Reported three times across three sessions, two days apart, and it
survived a whole UI rebuild: "i cant see whos support i accept".
It was not styling. The move button's label was format!("{c:?}"), so the
player read `RespondToSupport { response: AcceptBond }` — Rust struct
syntax with no name in it. And the command does not carry the
counterparty, so nothing rendering it alone could have said who; it comes
off the view, as whoever played Support at this seat.
Four more seat-panel fields had the same defect, including `support
AcceptBond` — the one the report names. CB-WP-0020 fixed exactly this for
selections and left its four neighbours as they were.
command_label has no catch-all arm, and that earned its keep before any
test ran: GroundChoice::RejectReverse and SupportResponse::BreakRivalry
both failed to compile — two moves that would have shipped as struct
dumps. An offer the view cannot see is said to be unseen rather than given
an invented name.
The finding underneath: the coverage probe that exists to prove every view
field reaches the PLAYER was matching "player: Some(PlayerId(1))" and
"members: [PlayerId(1)". It was certifying the defect as coverage and
would have gone red had anyone fixed it. Second confirmation of
CB-WP-0024's finding, from the sharper side: a probe naming Debug output
does not merely fail to protect, it pins the defect in place.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
make trials reported "positions unreachable: 4, target 0 — the recording
exists but the position moved". All four were false. A recording holds one
hash, the final state, and reachability asked whether the note's hash was
in that file — so a mid-game note could never match, and a post-game note
from any but the last game could not either. Instance 8 of the ADR-0018
family: vary only WHEN a note was written and the answer flips, with
nothing having moved.
The root cause was not the metric. play again reused state belonging to a
game: it overwrote the previous game's recording (data loss), never
cleared the journal (game 2's log opened with game 1's commands), and so a
note's command index pointed into a recording without those commands.
Fixing reachability alone would have gone green while a session still
destroyed its own evidence.
A note now binds by (game, after) — an index into the recording's own
commands list, which a reader can replay to. The hash keeps a job as the
integrity check at the end of a game, where it can actually fail. Game 1
keeps the path it was given, so GameDesign §5's documented invocation is
unchanged; later games get -2, -3 and nothing is overwritten. Legacy
5-column logs stay readable and are reported as legacy, never as orphans —
an unsubstantiated orphan claim is the defect being fixed.
All three fixes mutation-proven, including at the call site via a real
two-game session.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Notes were written to a file and shown back nowhere — the shape
trials.py's own docstring calls this project's signature failure in a new
medium. The log now carries the player's comments where they were made.
Position is the feature: a remark like "why did that do nothing?" is about
the move above it, and collected at the bottom it is a sentence with no
subject. round/step cannot order a note against the log because several
commands share a step, so the server records how many commands had been
played — which it knows exactly — and the page places it there.
A comment must not be readable as something the game did. The log is the
recorder's vocabulary; notes render in their own block, attributed to the
player, quoted. The .note CSS already existed and nothing had ever used it.
No column was added to the trial log. trials.py skipped any row that was
not five cells, silently, so a sixth column would have made `make trials`
report zero notes for every log at once. That latent defect is fixed on
its own terms: a wrong column count now raises, and the walk reports the
real reason rather than blaming a missing block for every failure.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The note channel closed at the moment it is worth most — a player who has
just seen the outcome is the one with something to say, and that reading
was collectable at every moment of the game except the one after it.
Two independent defects: doc::ending never rendered the box, and
serve_end had no POST /note arm, so even a hand-built post fell through to
404. Fixing either alone leaves the channel shut, so the test asserts both
and is mutation-proven to fail on each half separately.
A post-game note binds to the final position but is not an observation
made at the last decision point. RoundStep::End is the last step of a
ROUND, not the end of the game, so record_note now takes the step as an
argument and the post-game path passes "after the end" — otherwise an
after-the-fact reading is filed as an in-play one, which is the
wrong-subject family ADR-0018 was written for.
A note does not end the session: every other POST in that loop breaks it,
and a player must be able to write a second one and then still play again.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Seven claims in this project have been wrong the same way — the arithmetic
right, the subject wrong. Four reached a document or another repo.
ADR-0018 states the rule: a measured claim names the variable it depends
on and shows the measurement moving with it. GameDesign §1.4 adds it as a
fourth admissibility clause; the existing three are untouched.
Honest yield is 3 clear + 1 partial of 7. It misses a count never made and
a claim that is not a number, and D3 names those so nobody stops looking.
The tempting fix — closing facts-check's 62 untagged literals — would have
caught 1 of 7 and is deferred as separate work rather than bundled.
loop-lint gains check_sensitivity_stated over arithmetic register rows,
mutation-proven red on F17 and green when restored, with four self-test
controls. It checks presence, never adequacy, and says so in its output.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Declaration 1 of chaos window 3 (opened by ADR-0017), tier M, chaos d8=7,
no override.
Seven claims in this project have been wrong the same way, and the
workplan tabulates all seven with what caught each.
MEASURED BEFORE PROPOSING: facts-check reports 62 untagged literal copies
and fails on none, which is a real drift surface it names and does not
close. CLOSING IT WOULD HAVE CAUGHT AT MOST ONE OF THE SEVEN. Instance 1's
denominator WAS a registry fact and WAS correct -- the error was that
AM-4a does not cover the code. Four others were fresh measurements with no
registry key. A tag proves a number came from somewhere; it cannot prove
the somewhere was the right place. Named as separate work rather than
bundled, because bundling would let this pass claim a victory it did not
win.
What did work: every instance was caught when the number was computed a
SECOND TIME by something with a different purpose. Two of them share a
sharper property -- the claim was tested against a variation. #4's cost
FELL as branching ROSE, which no per-node cost can do, and the tell was in
the published output. #5's win rate was attributed to the game, and a
second policy on the same deals gave 0% where the first gave 100%.
So the rule generalises what this project already builds after each
instance (the policy panel, the attack-value sweep) and requires it
before: a measured claim must name the variable it depends on and show the
measurement moving with it. A number that does not move when its subject
moves is measuring something else.
HONEST YIELD: 3 or 4 of 7, and it catches the two that travelled furthest.
It does not catch a count never made, or a claim that is not a number.
Claiming it closes the family would be the family's own error applied to
itself.
T03 requires the lint to fail on a real row by mutation, and to say in its
own output that it checks PRESENCE and not ADEQUACY -- so nobody reads a
green lint as 'the claim was verified'.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
not
Tier M (changes how the loop constrains its own operation), declared at
d8 because the rate for window 3 is what this document decides and
declaring at a rate it invents would be circular. chaos d8 = 7, no
override.
I CLAIMED WINDOW 2 PRODUCED ZERO OVERRIDES, FIVE TIMES, AND IT IS FALSE.
Declaration 3 (CB-WP-0020) rolled d8 = 8, overrode, drew S against a
structural S, and changed nothing -- and CB-WP-0020 recorded it correctly
at the time, in those words: "the first override at d8... It changed
nothing... One."
