CB-WP-0038: variant selection, H1 implemented, and H1 measured
ground-game packages hypotheses as selectable rules variants — a catalog,
a rules_delta.yaml, and prose — and their note is explicit that CSV text
alone is not executable here. So the kernel gains a Variant in game state:
in the state, therefore in the hash, therefore in the recording, because a
scenario replayed under a different variant would diverge silently.
Baseline is bit-for-bit what it was, asserted across seat counts and
seeds. A variant system that perturbs the baseline invalidates every
measurement this repo has.
H1-A and H1-B implemented from rules_delta.yaml and mutation-proven on
their own defects: "unclaimed" misread as face-up-and-unsolved, and the
attacker's Stress read after the attack's effects. Their `unchanged:` list
is asserted rather than trusted — that list is their claim about their own
experiment.
Measured, and three of their four criteria fail. DARVO arm rate is still
0 under greedy; ATTACK selection does not rise and falls for the rank-75
policy; group success collapses from 165/190/200 to 0 at 3/4/6 seats.
The mechanism is not the assumed one: greedy answers the pressure by
regulating, Stress plateaus at 3, so it never reaches the gate at 4 or the
arm at 5 — H1-A acts as a solve-rate tax and H1-B is unreachable under
competent play.
A harness defect was caught before the claim: sweep discarded refused
games silently and never reported its count, so "nobody won" and "nothing
played" printed identically. Reporting H1 as unwinnable on that basis
would have been the ADR-0018 family aimed at another repo's design. All
200 games ran in every cell; the zeros are real.
Chaos d8 = 8 — the window's first override, redrew L against a structural
L, so it changed nothing. Window 3 recorded in ChaosRollHistory.
NOT REVIEWED: tier L owes a separate-agent adversarial review, and no H1
result may reach ground-game until it has run.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-08 00:50:08 +02:00
# CB-EV-0030 — H1 measured
CB-WP-0038. Tier L (structural L; **chaos d8 = 8 → override, redraw = L**
— the override changed nothing). **Declaration 10 of chaos window 3, and
the window's first 8.** Closed 2026-08-08.
> **This evidence has not had its tier-L adversarial review** (InnerLoop
> Step 2). **It may not go to `ground-game` until it has.** It is a
> measurement of another repo's design hypothesis, which is precisely the
> case the review exists for.
**Delivered:** variant selection in the kernel, H1-A and H1-B implemented
from `rules_delta.yaml` , ground-game's `unchanged:` list asserted rather
than trusted, and the ATTACK panel re-run for both rule sets.
---
## 1. The result, against their own criteria
`ground-game` 's success criteria (design note §3.2), answered one by one.
**Three of four fail.**
CB-REV-0002: round 2, and the corrections were not approvable either
Three FATAL, five SERIOUS. The substance of round 1's corrections held —
Reactive is genuinely one arm different, the five replacement controls are
non-inert, the inert metric is right, the numbers reproduce. What failed
were the CLAIMS about them, and two defects the corrections introduced.
FATAL 1: the fix for round 1's #11 did not fix it. The assertion was
`games + setup_fails == 200`, and a refused setup increments setup_fails
while skipping games — so the sum is invariant under exactly the failure
it claimed to catch. Injecting setup failures gave exit 0 over 196-game
columns. Now asserts games == GAMES, verified to exit 101.
FATAL 2: the correction to the selective-column FATAL was itself
selective. "81-1000 per cell, baseline AND H1" and "31-1000" twelve lines
apart, both taken from the baseline row; under H1 rank-75 arms are
59/0/0/0. Every cell is now printed rather than summarised, and the
corrected verdict is the opposite of the one it replaced: under rank-75,
H1 REDUCES DARVO arms to zero at 3p and above.
FATAL 3: "DARVO arms 2 per seat per game" is 1 per seat per game, exactly,
at every band.
SERIOUS: the tiebreak oracle asserted only that the winner set CHANGED, so
reversing the tiebreak left it green; the #13 defect's impact was claimed
and never measured (72,000 games: zero divergences — real in principle,
witnessed only by a constructed board); a 29-of-363 citation pointed at a
file that did not contain it (round 1's reviewer did report it, and it was
never transcribed — the record was wrong, not the number); the harnesses
were run by NO GATE, so every published figure came from a manual run of
an ungated binary, including the assertion added for #1; and edition-check's
sibling handling — added by the last correction — was self-certifying,
crashed instead of failing, and counted Markdown lines as coverage. Now
discovered on disk, and it found a real gap on its first run: Rules_Text.csv
vendored with no digest.
