# Symmetry test — the faction rule applied to the seeds

**Run:** 2026-09-21, by Claude, discharging the obligation set in
`AMENDMENT-2026-09-21-locus-and-s5.md` §4: *"before §3 is used to score anything, it should be applied
to at least one seed whose score it would raise, and the result published."*

**Why it was owed.** The faction ruling (amendment §3) holds that resistance, reversal under pressure
and internal turmoil are downstream battle rather than evidence against the signature. That is a
**construct refinement, not a T1 allowance**, so the packet's double-standard condition makes it
available to every architecture in the register or to none. Applied to the seeds it can only ever
*raise* scores, since each seed's disconfirming evidence becomes reassignable to a non-sabbotarchic
remainder. The seeds already sit 76–91 with nothing below 70, which is the distribution Part VIII
concedes a ceiling-slamming instrument produces. **An unbounded faction rule is the ceiling-slamming
charge arriving through a side door, and nothing in the register would catch it there.**

**Result, stated first: the rule cannot materially inflate the seeds. Maximum inflation is +2 composite
points on any seed, and no seed changes band.** The reason is structural and is the useful finding.

---

## §1 — Which seed moderations the rule actually reaches

Every seed dimension was read for moderations of the type §3 neutralizes — a score held down because a
fix was reversed, a purge was incomplete, or resistance was contested from inside.

| Seed | Dim | Score | Moderation present | Does §3 reach it? |
|---|---|---|---|---|
| IRS | S4 | **76** | "7,315 probationary termination notices… **reinstated by court order Mar 17 2025**; durable reduction came via deferred-resignation and resignations, with ~3,000 returning" | **Yes** — a reversal extracted by court order is the paradigm case |
| DOGE | S3 | 86 | "a Sept 2025 federal ruling found the firings unlawful while declining to order reinstatement… the posture is **partly judicial-procedural rather than purely executive**" | **Partly** — judicial, so external to the architecture; §3 concerns the non-sabbotarchic remainder *inside* it |
| VA | S4 | **64** | "career-clinical-staff… **pressured but not categorically displaced**; senior medical-officer architecture **retained largely intact**" | **Only on a loose reading** — incompleteness is not reversal. Tested below as the worst case |
| VA | S3 | 71 | "cheap fixes resisted **with delay rather than categorical refusal**" | **No** — manner of resistance, not reversal |
| OBBBA | S4 | 73 | "career-staff… marginalized or RIFed" | **No** — completion, not reversal |
| NIH | S1 | 82 | magnitude claim withdrawn as unsupported | **No** — epistemic restraint, a different class entirely |
| IRS | S4 | 76 | "'specifically targeted the competent' is an **inference beyond the record and not asserted**" | **No** — epistemic restraint; must survive §3 untouched |

**The separation that matters:** most seed moderations are *epistemic restraint* — a claim withdrawn
for want of evidence, a figure ranged, an inference declined. The faction rule does not reach those and
must never be read to. It reaches only moderations that credit the architecture for a correction it did
not author.

## §2 — The arithmetic, at maximum inflation

Worst case: assume §3 is read as loosely as possible, every reversal-and-incompleteness moderation is
removed across all seven seeds, and each seed's S4 rises to the highest S4 in the register (DOGE and
Rollback's 84). S3 is left alone, since no seed's S3 moderation is a reversal by the architecture's own
remainder.