Counting the workplans takes one command and I never ran it. CB-EV-0024
asserted "zero" without checking; CB-EV-0025, 0026, 0027 and CB-WP-0029
each cited the one before. A claim propagated five times by citation
rather than by measurement, in files whose subject was that exact failure.
facts-check catches a copied number that disagrees with its source;
nothing catches a number with NO source, asserted once and repeated.
Registered F23, and all four evidence files carry an in-place correction
rather than a silent edit (ADR-0012 D5).
THE ACTUAL VERDICT: THE RATE IS WORKING. Eleven rolls at d8 -- declaration
1 opened the window at the old d4 -- against 1.375 eights expected, 1
observed. Not a shortage of evidence; the design.
BUT THE RETIREMENT CONDITION GENUINELY CANNOT FIRE, and that took
computing to see. "An override changes nothing twice running" needs a
consecutive pair at P=1/3 each, so ~12 overrides expected, at ~1.4 per
window: ~9 windows, roughly 100 declarations. A gate that cannot cash out
on any realistic horizon is decoration, which ADR-0006 D3 forbids.
Restated to be evaluated PER WINDOW: retire if a full window's overrides
all change nothing, met in two consecutive windows. A window with no
overrides is inconclusive and advances nothing. ~24 declarations rather
than ~100. Window 2 counts as the first; window 3 opens at d8 and decides.
Recorded and deliberately not acted on: declarations 5-9 rolled six five
times running, ~1 in 370 for some run of five in eleven rolls. shuf tested
over 200 rapid successive calls looks uniform, longest run three. Found
post hoc, which is how coincidences become findings, so it is logged for a
future window to check rather than treated as evidence.
InnerLoop.md then crossed the loadability limit, and so did
InnerLoopReference.md. The window log moved to specs/ChaosRollHistory.md:
it grows by one entry per window, and a log inside a reference eventually
crowds out the reference.
make all: exit 0.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
unplayed was ours
Tier S (a fix and a measurement inside a boundary; chaos d8=4 from
CB-WP-0029's roll, no override).
cb-play built EVERY game with ScoringMode::SharedGround and passed an
empty patch. The mode was settable in scenarios and not from the driver,
so two of the three shipped modes were unreachable from the only way
anyone actually plays. F14 sat open for a week because nobody could reach
the thing it was about.
--mode added. All three now play out and give DIFFERENT WINNERS FROM
IDENTICAL PLAY: shared -> all four seats (mastery 4), common -> P3 alone
(top personal scorer), coalitions -> P1+P2 (best Bond network, 4>3>2).
Same 37 commands, three answers.
AND THEY ANSWER F17'S OPEN QUESTION. I had flagged that ATTACK might earn
its place where Blame costs personal score. It does not, in any mode:
SHARED GROUND 132/165/190/200 -> identical free but pointless
COMMON PROBLEM 59/52/48/44 -> 59/52/48/34 a cost at six seats
BONDED COALITIONS 131/134/132/116 -> 59/52/48/34 roughly halved
The coalitions row has a mechanism and the data confirms it unprompted.
GR-A07 flips a Bond to a Rivalry on Attack, and GR-E04 scores Bond
NETWORKS -- so attacking destroys the thing that scores. And the attacking
numbers in E04 are IDENTICAL to E03's, which is exactly what that
predicts: break every Bond and each seat is a coalition of one, so GR-E04
degenerates into GR-E03. That check was not designed; it fell out.
F14 -> applied. F17 strengthened and no longer bounded to co-op: ATTACK
has no mode in which it helps, and one where it actively destroys your
score.
Still framed as a question rather than a verdict. DARVO is the pattern the
game is about not falling into, so a self-destructive ATTACK may be the
design. What ground-game has to decide is whether the namesake mechanic
being unreachable in competent play -- in all three modes -- is intended.
make all: exit 0.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
says
Five games per seat count, 2-6 players. NO ANOMALIES: every game reaches
5 rounds with an outcome, no stalls, no stress above the cap, no
over-claimed Problems.
But the series showed something a crash never would. DARVO NEVER FIRED IN
25 GAMES and stress never exceeded 2. Measured wider: GreedyPolicy plays
ATTACK exactly ZERO times in 10,000 selections across 500 games.
THAT NUMBER IS ABOUT OUR BOT, NOT THE GAME. bot.rs ranks
`Action::Attack => 10`, below everything. Reporting "the game gives no
incentive to attack" from a policy we programmed to rank attack last would
have been CB-WP-0025's C4 error committed again -- a single policy's
behaviour presented as the game's.
So the artifact varies exactly one number: ATTACK's rank in an otherwise
identical policy, 200 games per cell.
rank 10 (below all): 132/165/190/200/200 wins, 0 attacks, 0 DARVO
rank 75 (above SUPPORT): 132/165/190/200/200 wins, 315-923, 13-218
rank 95 (above SOLVE): 0/0/0/0/0 wins, 1400-5170, 400-1000
THE MIDDLE ROW IS THE FINDING. Identical win counts at every seat count,
while attacking hundreds of times and arming DARVO repeatedly. Attacking
is not punished -- it is INERT with respect to the goal. Group success is
a function of SOLVE alone, and ATTACK costs anything only when it ranks
above SOLVE and displaces it.
The maintainer was right and the reason is sharper than his phrasing:
there is no incentive because there is no PATH. ATTACK's effects (Stress,
Rivalry, DARVO) feed nothing that decides group_success.
Bounded honestly to SHARED GROUND. Blame costs PERSONAL score, so ATTACK
may earn its place in GR-E03 and GR-E04 -- which have never been played to
the end (F14), and that is where to ask next. And this is NOT a claim the
game is broken: DARVO is the pattern the game is about not falling into,
so a self-destructive ATTACK may be the design. The question for
ground-game is whether the namesake mechanic being unreachable in
competent co-op play is intended.
F17 promoted from note to raised, with games/ground/examples/attack-value.rs
as its reproduction. Register: 18 findings, 8 with a resolving
reproduction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
evaluate
CB-EV-0027.
The supply is never exceeded, and the link row explains why: GR-L01's two
slots per seat ARE the twelve link tokens, the same constraint written
twice. Nothing was designed to make that true; it fell out.
The Focus/Blame check was wrong and I caught it -- fifth instance of the
family CB-EV-0019 named (a number computed correctly about the wrong
subject) and the FIRST caught before leaving the repo, rather than by an
adversarial reviewer or the maintainer. That is one data point, not a
trend. What differed is that this pass read its own output and asked what
exactly it had compared. The family still has no control: facts-check
catches copies that disagree and nothing catches a correct computation
over the wrong subject.
The height limit went 460 -> 500, and the evidence argues why that is a
correction and not a concession. 460 was a number picked while fixing the
unplayable table -- no derivation. 500 has one: ~800px viewport less ~120
header and ~150 controls leaves ~530, and the version that broke dragging
declared 620. An underived limit is a measurement defect; replacing it
with a derived one is CB-WP-0021 T06's rule applied rather than evaded. A
later pass wanting 560 has to move the viewport budget, not the number.