Also: "peak Stress held" was dead code kept quiet by `let _ = held;` — the
numbers were right by coincidence.
make panels is now a registered gate. Round 3 is owed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-08 09:44:00 +02:00
> **CORRECTED TWICE.** Round 1 found criteria 1 and 2 rendered as flat
> failures **on the greedy column alone**, while this pass's own harness
> printed non-zero DARVO arms in the rank-75 and rank-95 columns — the
> selective-column move, in the file that names it.
>
> **Round 2 found the correction was itself selective, and its numbers
> wrong** ([`CB-REV-0002`](../reviews/CB-REV-0002-h1.md) #2): it quoted
> "81–
> "31–
> **H1**, rank-75 arms are `2p 59, 3p 0, 4p 0, 6p 0` — three cells at zero
> and the fourth below the stated floor.
>
> **Every cell is now given**, because two attempts at summarising this
> into a range have now produced three wrong numbers.
**DARVO arms per 200-game cell, SHARED GROUND** (all three modes agree on
arms; they differ only in who wins):
| | greedy | rank-75 | rank-95 |
|---|---:|---:|---:|
| **baseline** 2p / 3p / 4p / 6p | 0 / 0 / 0 / 0 | 81 / 125 / 218 / 31 | 400 / 805 / 1000 / 1000 |
| **H1** 2p / 3p / 4p / 6p | 0 / 0 / 0 / 0 | **59 / 0 / 0 / 0** | 600 / 870 / 941 / 970 |
CB-REV-0001: the adversarial review, and it was not approvable
Thirteen challenges, five FATAL, all five conceded. Nothing had reached
ground-game, which is the only reason this is a correction and not a
retraction.
The worst: `Reactive` was not "greedy with one preference changed". It
differed in five, including SpendFreedom — ranked 95 unconditionally
against greedy's `95 if gated else 0` — so the seat burned its Freedom
token in round one of every game. A second change to the exact mechanism
under study, and every number in CB-EV-0031 was measuring it. The pass
claimed ADR-0018's one-varying-parameter discipline in its own workplan
while violating it. GreedyPolicy::rank is now public and the policy
delegates, overriding one match arm, so the control is structurally true.
Withdrawn entirely: "H1-B suppresses DARVO in the attacker". Disabling
H1-B under the corrected policy changes the arm count by exactly zero.
The pass hedged the wrong variable — it disclaimed "the number 2" and
defended "the direction", and the direction is what failed. The
supporting inference was invalid anyway: final Stress cannot show who
armed, because DarvoEnded resets the stage and REVERSE gives its owner -2.
Corrected: criterion 1 was failed on the greedy column while the pass's
own printed table showed 31-1000 arms in the other columns — the
selective-column move, in the file that names it. "Peak Stress was 1" was
a maximum over StressSet payloads, not held state (true: 2); the baseline
game count was 1,600 not 3,200; and "a reckless policy plays identically
to a careful one" is refuted by this repo's own rank-95 policy.
Inert controls replaced, each verified red against the reviewer's own
mutation: the baseline hash test compared two identically-constructed
states (serde(skip) on variant left 57/57 green); the `unchanged:` test
checked 3 of 7 entries and passed with SOLVE made illegal; H1-A's ordering
and H1-B's OU-cancel path had no test at all.
edition-check now covers catalog.yaml and rules_delta.yaml, whose digests
CB-WP-0038 claimed and never recorded — the review found it and reported
it unverified rather than absent, which was the right call.
Still open: H1-B on the DARVO extra-Attack path is untested, regulation.rs
still skips setup failures silently, and round-5 arms are counted though
they can never act.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-08 02:02:35 +02:00
CB-WP-0038: variant selection, H1 implemented, and H1 measured
ground-game packages hypotheses as selectable rules variants — a catalog,
a rules_delta.yaml, and prose — and their note is explicit that CSV text
alone is not executable here. So the kernel gains a Variant in game state:
in the state, therefore in the hash, therefore in the recording, because a
scenario replayed under a different variant would diverge silently.
Baseline is bit-for-bit what it was, asserted across seat counts and
seeds. A variant system that perturbs the baseline invalidates every
measurement this repo has.
H1-A and H1-B implemented from rules_delta.yaml and mutation-proven on
their own defects: "unclaimed" misread as face-up-and-unsolved, and the
attacker's Stress read after the attack's effects. Their `unchanged:` list
is asserted rather than trusted — that list is their claim about their own
experiment.
Measured, and three of their four criteria fail. DARVO arm rate is still
0 under greedy; ATTACK selection does not rise and falls for the rank-75
policy; group success collapses from 165/190/200 to 0 at 3/4/6 seats.
