clay-borg/workplans/CB-WP-0028-the-table-you-sit-at.md

id	kind	title	status	state_hub_workstream_id
CB-WP-0028	product	The table you sit at: the cards' own words, an overhead view, and a game you solve rather than survive	done	d37c8671-54e4-447f-af64-56f482483282

Purpose

structural tier  M   (imports more of the authoritative edition, which is
                      an external dataset under AM-4's budgets, and touches
                      the rendering the coverage gate is written against)
chaos            d8 = 1  → no override
declared tier    M

Declaration 11 of chaos window 2. Tier M: survey and decision merged; adversarial review optional.

Nine observations from play

The maintainer played with the new meta panel and reported nine things. Two are findings about the game, seven are about the engine, and the split matters because they go to different places.

#	observation	where it goes
1	no incentive to attack while holding positive cards	register — a game-design finding
2	"I don't understand the GROUND card"	here — and it is not a design gap (below)
3	overhead view: players around a table, stacks on it	T03
4	click the draw stack to take cards	T04
5	optional auto-draw	T04
6	play again / stop belong in the meta column	T05
7	so does the full log	T05
8	"Game over" is wrong when you won	T06
9	rankings — MVP, most problems solved	T07

Observation 2 is a data-import gap, and it reframes the pass

"I don't understand the GROUND card" reads as a design problem. It is not: Actions.csv carries that card's tagline ("Regulate. Restore the frame. Decide.") and its full rules text, and we vendored one file of nineteen. Everything else the engine knows is a hand-transcription into GroundRules.md's 59 rules — enough to play the game, and nothing to read. The page shows Clarify where the card says "Ask What Happened."

Found by a player saying he did not understand something. Rule coverage is 59/59 and has been for weeks. See F18 and ADR-0015.

Task: decide what else to import, and what it costs

id: CB-WP-0028-T01
status: done
priority: high
state_hub_task_id: "17e6f38c-1903-4bb2-aa20-37e76c8fc4e0"

decisions/ADR-0015-*.md (tier M merges survey and decision).

ADR-0011 is the precedent and it constrains this. It vendored Problems.csv with a checked digest and a ~50-line hand reader, and refused the csv crate on proportion — 17,651 lines against AM-4b's 19,742 of remaining headroom, 89% of the budget to read 20 rows.

Decide:

• which files, and the answer is not "all of them". Actions, Solutions and Modes carry text a player reads. BOM, Print_Manifest, Back_Designs, Symbols are production artifacts for a physical print run and have no business here.
• whether the hand reader survives contact. Solutions.csv has microcopy and rules_text — prose fields with commas, quotes and possibly embedded newlines. ADR-0011 named exactly this as the revisit condition: "nested quoting, embedded newlines, multiple dialects." If the reader cannot parse them correctly, say so and re-run the dependency argument rather than writing a fragile parser.
• what happens when text and rules disagree. GroundRules.md was derived by hand from these files. If a rules_text contradicts a numbered rule, that is a finding, and it goes to the register — not a quiet edit of either.
• Extensions.csv exists and names content deliberately outside the core. Decide whether it is imported at all; the answer is probably no, but "we did not know it existed" must not be the reason.

Done 2026-08-06. ADR-0015, six decisions.

The gap is bigger than "one file of nineteen", and the measurement is the decision. Of the file we did vendor the engine reads 5 of 13 columns; the discarded ones are player-facing text that had been in the repo for eight days. And SCN_01 is hardcoded — the edition ships four scenarios and the engine has never dealt three.

ADR-0011's revisit condition is measurably absent — zero doubled quotes, zero embedded newlines across every candidate file — so the dependency argument does not get re-run. Vendored Actions, Solutions, Modes. Not the production artifacts, and not Extensions.csv, which names deliberately non-core content; it is now known to exist, which was the real risk.

Task: the cards say what they do

id: CB-WP-0028-T02
status: done
priority: high
state_hub_task_id: "aa4453d2-a84e-4afc-8534-1feb835c4d5b"

Every card a player can see carries its own title, tagline and rules text, from the edition — not from a phrase we invented.

• action cards: GROUND, SOLVE, INVESTIGATE, SUPPORT, ATTACK;
• Solution cards: the title and microcopy, not just the suit;
• the mode in play, with its own description.