CHAOS WINDOW 2 IS CLOSED. Twelve declarations, zero 8s, zero overrides.
Its retirement condition -- retire if an override changes nothing twice
running -- was untestable from the first declaration to the last, and
CB-EV-0024, 0025 and 0026 each said so; this is the fourth and final.
Window 1 at d4: 12 declarations, 2 overrides, both changed the outcome.
Window 2 at d8: 12, 0, none.
The verdict this supports is that d8 bought rarity by spending evidence --
stronger than "the rate is too low", because the rate was chosen without
asking what sample size the retirement condition needs. Recording that
changes how the loop constrains its own operation, which is a tier-M
trigger in its own right, so it is OUTSTANDING and named rather than
bundled here. Leaving it implicit is exactly how ground-game's ten rulings
went uncollected.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ADR-0016, Tokens.csv vendored, tracks and discs on the table, and a supply
audit that found nothing and says so.
T03 MEASURED BEFORE T01 DECIDED. 750 games, 2-6 seats, greedy and random:
Protection reaches 1 per seat and 2 on the table against a supply of 6;
Denied 3 of 5; relation links EXACTLY 12 OF 12 and never more; Focus/Blame
0 conflicts. The link row is the interesting one -- GR-L01's two-slot rule
IS the twelve-token supply written twice, which is the shape of a supply
needing no separate enforcement.
AND THE FIRST VERSION OF THE FOCUS/BLAME CHECK WAS WRONG. It compared a
seat's own placed Focus against its OWN blame_from -- but that list holds
OTHER players' discs, so they are different tokens. It reported 2
conflicts; corrected, it reports 0. Fifth instance of this project's
recurring defect, a number computed correctly about the wrong subject, and
the first caught before it left the repo rather than by a reviewer.
D2: a token is a VIEW, not a type. The aggregate gains no `Token` --
adding one would create a second source of truth for Stress, and the first
time they disagreed the bug would be invisible because both would look
internally consistent.
D3: quantity does NOT bind, and the reason is not the measurement. A
component limit the rules do not state is not a rule. Refusing a seventh
Protection token would enforce something nobody ruled -- CB-WP-0023's
error inverted: SOLVE was OFFERED where it could not act; this would
REFUSE where the rules allow. The check ships as a standing control, so a
future violation becomes a question for ground-game (does the box bound
the game, or do the rules?) rather than a bound the engine invented.
Registered as F22, withdrawn: a stated negative, because a survey that
finds nothing and leaves no trace cannot be told from one never run.
D4: Stress on a 0-5 track that turns red at 5 where DARVO arms, DARVO on
OFF/DENY/ATTACK/REVERSE, Freedom as the two-sided disc the edition says it
is, Protection and Blame counted, Lead and Round on the table.
Two tests broke on token discs and both were FIXTURE defects:
seat_centres matched every <circle> and track stops are circles. Seats now
carry class="seat".
The table height limit went 460 -> 500 as a CORRECTION, not a concession.
460 had no derivation; 500 does -- ~800px viewport less ~120 header and
~150 controls leaves ~530, and the version that broke dragging was 620.
CB-WP-0021 T06's rule is to fix the measurement rather than lower the
floor, and an underived number is a measurement defect.
make all: exit 0. 66 render tests, 26 cb-play.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Declaration 12 of chaos window 2, which CLOSES IT. Structural M (imports
another edition file under AM-4's budgets), chaos d8=4, no override.
Tokens.csv turns out to carry more than names. Every token has a
`quantity` -- 6 Protection, 5 Denied, 12 relation links -- and a component
count is a limit a simulation does not have unless it is told. protection
is saturating_add(1) at lib.rs:986 with no upper bound, against an edition
that ships six tokens for six players. T03 measures whether that matters
rather than assuming either way.
And `sides` is the second interesting column: Freedom and Focus/Blame are
each ONE DOUBLE-SIDED TOKEN, not two states. The engine holds Focus in a
map and Blame in a list; whether those can disagree with a single physical
token is worth checking.
T03 is shaped against a specific error: whatever it finds is a FINDING,
not a fix. A component limit the rules do not state is underdetermined and
belongs to ground-game, and the engine must not start enforcing a bound
nobody ruled. CB-WP-0023 exists because SOLVE was offered where it could
not act; inventing a limit is the same error in the other direction.
The workplan also names the window-2 verdict as DUE. Its retirement
condition was untestable throughout -- eleven declarations, zero 8s, zero
overrides -- and recording that is a change to how the loop constrains
itself, so it is its own tier-M pass rather than a task here. Named so it
is not lost between passes, which is how the ten rulings went uncollected.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Tier S (fixes inside a boundary; chaos d8=3, no override). Two
observations from play, both about what the page says has happened.
The controls moved BELOW the log: you read what happened, then decide what
to do next. They were above it, which asks for the decision first.
And sealing now marks the WHOLE PAGE inert, not just the controls. A
greyed-out button beside a full-colour table still reads as a live game
with one broken control; the session has ended and everything on screen is
a record of it. The status line stays legible on purpose -- it is the one
thing still worth reading.
THE STUB NEEDED A BODY classList TO MAKE THAT TESTABLE. Without it "the
session visibly ended" would have been a claim about CSS with nothing
checking it, which is precisely CB-WP-0016's finding: a stub too thin to
express a failure is how the failure survives. The harness reports the
sealed page through the status channel with a NUL-separated marker --
ugly, deliberate, and documented, because widening the return type would
touch every caller for one boolean.
Both directions asserted: a `closed` reply seals the page, an `ok:
dealing` reply does NOT -- otherwise the seal test would pass for a page
that greys itself whenever it is touched, breaking `play again`.
63 render tests, 26 cb-play, check and loop-lint clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Tier S (a fix inside a boundary; chaos d8=7 from the previous roll stands
for this continuation). Two observations from play that are ONE defect.
`play again`, `end session`, `pass` and the move buttons carried `.pick`,
which is cursor:grab. The stylesheet has .btn{cursor:pointer} BEFORE
.pick{cursor:grab}, so grab won.
A GRAB CURSOR INVITES A DRAG. A drag released over nothing posts nothing,
so the player picked up the button, let go, and the page did nothing. It
looked dead because the affordance told them to do the one thing that does
not work. Reported as two separate things -- "the button shows a hand to
pick up that it probably shouldn't" and "I can't start another game or
stop the server" -- and the first causes the second.
The click path itself was never broken: driving again->again and
done->done through the JS harness posts correctly. The logic was fine and
the invitation was wrong.
Click targets now carry `.tap` -- pointer cursor, same press affordance.
This extends CB-WP-0017's rule (interactive and inert must not look
identical) to: click and drag must not look identical either. The test
asserts both directions, because checking only that buttons lost `.pick`
would pass for a page with no affordances at all.