The mechanism is not the assumed one: greedy answers the pressure by
regulating, Stress plateaus at 3, so it never reaches the gate at 4 or the
arm at 5 — H1-A acts as a solve-rate tax and H1-B is unreachable under
competent play.
A harness defect was caught before the claim: sweep discarded refused
games silently and never reported its count, so "nobody won" and "nothing
played" printed identically. Reporting H1 as unwinnable on that basis
would have been the ADR-0018 family aimed at another repo's design. All
200 games ran in every cell; the zeros are real.
Chaos d8 = 8 — the window's first override, redrew L against a structural
L, so it changed nothing. Window 3 recorded in ChaosRollHistory.
NOT REVIEWED: tier L owes a separate-agent adversarial review, and no H1
result may reach ground-game until it has run.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-08 00:50:08 +02:00
| # | criterion | verdict |
|---|---|---|
CB-REV-0002: round 2, and the corrections were not approvable either
Three FATAL, five SERIOUS. The substance of round 1's corrections held —
Reactive is genuinely one arm different, the five replacement controls are
non-inert, the inert metric is right, the numbers reproduce. What failed
were the CLAIMS about them, and two defects the corrections introduced.
FATAL 1: the fix for round 1's #11 did not fix it. The assertion was
`games + setup_fails == 200`, and a refused setup increments setup_fails
while skipping games — so the sum is invariant under exactly the failure
it claimed to catch. Injecting setup failures gave exit 0 over 196-game
columns. Now asserts games == GAMES, verified to exit 101.
FATAL 2: the correction to the selective-column FATAL was itself
selective. "81-1000 per cell, baseline AND H1" and "31-1000" twelve lines
apart, both taken from the baseline row; under H1 rank-75 arms are
59/0/0/0. Every cell is now printed rather than summarised, and the
corrected verdict is the opposite of the one it replaced: under rank-75,
H1 REDUCES DARVO arms to zero at 3p and above.
FATAL 3: "DARVO arms 2 per seat per game" is 1 per seat per game, exactly,
at every band.
SERIOUS: the tiebreak oracle asserted only that the winner set CHANGED, so
reversing the tiebreak left it green; the #13 defect's impact was claimed
and never measured (72,000 games: zero divergences — real in principle,
witnessed only by a constructed board); a 29-of-363 citation pointed at a
file that did not contain it (round 1's reviewer did report it, and it was
never transcribed — the record was wrong, not the number); the harnesses
were run by NO GATE, so every published figure came from a manual run of
an ungated binary, including the assertion added for #1; and edition-check's
sibling handling — added by the last correction — was self-certifying,
crashed instead of failing, and counted Markdown lines as coverage. Now
discovered on disk, and it found a real gap on its first run: Rules_Text.csv
vendored with no digest.
Also: "peak Stress held" was dead code kept quiet by `let _ = held;` — the
numbers were right by coincidence.
make panels is now a registered gate. Round 3 is owed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-08 09:44:00 +02:00
| 1 | DARVO arm rate non-trivial, "not 0/500" | **depends entirely on the policy, and H1 does not improve it for the middle one.** Zero under greedy in both. Under rank-75, H1 **reduces** arms to zero at 3p and above. Only rank-95 — a seat that always attacks — arms freely, and it does so **under the baseline too** |
| 2 | ATTACK selection rises for some subpopulation | **no, for the policy the criterion is about.** Greedy attacks 0 times under both. rank-75's attacks fall 604 → 180 at 2p and **592/923/389 → 0** at 3p/4p/6p. Only the deliberately unregulated seat of CB-EV-0031 rises, and it never wins |
CB-WP-0038: variant selection, H1 implemented, and H1 measured
ground-game packages hypotheses as selectable rules variants — a catalog,
a rules_delta.yaml, and prose — and their note is explicit that CSV text
alone is not executable here. So the kernel gains a Variant in game state:
in the state, therefore in the hash, therefore in the recording, because a
scenario replayed under a different variant would diverge silently.
Baseline is bit-for-bit what it was, asserted across seat counts and
seeds. A variant system that perturbs the baseline invalidates every
measurement this repo has.
H1-A and H1-B implemented from rules_delta.yaml and mutation-proven on
their own defects: "unclaimed" misread as face-up-and-unsolved, and the
attacker's Stress read after the attack's effects. Their `unchanged:` list
is asserted rather than trusted — that list is their claim about their own
experiment.
Measured, and three of their four criteria fail. DARVO arm rate is still
0 under greedy; ATTACK selection does not rise and falls for the rank-75
policy; group success collapses from 165/190/200 to 0 at 3/4/6 seats.
The mechanism is not the assumed one: greedy answers the pressure by
regulating, Stress plateaus at 3, so it never reaches the gate at 4 or the
arm at 5 — H1-A acts as a solve-rate tax and H1-B is unreachable under
competent play.
