, what object it targets, and what
13+
execution context it threads. The figures below are generated, not written —
14+
rebuilt from the tree by `scripts/tenant-audit-census.mjs`, and
15+
`scripts/check-tenant-audit-census.mjs` fails the build when what is committed
16+
and what the tree says disagree.
17+
18+
The per-site rows live next door, in
19+
`docs/audits/2026-08-tenant-audit-write-call-sites.counts.md`, and are written by
20+
the same command. The split is the one `packages/spec`'s strictness ledger
21+
already uses and is there for the same reason: this page has prose worth
22+
preserving and the row table has none, so the table is regenerated whole while
23+
the page keeps a small generated region for the figures its prose reasons about.
24+
25+
## ⚠️ Why this page exists at all
26+
27+
The measurement this page carries used to live in **a comment on an issue**. That
28+
issue became unreachable — it returns 404 on unauthenticated REST, on the rendered
29+
web page and on authenticated API access alike, while its immediate neighbours
30+
answer 200 — and it took the census with it. Three open cards named that comment
31+
as their input, and nothing anywhere noticed that their input had gone.
32+
33+
What survived did so by luck. A changeset author had happened to quote two of the
34+
figures in prose, so "175 write call sites, 24 of them carrying no tenant context"
35+
is still readable on `main` in
36+
`.changeset/tenant-audit-update-delete-half-repairs.md` — while the list of 24
37+
that a downstream card was briefed to classify is not recoverable at all.
38+
39+
⇒ A census that decides a repair family's severity and a ruling's scope is not a
40+
comment. It is a re-runnable instrument plus a committed page. Losing any issue
41+
now costs nothing, and the population is **re-derivable** rather than quoted.
42+
43+
## What the control does, and what a call site controls
44+
45+
`auditMissingTenant(object, op, options)` is gated in this order:
46+
47+
1. `OS_TENANT_AUDIT=0` — off entirely;
48+
2. `options.bypassTenantAudit` — the engine sets this for **every**
49+
`ExecutionContext.isSystem` write (ObjectQL's `buildDriverOptions`);
50+
3. a present `options.tenantId` — the engine fills it from `execCtx.tenantId`;
51+
4. the deployment posture — the warning is only meaningful on a walled install;
52+
5. the object having a tenant field.
53+
54+
Both of the guards a *caller* can move are downstream of one decision at the call
55+
site: **whether it threads an execution context at all**. Thread an elevated one
56+
and guard 2 silences the warning; thread a tenant-bearing one and guard 3 does;
57+
thread none and the site is what this census counts.
58+
59+
## How the population is decided
60+
61+
A site is a call to one of the three `IDataEngine` write doors — `insert`,
62+
`update`, `delete` — on a receiver whose declared type is an engine, in tracked
63+
non-test sources under `packages/services/` and `packages/plugins/`.
64+
65+
**The verb name is not the signal.** `.delete()` alone answers some 250 call sites
66+
in this corpus, and most of them are `Map.delete`, `Set.delete`, `Headers.delete`,
67+
a crypto `Hash.update`, a blob-storage delete-by-key and a search-index de-index.
68+
So the receiver is typed structurally: a declared interface or type literal counts
69+
as an engine when it declares a write door whose first parameter is named
70+
`object` / `objectName` / `objectApiName` / `name` and typed `string` — the
71+
`IDataEngine` door signature. Interfaces that extend one inherit it; aliases that
72+
narrow one (`Partial<Pick<IDataEngine, …>>`) carry it.
73+
74+
**The expensive failure direction is a keyword.** Sites whose receiver the author
75+
typed `any` have no type to read, and there are 45 of them — better than a fifth
76+
of the population, concentrated in exactly the seed and bootstrap paths this
77+
control exists for. Scoring an unreadable receiver as "not an engine" would have
78+
dropped every one of them silently, with a clean exit and a smaller number that
79+
reads exactly like a smaller truth. So `any` is not a classification here: it goes
80+
to the unreadable pile, and the unreadable pile is placed by facts about the tree
81+
(the argument names a declared object; the argument is a parameter declared
82+
`object: string`) or by a written `UNTYPED_RECEIVERS` row. **An unreadable
83+
receiver that none of the three place is an error, never a default.**
84+
85+
Tenancy itself is enabled *by default* — `isTenancyDisabled()` reads
86+
`tenancy.enabled === false` and nothing else — so the object registry only has to
87+
find the opt-outs. Across 297 declared objects — the dated, ⛔ unenforced
88+
corpus-scale figure below — exactly two opt out (`sys_api_key`,
89+
`sys_sso_provider`), and no write call site on this surface targets either.