Registered F20 (applied) and F21.
F21 IS THE ONE I COULD NOT REPRODUCE: dragging did not work until after
the first note was saved. Ruled out the plausible mechanisms -- the
gesture logic posts correctly against the served page, the drag ghost
carries pointer-events:none so it cannot intercept the drop, and the
markup is identical before and after since the 303 re-renders the same
page from the same state. Remaining candidates are a <details> toggle
shifting layout mid-drag, a first-load timing difference, or browser-level
pointer capture. Reproducing it needs a browser, which no test here has --
the same gap F19 named. Recorded as unreproduced rather than given a
speculative fix.
And the fourth observation is confirmation, not a bug: "drawing my cards
from the deck is not implemented, I did not need to do that" is exactly
what CB-WP-0028 T04 determined and deliberately did not build. It is the
first evidence that importing the card text closed the comprehension gap
that produced the earlier click-the-deck request.
make all: exit 0. 62 render tests, 26 cb-play.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CB-WP-0028's overhead view was 620px tall, so the action cards sat a
screen below the Problems and dragging between them was physically
impossible. Seats were drawn inside the table; the `table` drop target was
a card among the buttons rather than the drawn surface.
make all passed throughout. Every test asserted the DOM was correct --
which it was -- and the JS harness posts a correct gesture against a page
a human cannot drag on.
Kind `degenerate`: it fires and collapses play. State `applied`: fixed the
same day, with height, seat-position and single-drop-zone proxies added.
They are PROXIES -- nothing here lays out a browser, and the gap CB-EV-0026
named a day earlier is unclosed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Tier S (fixes inside a boundary; chaos d8=7, no override). Four
observations from play, three of them caused by CB-WP-0028 -- and make all
was green for all of them.
THE NOTE BUG, WHICH I GOT WRONG TWICE. The first fix put the token in the
form's action, and that worked. But the 303 afterwards pointed at bare `/`
with no token, so the note WAS SAVED and then the browser followed a
redirect control 1 refuses. The player sees "no session token" for a note
that already landed. A redirect is a request the browser makes on your
behalf and is subject to every control the others are. I had tested the
POST and stopped there -- the same mistake as the first fix, one step
further along. Guard::page_path() now carries the token, and the test
asserts the redirect target is ADMITTED rather than merely non-empty.
WHY DRAGGING BROKE, WHICH WAS NOT THE DRAG. The gesture logic was fine: the
JS harness posts correctly against the served page, and all 14 drop
targets are present. The table was 620px tall, which pushed the action
cards a full screen below the Problems -- and you cannot drag between two
things that are never on screen together. Now 440, with a test asserting
the declared height stays under 460 and saying why. That is a proxy for a
browser layout, not the property itself, and the test says so.
Seats sat ON the ellipse: they were placed at 0.83 of the table radius,
which is inside it. Now outside, asserted numerically at 2 through 6 seats
against the ellipse equation rather than eyeballed.
And the `table` drop target was a separate CARD among the move buttons,
which is exactly why a player looking at a picture of a table could not
find anywhere to drop. The drawn ellipse is the drop zone now, and a test
asserts there is EXACTLY ONE table target and that it is the drawn one --
two elements claiming to be the table is worse than none.
The gap this exposes is the one CB-EV-0026 named a day earlier: every test
asserted the DOM was correct, and it was. Nothing asserted the page was
usable, and the drag test passes on a page you cannot physically drag on.
What is added here are proxies a browser-less test can check.
make all: exit 0. 61 render tests, 26 cb-play.
Verified over real HTTP rather than by inspection: note POST 303, the
redirect carries the token, following it returns 200, the table declares
440, one drop zone, seats at (380,421)/(113,112)/(647,112) against a table
of rx=200 ry=118, and both notes reached the trial log.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
turned out to be
T03. One table_svg: seats around an elliptical table starting at the
BOTTOM -- the viewer sits nearest the reader, as at a real table --
Problems and both stacks in the middle, each seat's played card between it
and the centre, relations drawn between seats.
Two renderers were DELETED: relations_svg and piles_svg. The task said one
table not two diagrams, and leaving the old ones would have meant drawing
the same thing twice and letting them drift.
No coverage probe cost, through a restructure that merged three diagrams
and removed two functions. Second confirmation of CB-WP-0027's finding: a
probe naming a FACT survives a reflow, one naming a PRESENTATION does not.
CB-WP-0024's "17 remaining" broke on a rendering change; this far larger
reflow broke nothing.
The new control is per seat count -- no two seat circles closer than 70px
at 2 through 6 -- asserted rather than eyeballed at three, which is the
only count anyone ever looks at.
T08 (CB-EV-0026). Seven of nine observations were engine defects, one was
a design finding, one was already true and nobody could tell.
Observations 4 and 5 both dissolved and had ONE cause: nothing on the page
said how drawing works, so a player built a mental model to fill the gap
and reported the gap as two feature requests.
The import gap was worse than "one of nineteen" -- 5 of 13 columns read
from the file we DID vendor, discarded at parse time for eight days. Rule
coverage was 59/59 throughout. The gate measures whether rules are
EXERCISED; nothing measures whether a player can READ the game, and
nothing cheaply could, which is why the person playing it is the
instrument.
TWO GATES WERE WRITTEN FOR A SMALLER WORLD, and neither was wrong when
written. edition-check compared one recorded digest against Problems.csv
regardless of which file it described -- correct with one vendored file,
comparing across files with four. And a cb-play test asserted the literal
"game over" and went red when a won game said "solved", which was T06
working; it now asserts the heading against the OUTCOME and covers the
no-outcome case the original never touched.
Chaos window 2 closes with zero overrides in eleven declarations at d8.
Third and final statement of it: d8 bought rarity by spending evidence,
and a mechanism producing no data across a full window cannot be evaluated
by that window.
make all: exit 0. 57 render tests, 26 cb-play, loop-lint clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
a feature deliberately not built
T04 BUILT NOTHING, WHICH IS THE CORRECT OUTCOME. "Click the draw stack to
get hand cards" is not a legal move: GroundCommand has no standalone draw,
and the edition's own INVESTIGATE text settles it -- "Choose one hidden,
non-Denied Problem and reveal it. THEN DRAW ONE SOLUTION." Drawing is a
consequence, never an action. Implementing click-to-draw would have
invented a rule, which is what CB-WP-0023 exists to stop.
And observation 5 is already true: drawing is ALREADY automatic, inside
INVESTIGATE, with no player input. An auto-draw option was asked for a
thing that has never been manual.
Both observations have one root cause and T02 fixed it. The maintainer
expected to take cards from the deck because NOTHING ON THE PAGE SAID HOW
DRAWING WORKS -- the INVESTIGATE card's own text was in a file we had not
imported. No finding raised: a player's instinct differing from a legible
rule is a comprehension gap, not a rules gap. Whether the instinct recurs
now that the text is present is a testable question and was not before.