A harness defect was caught before the claim: sweep discarded refused
games silently and never reported its count, so "nobody won" and "nothing
played" printed identically. Reporting H1 as unwinnable on that basis
would have been the ADR-0018 family aimed at another repo's design. All
200 games ran in every cell; the zeros are real.
Chaos d8 = 8 — the window's first override, redrew L against a structural
L, so it changed nothing. Window 3 recorded in ChaosRollHistory.
NOT REVIEWED: tier L owes a separate-agent adversarial review, and no H1
result may reach ground-game until it has run.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-08 00:50:08 +02:00
| 3 | group success does not collapse | **fails hardest** — greedy's wins go **165 → 0** (3p), **190 → 0** (4p), **200 → 0** (6p) |
| 4 | Bond/GROUND stay better than DARVO | holds, but **vacuously** : DARVO never fires |
SHARED GROUND, greedy column, 200 games per cell:
| seats | baseline won | H1 won |
|---|---:|---:|
| 2p | 132 | **68** |
| 3p | 165 | **0** |
| 4p | 190 | **0** |
| 6p | 200 | **0** |
## 2. Why — and the mechanism is not the one H1 assumed
H1-A was meant to raise Stress toward the DARVO arm at 5. **It does not
get there under competent play.** One 3-player game, seed 1:
| | rounds | final Stress | Problems unclaimed | DARVO armed | outcome |
|---|---|---|---|---|---|
| baseline | 5 | 0, 1, 2 | 0 | 0 | 9 of 7 — **success** |
| H1 | 5 | 3, 3, 3 | 2 | 0 | 4 of 7 — **failure** |
**Greedy spends its actions on Stress instead of Problems.** The pressure
is answered by regulating, so Stress plateaus around 3 — below the gate at
4 and well below the arm at 5 — while two Problems go unclaimed and the
total lands under the threshold.
**So H1-A behaves as a solve-rate tax, not as DARVO pressure.** And since
Stress never reaches 4, **H1-B never fires at all under competent play** :
the self-soothe is unreachable by the policy it was designed to tempt.
**The seat-count pattern is the tell.** At 2p — three Problems, so a table
can clear the board — H1 *helps* in SHARED (132 → 68 is a fall, but 2p is
the only band that still wins at all). At 3p and above the board is bigger
than the pressure allows a table to clear, and success goes to zero. The
tax scales with the number of Problems; the intended effect does not.
## 3. The harness defect caught before the claim
The first run showed `won` , `atk` and `darvo` **all exactly 0** at 3+
seats. That is the signature of games being *discarded* , not lost.
`sweep` had `let Ok(g) = play(st, &mut ps) else { continue };` and **never
reported its game count**, so a run in which nothing played and a run in
which nothing won were the same output.
> **Reporting "H1 makes the game unwinnable" when the truth might have
> been "our harness threw every H1 game away" would have been the
> ADR-0018 family — aimed, this time, at another repository's design
> decision.**
Instrumented: setup failures counted, `play` errors captured and printed,
`games != 200` reported. **All 200 ran in every cell.** The zeros are
real, and now they are known to be real.
## 4. What this evidence does not say
- **It does not say H1 is a bad idea.** It says these two deltas, at these
magnitudes, measured by this panel, do not do what §3.2 asks. Their §6
already has the branch for it: *"H1 measured | DARVO still ~0 | add Idea
2 last-place stress (semi only) as H2"*.
- **The panel is greedy-family throughout** — greedy, and greedy with
ATTACK reranked to 75 and 95. Plural, per RetrospectiveAnalysis §4.1,
but not diverse. A policy that *manages Stress badly* would reach 5 and
arm DARVO, and none of these do. **That is the obvious next measurement
and it has not been run.**
- **No felt-play.** GROUND-WP-0003 T01–
and 200-game aggregates cannot say what a table feels like.
## 5. A tuning direction, offered as a question and not a finding
If the aim is Stress reaching 5 sometimes, the tax must outrun regulation.
Candidates, none measured: pressure scaling with the *number* of unclaimed
Problems rather than a flat +1; applying it only above a Round threshold;
or lowering the DARVO arm below 5. **Each is a rules change and therefore
ground-game's to make, not ours to propose as a fix.**
## 6. Cost
`make cost` , not inlined. One vendored catalog and one experiment package;
no new dependency.
## Open after this pass
- **The tier-L adversarial review**, before any of this travels.
- **A policy that regulates badly**, which is the measurement §4 names.
- **H1-B has never been observed firing** in a real game — only in a unit
test. A delta that competent play cannot reach is untested where it
matters.