90+
91+
## ⛔ What this instrument cannot decide
92+
93+
A site whose object name is an inline literal or a local `const` is statically
94+
decidable. A site whose name is a parameter or a field — `ql.insert(object, …)`
95+
inside a shared seed helper, `this.objectName` on a per-object service — is not,
96+
and no amount of AST work makes it so: the object is chosen at run time. Those
97+
are reported as `undecidable` rather than assumed either way.
98+
99+
The same holds twice over for the context. An options argument spelled as a
100+
literal can be read; one spelled `options`, `{ ...opts }`, or handed through a
101+
forwarding shim cannot, and **67 of the 215 sites are spelled that way**. A
102+
context resolved from an inline literal or a local `const` can be tested for
103+
`isSystem`; one arriving from a helper call cannot.
104+
105+
⇒ **A third of this population is undecidable in one dimension or another, and
106+
that is reported rather than smoothed over.** A census that quietly guessed on
107+
those sites would publish a number in the one direction this page cannot survive
108+
— claiming completeness while short — and a gate seeded from it would then hold
109+
the page to it.
110+
111+
## ⚠️ Two over-claims this instrument made, and what they cost
112+
113+
Both were found by the drift gate firing on this page's own artifact, and both
114+
are recorded here rather than quietly corrected, because the numbers they moved
115+
were published first.
116+
117+
**The elevation classifier read a spread as a denial.** It walked an object
118+
literal's named properties looking for `isSystem` and returned "not elevated"
119+
when the loop ended — so `{ ...SYSTEM_CTX }` was scored **decidably NOT
120+
elevated** while every `SYSTEM_CTX` in the tree is `{ isSystem: true, … }`. It
121+
also never unwrapped `as const`, so a bare `context: SYSTEM_CTX` read as
122+
undecidable. Fixing both moved **51 sites' verdicts**: 45 from undecidable to
123+
elevated, and 6 from *decidably not elevated* to *elevated* — an inversion, on a
124+
security-relevant classification. The count of decidably-not-elevated sites is
125+
now **0**: nothing on this surface threads a context that provably lacks the flag.
126+
127+
**"No tenant context" counted sites it had not read.** An options argument the
128+
walker could not parse was folded into the same bucket as one it had read and
129+
found empty. That published **84 sites "carrying no tenant context at all"**
130+
when 17 said so and 67 were simply unread — an over-claim in the *alarming*
131+
direction, on the very figure this page tells other cards to cite. `carries` is
132+
now three-valued, and an unreadable argument can never contribute to the
133+
provable count.
134+
135+
⭐ Both are the same shape as the failure this artefact exists for, wearing
136+
opposite hats: one scored an unread thing as *nothing to report*, the other
137+
scored an unread thing as *a finding*. `node
138+
scripts/check-tenant-audit-census.mjs --self-test` pins all of it — 18 classifier
139+
cases, 6 of which red against the old reading, driven from the gate's own
140+
self-test rather than from a flag on the generator.
141+
142+
## ⭐ Deviations from the carried figures, untrimmed
143+
144+
Four figures were carried forward from the lost census. They are **control inputs
145+
for this re-derivation, not targets**, and this instrument was not shaped to
146+
reproduce them. Where it disagrees, it disagrees on the page:
147+
148+
| carried figure | where it survives | this census |
149+
| :--- | :--- | ---: |
150+
| 175 write call sites | quoted in the merged changeset | **215** |
151+
| 24 carrying no tenant context | quoted in the merged changeset | **9** provable and tenancy-enabled; **32** more whose options argument is unreadable |
152+
| 127 of 175 statically decidable, 48 runtime-parameter-name sites | restated on the `isSystem`-scoping card | **143 of 215** decidable, **72** undecidable |
153+
| 135 (77%) silenced by the `isSystem` guard before the posture gate | the lost issue body — **no surviving corroboration** | **not reproduced**: 97 decidably elevated, 0 decidably not, 101 undecidable |
154+
| 141 and 132, two independent re-derivations | the card that filed this work | — |
155+
156+
**The differences are not reconciled, and deliberately so.** The old census's
157+
population definition is not recoverable — that is the whole reason this page
158+
exists — so no honest reading can attribute the gap to a specific cause. What can
159+
be stated is what this instrument counts, which is written above and re-runnable
160+
at any commit.