T06: "game solved" on a win, "game over" on a loss, "the game stopped"
with no outcome. Asserted all three ways, because a test checking only the
win case passes for a page that always says solved. More than tone --
GROUND is co-operative and about repairing something, and arcade
failure-state vocabulary for a win tells a player the wrong thing about
what they just did.
T05: the ending page gets the table's two-column shape. Result left;
rankings, controls and the full log right. The seal still removes every
control wherever they now live.
T07's interesting decision was NOT TO RANK. Modes.csv defines
scoring_tiebreak per mode, so ordering is the GAME's where one exists --
"Lower combined Stress, then fewer Blame tokens" for coalitions. For
SHARED GROUND it says "Not applicable", because the table succeeds or
fails together. So co-op shows contributions and refuses to order them,
and says why. Drawing a leaderboard because a leaderboard is easy would
invent scoring the rules do not have. The one derived superlative is
labelled "clay-borg's reading, not a rule" and appears only where the mode
ranks; ties are shown as ties.
52 render tests pass. check clean, loop-lint clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ADR-0015 and the import. F18's fix: "I don't understand the GROUND card"
was never a design gap -- the card explains itself in the edition and we
never imported the explanation.
THE MEASUREMENT IS THE DECISION, and the gap is bigger than "one file of
nineteen". Of the file we DID vendor, the engine reads 5 of 13 columns:
title, problem_text, front_rules, reveal_effect and unresolved_effect were
discarded at parse time. The cheapest part of this pass costs no new bytes
and was sitting in the repo for eight days. And SCN_01 is hardcoded at
lib.rs:1824 -- the edition ships FOUR scenarios and the engine has never
dealt three of them. Nobody had said so.
ADR-0011's revisit condition is measurably absent, so the dependency
argument does not get re-run: across Actions, Solutions, Modes and
Scenarios there are ZERO doubled quotes and ZERO embedded newlines. The
hand reader's only job is comma-in-quoted-field, which it already did.
Refusing csv on a measurement rather than on a preference.
Vendored Actions, Solutions and Modes -- the text a player reads. Not the
production artifacts (BOM, Print_Manifest, Back_Designs, Symbols). NOT
Extensions.csv, which names content the designer placed outside the core;
importing it would break the claim that this engine plays the edition as
printed. It is now known to exist, which was the real risk.
One Table reader with four callers, because a per-file copy is how a
parser acquires four subtly different bugs. The GROUND card now shows
"Regulate. Restore the frame. Decide." with its GR/OU/ND text on demand;
Problems show their own titles where a priority number used to be.
The load-bearing test asserts the text is a SUBSTRING OF THE VENDORED
FILE rather than equal to a Rust literal -- a test comparing against a
hardcoded expectation would pass for a hand-copied string, which is the
drift this ends.
`edition` came out from behind #[cfg(feature = "scenarios")]. It was gated
because its only consumer was; the edition is the game's own data and the
shipped runtime now reads it. Test machinery and game content are
different things and only one of them is optional.
And edition-check was written for a single-file world: it compared the
first recorded digest against Problems.csv regardless of which file that
digest described. It now checks every file both ways -- a vendored file
with no digest fails, a digest naming an absent file fails -- and asserts
ADR-0015 D3's falsifier directly rather than trusting it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Nine observations from the maintainer's session. Two are about the game
and go to the register; seven are about the engine and are this workplan.
F18 IS THE ONE THAT REFRAMES THE PASS. "I don't understand the GROUND
card" reads as a design problem. It is not: Actions.csv carries that
card's own tagline -- "Regulate. Restore the frame. Decide." -- and its
full rules text, and clay-borg never imported it. We vendored ONE OF
NINETEEN edition files. Everything else the engine knows is a
hand-transcription into GroundRules.md's 59 numbered rules, which is
enough to PLAY the game and gives a player nothing to READ. The page shows
`Clarify` where the card says "Ask What Happened -- Invite a concrete
account before judging." Registered as `inert`: the data exists and cannot
fire, because nothing reads it.
Found by a player saying he did not understand something. Rule coverage is
59/59 and has been for weeks.
F17: no incentive to ATTACK while holding useful Solutions. Registered as
a NOTE, not a finding -- no artifact demonstrates it, and under GameDesign
§3.1 it may not go to ground-game until one exists. One is cheap (count
ATTACK selections across the policy panel against hand quality). Owner is
ground-game if it survives, since it would be a design finding.
The workplan (M, chaos d8=1, no override, declaration 11 of window 2)
carries the seven engine observations. Two tasks are deliberately shaped
against past mistakes:
T01 must decide whether ADR-0011's hand-rolled CSV reader survives
Solutions.csv, whose microcopy and rules_text are prose with commas and
quotes -- ADR-0011 named exactly that as its revisit condition, so if the
reader cannot parse them the dependency argument gets re-run rather than a
fragile parser written.
T04 must first establish whether "click the deck to draw" is a legal move
at all. GR-A01 draws as part of INVESTIGATE; the deck is not a thing a
player may take from. If it is not legal, that is a FINDING for
ground-game -- the maintainer expected an interaction the rules do not
offer -- and not a feature. CB-WP-0023 exists because SOLVE was offered
where it could not act.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Tier S (a fix inside a boundary; chaos d8=5, no override). Reported by the
maintainer: "I can't save notes, I get 'refused: no session token'."
The form posted to a bare `/note`. Control 1 requires the token on EVERY
request, so the guard refused all of them.
WHY THE TESTS MISSED IT IS THE PART WORTH RECORDING. I verified the note
channel over real HTTP and got 303 -- but I appended the token to the URL
by hand. I tested the ENDPOINT and not the PATH A PLAYER TAKES, so the one
thing standing between the feature and the user was the one thing not
exercised. Same family as timing the wrong span and counting the wrong
denominator: a correct measurement of the wrong subject.
Fixed with Guard::note_endpoint(), so the form's action carries the token
like every other request. The assertion now pins the token's PRESENCE
rather than the bare path, so reverting the fix turns it red.
Verified the way it should have been done first: read the form's `action`
out of the SERVED page and POST to exactly that, nothing added by hand.
303.
Clippy then flagged document_with_log at 8 arguments. It was right -- the
signature had grown across three passes -- so the two endpoints are now
one `Endpoints` struct rather than an #[allow]. They are one concept: the
guarded surface this page may talk to, one channel that becomes commands
and one that provably cannot.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Tier S (utility inside a boundary; chaos d8=5, no override). Provenance in
this message per InnerLoop §Loop tiers.
The maintainer asked how to start the UI. The answer was a five-flag
cb-play invocation nobody could be expected to remember, in a Makefile
whose own header calls itself "one command surface" -- a surface you have
to read the source of is not one.
make ground # play, recording a trial
make ground PLAYERS=2 SLUG=darvo-confusion
make trials # read the notes back
`ground` timestamps the trial name so two sessions in one day cannot
overwrite each other's notes, and keeps the date prefix because
tools/trials.py reads the age from there.