| Seed | S1 | S2 | S3 | S4 → | S5 | Composite now | Max-inflated | Band |
|---|---|---|---|---|---|---|---|---|
| IRS | 94 | 95 | 89 | 76 → 84 | 92 | 91.05 → **91** | 91.85 → **92** | Extreme → Extreme |
| DOGE | 88 | 91 | 86 | 84 (no change) | 93 | 89.40 → **89** | **89** | Extreme → Extreme |
| Rollback | 88 | 85 | 86 | 84 (no change) | 92 | 87.90 → **88** | **88** | Extreme → Extreme |
| ICE | 87 | 88 | 81 | 79 → 84 | 86 | 85.20 → **85** | 85.70 → **86** | Extreme → Extreme |
| OBBBA | 92 | 79 | 82 | 73 → 84 | 89 | 85.10 → **85** | 86.20 → **86** | Extreme → Extreme |
| NIH | 82 | 84 | 74 | 68 → 84 | 81 | 79.50 → **80** | 81.10 → **81** | High → High |
| VA | 80 | 79 | 71 | 64 → 84 | 76 | 75.65 → **76** | 77.65 → **78** | High → High |

**Distribution: 76–91 becomes 78–92. No band changes. The largest single move is VA, +2.**

## §3 — Why the rule is bounded, and where that stops being true

The moderations §3 reaches sit almost entirely on **S4 (weight 0.10)** and **S3 (weight 0.15)** — the
two lowest-weighted dimensions. The seeds' scores are driven by **S1 (0.25), S2 (0.20) and S5 (0.30)**,
where the moderations are of the epistemic-restraint class §1 excludes. S4's entire 0–100 range moves a
composite by 8.0 points; the realistic moves here are 5–20 points of S4, or 0.5–2.0 composite points.

**This is the load-bearing caveat.** The rule is safe *because* it only touches S3 and S4. Extended to
the high-weight dimensions it stops being safe:

- On **S5 (0.30)**, the argument would be "the external incoherence was the remainder fighting back."
  That reading moves a composite by up to 30 points and would cross band lines freely.
- On **S1 (0.25)** it would be "the diffuse incidence is the remainder blunting the extraction."
- On **S2 (0.20)** it would be "the public-facing function survived because the remainder defended it" —
  which would neutralize the S2 signature entirely, on every architecture.

**Note that T1's live S5 question has exactly the shape of the first bullet.** The faction rule and the
S5 structural argument are close relatives, and if the faction rule is ever allowed onto S5 it becomes
the most powerful score-raising device in the instrument.

## §4 — Recommended bound, now with arithmetic behind it

Combining this with the bound proposed in amendment §4:

1. **Faction attribution is available only where the record names *who* acted and *against whose
   objection*.** IRS S4 meets it (court order, dated, 7,315 notices, ~3,000 returning). VA S4 does not —
   incompleteness is not reversal.
2. **The faction rule applies to S3 and S4 only.** Using it on S1, S2 or S5 requires its own ruling,
   because only there is it capable of moving a band.
3. **It never reaches epistemic restraint.** A claim withdrawn for want of evidence, a ranged figure, an
   inference declined — these are the instrument working, and §3 must not be read to reverse them.
4. **A corrective act issuing from the architecture's own leadership in its own name, with no documented
   internal opponent, counts as the architecture remediating** and scores against the signature. This is
   the T1 R4/R5 case and it is unaffected by anything in this test.

## §5 — What this test establishes and does not

**Establishes:** that adopting the faction rule does not inflate the current seed distribution past
+2 composite points, changes no seed's band, and therefore does not smuggle in the ceiling-slamming
failure — *provided* it stays on S3 and S4.

**Does not establish:** that the rule is correct. It shows the rule is *contained*, not that it is
right. Nor does it establish anything about future architectures: a seed whose score rested more heavily
on an S3 or S4 reversal would move further, and one sitting within 2 points of a band line would move
across it. **OBBBA and ICE both sit exactly on the 85 boundary**, and under maximum inflation both go to
86 — still Extreme, but they are the two rows where a future application of this rule should be checked
before it is applied rather than after.

**Method:** scores read from the live `ARCHITECTURES` data in `sabbot/index.html`; composites recomputed
independently and verified against the canonical published values (IRS 91 · DOGE 89 · Rollback 88 ·
ICE 85 · OBBBA 85 · NIH 80 · VA 76) before any counterfactual was run. No seed score was changed by this
test, and none should be changed on the strength of it.