161+
162+
Two structural facts do plausibly widen this reading against any hand or regex
163+
one, and both are counted in the generated tables below: the 45 sites reached
164+
through an erased (`any`) receiver, and the 37 that name their object through a
165+
`const` rather than inline. An instrument that read either the way a person does
166+
would report a smaller number and would not say so.
167+
168+
The fourth row is the one worth flagging to anyone citing it. **The 135 / 77%
169+
figure has no surviving corroboration anywhere in the tree.** This census reads
170+
97 of 215 (45%) as decidably elevated, with 101 more whose elevation is a
171+
run-time fact — so the claim is neither confirmed nor refuted, and the honest
172+
answer is that a static reading cannot settle it.
173+
174+
⇒ **Cite `9 / 215`, and say what it is**: the sites whose options argument was
175+
READ and holds no tenant context, against a decidably tenancy-enabled object.
176+
That is the control's provable yield surface. ⛔ Do not cite it as "the sites
177+
without tenant context" — **32 further sites** have an options argument this
178+
cannot read, and they are neither in nor out.
179+
180+
{/* BEGIN GENERATED: tenant-audit-census (scripts/tenant-audit-census.mjs) — DO NOT EDIT */}
181+
182+
## The measurement
183+
184+
| what | count |
185+
| :--- | ---: |
186+
| write call sites on the application surface | **215** |
187+
| …whose object name is statically decidable | 143 |
188+
| …whose object name is chosen at run time | 72 |
189+
| …against an object with tenancy ENABLED | 143 |
190+
| …against an object that declares tenancy off | 0 |
191+
| threading a tenant context | 131 |
192+
| PROVABLY carrying none (options read, no context key) | **17** |
193+
| …of those, against a decidably tenancy-enabled object | **9** |
194+
| options argument UNREADABLE — may or may not carry one | 67 |
195+
| …of those, against a decidably tenancy-enabled object | 32 |
196+
| threading a decidably ELEVATED (`isSystem`) context | 97 |
197+
| threading a context that is decidably NOT elevated | 0 |
198+
| threading a context whose elevation is a run-time fact | 101 |
199+
200+
| how the instrument reached the site | count |
201+
| :--- | ---: |
202+
| receiver carried a readable engine type | 170 |
203+
| receiver erased, placed by the object NAME | 19 |
204+
| receiver erased, placed by an `object: string` PARAMETER | 15 |
205+
| receiver erased, placed by an `UNTYPED_RECEIVERS` row | 11 |
206+
207+
| object name spelled inline | 106 |
208+
| object name spelled through a `const` | 37 |
209+
| object name is an `object: string` parameter | 19 |
210+
| object name is some other run-time expression | 53 |
211+
212+
The corpus walked is every tracked non-test source under `packages/services/`
213+
and `packages/plugins/`; calls to a same-named method on something that is not
214+
a data engine were subtracted. Every site is listed in
215+
[`docs/audits/2026-08-tenant-audit-write-call-sites.counts.md`](https://github.com/objectstack-ai/objectstack/blob/main/docs/audits/2026-08-tenant-audit-write-call-sites.counts.md),
216+
regenerated by the same command.
217+
218+
### Corpus scale — present and dated, ⛔ NOT enforced
219+
220+
⛔ These four describe the CORPUS this census walked, not the population it
221+
certifies, and the gate deliberately does not hold them to the tree — a source
222+
file arriving anywhere under the two roots moves them while every verdict above
223+
holds still. They are required to be HERE and to say WHEN they were true;
224+
their values are not compared. The reasoning, and the measurement behind it,
225+
are in `scripts/check-tenant-audit-census.mjs`.
226+
227+
Measured on 2026-08-31 at `fc8858a24`.
228+
229+
| corpus scale (not enforced) | count |
230+
| :--- | ---: |
231+
| tracked non-test sources scanned | 534 |
232+
| engine-shaped types recognised | 56 |
233+
| declared objects in the registry | 297 |
234+
| same-named calls subtracted as non-engine | 119 |
235+
236+
{/* END GENERATED: tenant-audit-census */}

0 commit comments

Comments
 (0)