`make` with no target now lists targets instead of running the heaviest
thing in the file. It extracts the `## ` comments that were already the
convention -- ten of them existed and NOTHING HAD EVER RENDERED THEM.
Which is how the first run found a real defect: `## run all GROUND
scenarios through cb-sim` sat above `dep-weight:`, 167 lines from the
`sim:` it describes. A doc comment nothing reads rots, and this one had --
same family as the message that sat unread for four days and the ten
rulings nobody collected, in a third medium. Moved to `sim`, and
`dep-weight` given its own.
Documented the targets a person reaches for (coverage, status, all,
self-tests, sim, design, difficulty, trials) and listed the undocumented
instruments compactly, so nothing in the file is invisible.
Verified end to end rather than by inspection: `make ground` served a
page (200), a note posted through it (303), and `make trials` reported it
bound to its position.
loop-lint clean, check clean, self-tests 0 failures.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CB-EV-0025. T05 asks whether a comment actually got written during play,
unprompted. It has not been, and that is the pass's real acceptance test.
Everything built here is evidence the mechanism works; none of it is
evidence it gets used. I exercised it myself, which proves the plumbing
and proves nothing about the ergonomics -- the person who wrote the
feature is the worst possible witness to whether it is worth using.
Stated plainly rather than answered optimistically, because CB-WP-0022
built a register whose first run found ten answers nobody had collected:
this project's failures are about whether things get READ.
ADR-0007 D5 turned out to be the easy part. PointerFact::parse already
refuses any unrecognised field, so a comment could not reach the command
path even by accident. That is strictness at a boundary paying off years
early, and the clearest instance in this project of it being cheaper than
vigilance across one.
esc() held against its first hostile input, and the test asserts the
harder half: the player's words are still READABLE after escaping. An
escaper that dropped the text would pass "no script tag in the output"
while destroying the note.
The reflow cost no coverage probe -- all 41 pre-existing render tests
passed through a full restructure into two columns. CB-WP-0024 saw the
opposite when a probe was tied to a rendering ("17 remaining"), and the
contrast is the useful part: a probe that names a FACT survives a reflow,
a probe that names a PRESENTATION does not.
Running it changed the design for the third pass running, after
CB-WP-0024's role column and CB-WP-0025's K=2 infeasibility. The pattern
is consistent enough to state: the ADRs in this project are good at
deciding and poor at predicting, and the loop's value comes
disproportionately from the code loop rather than from steps 1-3.
Chaos window 2 closes with zero overrides in twelve declarations, so its
retirement condition is untestable -- final now rather than projected. The
window's verdict should be that d8 made the mechanism unevaluable.
Also recorded rather than dismissed: make all failed once at env-test and
passed on re-run. A cb-play server and several cargo processes were
running concurrently, which is the likely cause -- but "likely" is doing
work in that sentence and I did not chase it. A gate that fails
intermittently and is re-run until green is a gate being trained not to
matter.
Not built, and named: promoting a comment to a register note is still
manual.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The meta view beside the table, and a note channel that provably cannot
carry a move.
T01 (ADR-0014). ADR-0007 D5 is SCOPED, NOT AMENDED, and the reason it was
easy is that PointerFact::parse already refuses any unrecognised field --
a comment could not reach the command path even by accident. So /command
carries pointer facts, /note carries text, and Note has no code path to
GroundCommand. Comments live in trials/<date>-<slug>.md, not in
ScenarioFile: a scenario is executed, replayed and hashed, and prose in it
is data the runner must ignore, which is how a format rots. The state hash
binds; round and step are for reading. And the retention question, decided
before any comment was written: RAW NOTES NEVER LEAVE clay-borg. A note
reaches ground-game only by being promoted to a register finding, by a
human, with the wording chosen then -- "the DARVO sequence is infuriating"
is useful signal and a bad way to open a message to the game's designer.
T02. CSS grid, minmax(0,1fr) on both tracks -- load-bearing, because a
grid child defaults to min-content width and without it the SVG table
refuses to shrink and pushes the meta column off-screen, looking correct
on the developer's monitor and broken everywhere else. Single-column
fallback under 64rem. The running tally moved into the panel so it is
visible WHILE PLAYING; it only appeared on the ending page before, and a
score you see once the game is over informs nothing.
T03. A plain <form method="post">, so the box works with the script
disabled; the command channel needs JavaScript because a drag is not a
form submission, a comment is one. 303 See Other so a reload does not
re-post. esc()'s first hostile input: <script>alert(1)</script> renders
escaped AND STILL READABLE -- escaping that eats the player's words is its
own defect. Verified over real HTTP: note posted 303, hostile note stored
as text, empty note refused 400, game did not advance.
T04. tools/trials.py and make trials. THE REPORT'S DESIGN CHANGED BECAUSE
I RAN IT: the first version called any note without a recording an orphan,
so a live session reported every note as broken -- the recording is only
written at game end. A metric that cries wolf is one nobody reads, which
is the exact failure this pass exists to prevent. Now ok / pending /
orphan, and only orphan is a target-0 number. The self-test exercises the
REPORTING path, not just the parser, because design-baseline.py had a
green self-test and an unexercised reporting path and that is where it
rotted.
And a latent Makefile defect surfaced: make trials did nothing, because
trials is also a directory and Make saw an up-to-date file. design,
difficulty and trials -- added by CB-WP-0022, CB-WP-0025 and this pass --
were ALL missing from .PHONY; only the one that collided revealed it.
make all: exit 0. 49 render tests, 26 cb-play, loop-lint clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The maintainer's ask: game interface left, meta and commentary right, with
comments bound to game state so a test player's perception, confusion and
reasoning become reviewable signal.
Declared M (chaos d8=7, no override, declaration 10 of window 2). It
touches two canonical interfaces: ADR-0007 D5's input contract, which
currently admits only raw pointer facts, and the recorded-session format,
which is deny_unknown_fields -- CB-WP-0026 just proved that bites.
FRAMED AS GameDesign §5's TRIAL PROTOCOL MADE ERGONOMIC, because that is
what it is. The protocol already exists and nobody would use it as
written: it asks the player to reconstruct their reasoning afterwards,
from memory, in a sibling file. A comment typed while the position is on
screen is contemporaneous, bound to something replayable, and captures the
one class of signal no gate in this project can see -- confusion and
frustration.
The reason it is worth more than a comment box: GameDesign §3.1 makes a
note unreportable until it has a reproduction, and three findings are
stuck there now (F12, F14, F15). A comment bound to a recorded state, plus
the recording, is the cheapest reproduction a play-derived observation can
have. So the pass closes the register's input path for findings that come
from playing rather than from building.
The failure it must not reproduce is named up front: a pile of prose
nobody reads is this project's signature defect -- the four-day unread
message, the ten uncollected rulings. Storage is not the deliverable,
surfacing is, and T04 carries it.
T01 (ADR-0014) must decide where comments live, what the page may send
(D5's transport widens from two ids to arbitrary user text -- amended or
merely scoped?), what binds a comment to a position, and the retention
question this project has not faced: these are the maintainer's own words
about his own game, one will eventually be unflattering, and whether they
travel to ground-game needs deciding before any are written.
T03 notes that esc() is about to face its first hostile input -- until now
it has escaped suit names.
loop-lint: no findings.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
T06. games/ground/examples/difficulty.rs, make difficulty, wired into make
self-tests, and a report file in ground-game under GROUND-WP-0005 with a
hub message pointing at it.
THE REPORT OPENS WITH THE RETRACTION, because what this task was written
to send was withdrawn by T02 and GROUND-WP-0005 is blocked on exactly that
number. They are told, in the first section, that we nearly sent them "the
game is too easy at 5-6 seats" and why it was wrong.
seats winnable greedy random first-legal spread
2p 60% 60.0% 5.0% 76.7% 71.7
3p 93% 88.3% 6.7% 25.0% 81.7
4p 100% 93.3% 6.7% 30.0% 86.7
5p 100% 100.0% 3.3% 0.0% 100.0
6p 100% 100.0% 3.3% 0.0% 100.0
SPREAD justifies the whole redesign: 71.7 to 100.0 points between three
trivial policies. The table now shows why no single rate is a difficulty
rather than asserting it.
And the 5-6 rows point the OPPOSITE way from the withdrawn claim --
first-legal 0% against greedy 100% is the widest spread in the table,
which suggests play matters MORE there, not less. Neither reading is
established and the report says so.
The confound is stated in the tool's own output, not only in prose:
`winnable` is conditioned on greedy's play up to the final round, because
searching from round 1 is unaffordable. Presenting it as a property of the
deal would repeat this pass's error in a subtler form -- which is exactly
how a corrected project reintroduces a defect. NO THRESHOLD CHANGES ARE
PROPOSED.
The instrument can fail (spec §5): a witness must replay to a win, an
unwinnable position must report searched-out rather than a budget cut, a
one-node budget must not claim exhaustion, and the policy panel must
actually disagree. difficulty-baseline.rs marked superseded, kept as the
survey's dated snapshot. Registered as F16, inconsistent / withdrawn.
T07. evidence/CB-EV-0024. Five of nine defects came only from the review;
four from execution, and all four of those were in work written after it.
The wrong-denominator family now has five instances and still no control
-- facts-check catches copies that disagree, nothing catches a number
computed correctly against the wrong base. Tier L was an over-declaration
(no port, structurally M) and paid for itself anyway, because the review
is L-only. Chaos window 2 will close with zero overrides, making its
retirement condition untestable.
Named as open rather than implied done: the witness is NOT wired to the
ending page. The search works; the browser cannot ask it yet.
make all: exit 0. loop-lint clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
affordability projection
games/ground/src/search.rs, five tests. It finds real winning lines and
replays them through validate/fold to group_success.
Two bugs in my own work, found and fixed here.
THE TRAVERSAL WAS WRONG. It branched on "the first seat with any legal
command" and stopped there, so a later seat never acted if an earlier one
was already selected but still had a legal move. Restructured around what
the rules oblige: a seat without a selection MUST select (GR-R02) and
nothing else can happen first; after Reveal the optional actions branch
freely and the aggregate rejects Resolve until the obligatory ones are
done -- so the search needs no phase logic of its own.
AND MY REWIND WAS OFF BY ONE ROUND, replaying the round it was meant to
search. That is why the first run reported 3 nodes and looked like a
working search.
Measured with the real search, rewinding real games to the start of their
last K rounds:
2p K=1 exhausted, 8,103 nodes, ~29 ms
2p K=2 budget cut at 2,000,000 nodes, ~5 s
3p K=2 win found, 41 nodes, ~157 us
The spec's own falsifier said "§3 fails if K=2 proves unaffordable at four
seats". IT FAILED AT TWO. The projection assumed a joint product per
round; the search explores sequential per-seat decisions, so orderings
multiply the tree far beyond width^seats. That is the second projection
this pass published in place of a measurement -- C1's timer was the first.
THE ASYMMETRY IS THE OPERATIVE FINDING. Finding a win is cheap: DFS
stumbles onto one in tens of nodes. Proving none exists needs exhaustion.
So the witness feature is affordable now at any K a player would ask
about, and the winnable fraction (ADR-0013 D4) is NOT, because its
negative half must exhaust every deal it counts. K=1 is the honest default
for exhaustive answers today; making K=2 exhaustible needs transposition
or move-ordering, neither of which this pass built. specs §3 and §3.1
corrected accordingly, and the K=2 default withdrawn.
The negative control that makes "winnable" falsifiable: 2p seed 7 over its
last round returns NoneFound with exhausted=true in ~8k nodes -- a real
negative, not a budget cut wearing a verdict's clothes. And the visible/
hidden marking is tested both ways, since a marking that can only say YES
is decoration.
make all: exit 0. loop-lint clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
caught me repeating C1
specs/RetrospectiveAnalysis.md v1.0 plus games/ground/benches/search.rs,
which exists because ADR-0013 D7 refused to let the spec quote either
disputed figure.
THE BENCHMARK'S OWN FIRST FIXTURE WAS DEFECTIVE, and it is the same defect
class the review caught one layer up. Stopping at a fixed step 20 put 2p
and 4p in states where seat 0 had NO legal commands, so it timed an empty
Vec (~120 ns) and silently skipped validate_fold because there was nothing
to validate. It now advances until the seat has a real branch and ASSERTS
it. A clone benchmark was added too: a search must copy state per branch,
and iter_batched excludes setup from timing, so without it the budget
would again rest on an unmeasured span.
Measured at real decision points: legal_commands 4.06-4.76 us, clone
378-639 ns, validate+fold 0.5-3.8 us. Per-child cost is NOT uniform --
some commands resolve cascades -- so budgets use the upper end (~5
us/child).
That settles D3 with real numbers. Joint branching over the last two
rounds is ~5x10^2 / 1.6x10^5 / 5.7x10^5 at 2/3/4 seats, so K=2 costs
negligible / 0.8 s / 2.9 s and holds at two to four seats. It does NOT
hold at five or six, where the tool must reduce K and say that it did
rather than silently searching less.
§4.1 is a normative prohibition, not a preference: a single policy's win
rate MAY NOT be reported as a difficulty. The spec carries the measured
reason -- greedy 100% against first-legal 0% on identical deals -- because
this project already made that error and nearly exported it to a repo that
is blocked waiting on the number.
§2.3 makes the empty-result wording normative: "no winning line found in
the last K rounds", never "unwinnable". A bounded search cannot establish
unwinnability and that sentence is what a player who just lost reads.
Also corrected: the T01 completion record still asserted all three
withdrawn claims as fact. It now carries claimed / withdrawn / survives
explicitly rather than being rewritten -- a retraction that does not
propagate to every place the claim lives is how the earlier ones survived.
make all: exit 0. loop-lint clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
does not measure the bot
Seven decisions. Two are not what T03 expected, because the review moved
the ground under both.
D1: strategy fusion DOES NOT APPLY, and that is why the affordable option
is also the honest one. Fusion is a defect of aggregating over
determinizations to choose a move -- the search picking different actions
in states a player cannot distinguish. After the game there is ONE WORLD:
the deal is known, so a search over it yields a line executable in the
only world there is. The survey treated fusion as this pass's central
obstacle; it is an obstacle to a playing engine, which we are not
building. The tool answers "given the deal as it actually was, was there a
line that reached the threshold" and is labelled that way on screen --
never "how you should have played".
D4: difficulty is the WINNABLE FRACTION, not a bot's win rate. C4 killed
the bot rate -- two trivial policies span 0-100% on the same deals, and
improving the bot would make the game "easier" without a rule changing. A
measure that moves when the measurer improves is not measuring the thing.
The solver supplies the alternative: over N deals, in what proportion does
a winning line exist. That is a property of the deal distribution and the
threshold, which is what ground-game tunes and what GROUND-WP-0005 is
blocked on. Ships as a table -- winnable fraction, named reference policy,
skill gap -- never one number, with policy/N/seed-range/K in the number's
name. Stated as a LOWER BOUND, since a K-round search cannot see a line
that needed round 1.
D2: search GroundState. The survey's view-only structural boundary is not
implementable -- a view cannot fold events, and it said so in §6. The
guarantee moves to something checkable: every move in a witness is marked
visible or hidden, computed from project(). A witness reads "you could
have won, but two of these six moves needed a card you had no way to know
was coming" -- more useful than either extreme.
D3: bounded exhaustive over the last K rounds, table as one co-operative
agent, K=2 default. Affordable once C6 corrected the premise: joint
branching over the last two rounds is ~5x10^2 / 1.6x10^5 / 5.7x10^5 at
2/3/4 seats. Wording is normative: "no winning line found in the last K
rounds", never "unwinnable". PIMC and ISMCTS rejected -- they add strategy
fusion to a problem that does not have it.
D5: the harness becomes an instrument before any figure is quoted (C3) --
positive controls, --self-test in make self-tests, a make target, and a
PLURAL policy panel, because the spread between policies is what C4
exposed and hiding it would restore the error.
D6: no new crate and no port. The L declaration was an over-declaration
and is recorded as one. D7: the node cost is disputed 5x between author
and reviewer; T04 benchmarks it with criterion and neither existing figure
may be cited, including by this ADR.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
premise never left the repo
Separate agent, second tier-L review in this project. Six of seven
challenges conceded. The survey's headline finding is WITHDRAWN, not
softened.
C4 kills it, and the reviewer ranked it fourth. A FirstLegal policy --
take legal[0], no heuristic at all -- scores 0% at five and six seats
where GreedyPolicy scores 100%, and 77.5% at two seats where greedy scores
66%. Two unsophisticated agents span the entire range at the same seat
count. "The game is too easy at 5-6 seats" is therefore a statement about
GreedyPolicy, not about GROUND. The rescue the reviewer offered -- greedy
hits the 12-point ceiling in 200/200 deals, so the 6p row is a rules claim
-- dies on the same data: FirstLegal reaches that ceiling never.
C1: the per-node cost was wrong by 30-50x. The timer started before the
seed loop, so "us/node" included two setups, an entire greedy game and a
full validate+fold replay, divided by player-decision count. The tell was
in my own published output and I did not look at it: the figure FELL
(161/139/112) as branching ROSE (4.7/7.4/9.1), which no per-enumeration
cost can do. Re-measured with the clock around legal_commands alone:
3.0/3.5/4.1 us, now rising with branching. The reviewer measured 15.6-20.4
by a different isolation; we disagree by ~5x and neither has established
which is right, so T04 must benchmark it with criterion rather than adopt
either number.
C6: "exhaustive search is out at any seat count" is false -- ~3 seconds
over the last two rounds at 3p. With C1's correction the budget is
~10^5-10^6 nodes and bounded endgame search fits, so ADR-0013 cannot open
with "exhaustive is impossible, therefore determinized sampling" --
especially as sampling carries strategy fusion that exhaustive search does
not.
C3: the finding failed the admissibility rule this project wrote nine
hours earlier. 6/9/12 are sums where GROUND-WP-0004 T02 requires
per-priority rows, and the harness has no assertions, no --self-test and
no make target, so nothing can turn it red -- a `default` artifact wearing
a `counterexample` label, by CB-WP-0022 T05's own distinction.
C2: the ratio story explains nothing; 3p and 4p share deal, threshold and
ratio and differ by 12.5 points of win rate. C5: "explains the
maintainer's report" is contradicted by lib.rs:2487, which records his
losses as 3-player games on the pre-ruling deal, arithmetically unwinnable
at 6 against 7.
T06 exists to report to GROUND-WP-0005, which is BLOCKED waiting on a
difficulty baseline. Had this proceeded they would have been invited to
move thresholds on the strength of one bot's behaviour. That is the fifth
wrong premise this project would have sent them, and the second stopped by
an adversarial review rather than by a control. Both tier-L reviews here
have now caught a false headline that every gate passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CB-RES-0008 with a runnable baseline
(games/ground/examples/difficulty-baseline.rs), and the measurement
produced a finding before any solver exists.
A GREEDY BOT WINS 200 OF 200 GAMES AT FIVE AND SIX SEATS. Median margin
+3, 11.8-12.0 points available against a threshold of 9. The curve is
66% / 82.5% / 95% / 100% / 100% across 2/3/4/5/6 seats.
Row-level, as GROUND-WP-0004 T02 requires: available points 6 / 9 / 12
against thresholds 5 / 7 / 9, so the ratio RISES with seat count (1.20,
1.29, 1.33) while the table also gains actions per round to clear it with.
Three multipliers pointing the same direction.
It also explains the maintainer's report without needing a solver at all:
"I felt it was too easy but then we lost" is two true statements about
different seat counts.
Cost measured and it rules out the obvious approach. Branching is small
(mean 4.7-9.1) but legal_commands costs 112-161 us per call because it
filters candidates through full validate. Exhaustive search is out at
every seat count; 10^4-10^5 nodes is 1.4-14 seconds, which is the budget
the ADR must design inside.
Prior art names the trap: determinized search (PIMC) suffers strategy
fusion (Frank, Basin & Matsubara 1998) -- the search picks different
actions in states a real player cannot distinguish, so the witness may
require knowing what was on top of the deck. Such a line still replays
green, so the checkability benchmark does not catch it. Honesty and
checkability are different properties; stated explicitly so T03 cannot
conflate them.
The survey states its own most likely killer up front (§6): a view-only
search cannot fold events, so making the information boundary structural
rather than a promise may not be affordable. Better found here than in
T05.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