Controls:

• a test asserts the text comes from the dataset, not from a Rust literal — the whole point is that the game's own words reach the player, and a hand-copied string drifts from the source it copies;
• the drag affordance still works: the explanation is additional, and CB-WP-0020 T02 already found that replacing a label with an explanation loses the label;
• the page stays readable. Five action cards with full rules text is a wall; the tagline is the default and the rules text is on demand.

Done 2026-08-06. One Table reader with four callers (a per-file copy is how a parser acquires four subtly different bugs), CardText for Actions/Solutions/Modes, ProblemText for the columns already vendored. The GROUND card now explains itself; Problems show their own titles.

The load-bearing test asserts the text is a substring of the vendored file, not equal to a Rust literal — 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, but the edition is the game's own data and the runtime now reads it. Test machinery and game content are different things and only one is optional. edition-check, written for a single-file world, now checks every file both ways.

Task: an overhead view of a real table

id: CB-WP-0028-T03
status: done
priority: high
state_hub_task_id: "2ff677ac-80ff-4618-ad7c-05b62c8f80c8"

"Players sitting around the table, the draw and discard stacks on it. An overhead view of a game table would instantly help with understanding what is going on."

The seats become positions around a table rather than a row of cards; the Problems, draw and discard sit in the middle; each seat's played card appears in front of that seat.

This subsumes relations_svg, which already places seats on a circle for the relationship graph — so the layout exists and is drawn twice, once as a circle and once as a row. One table, not two diagrams.

Controls:

• the coverage gate still passes: every view field appears in the parsed document. CB-WP-0027 showed a reflow costs nothing when probes name facts; this is a bigger reflow and the same rule applies;
• the viewer's own seat is identifiable at a glance — a table where you cannot find yourself is worse than a list;
• 2 through 6 seats all lay out without overlap, asserted per seat count rather than eyeballed at 3.

Done 2026-08-06. 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.

The new control is per seat count: no two seat circles closer than 70px at 2, 3, 4, 5 or 6 seats — asserted rather than eyeballed at three, which is the only count anyone ever looks at. Plus: the viewer's seat is labelled and visually distinct, and a spectator is not given one.

Task: take cards from the stack, or let the table do it

id: CB-WP-0028-T04
status: done
priority: medium
state_hub_task_id: "1106c60e-112b-401f-b854-eeec566ef3d1"

Observations 4 and 5. Clicking the draw stack to take a card is the natural gesture; auto-draw is for when that stops being interesting.

Both are subject to a rule the engine already has. Drawing is not a free action — GR-A01 draws a Solution as part of INVESTIGATE, and the deck is not a thing a player may simply take from. So this task must first establish whether "click to draw" is a legal move at all, and if it is not, it is a finding for ground-game, not a feature: the maintainer expected an interaction the rules do not offer, and that gap is the signal.

Do not implement a draw the rules do not have. CB-WP-0023 exists because SOLVE was offered where it could not act.

Controls:

• the answer to "is clicking the deck a legal move?" is in the task record with the rule that settles it;
• auto-draw changes no outcome — a game played with it and without it from the same seed produces the same end-state hash, or it is not automation but a rules change.

Done 2026-08-06 — and NOTHING WAS BUILT, which is the correct outcome.

Clicking the deck 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. If no hidden Problems remain, draw one Solution only."

Drawing is a consequence of INVESTIGATE, never an action. Implementing a 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: it happens inside INVESTIGATE with no player input at all. The maintainer asked for an auto-draw option for a thing that has never been manual.

Both observations have the same root cause, and T02 fixed it. He 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. It now reads on the card.

No finding raised. The rules are clear and were simply not visible; a player's instinct differing from a legible rule is a comprehension gap, not a rules gap. Worth ground-game knowing as design feedback if the instinct recurs with the text present — which is now a testable question and was not before.

Task: the meta column takes the controls and the log

id: CB-WP-0028-T05
status: done
priority: medium
state_hub_task_id: "6c78e44e-38fc-451b-b1a0-205b1d079ca0"

Observations 6 and 7. play again and end session move to the meta column, and the ending page gets the same two-column shape as the table — it currently has none.

The full log belongs there too, which CB-WP-0027 did for the live page and not for the ending page.

Controls:

• CB-WP-0024 T01's seal still works: acknowledging the end must still remove every control, wherever they now live;
• the ending page's existing tests pass unchanged or the change is a regression.

Done 2026-08-06. The ending page gets the table's two-column shape: result on the left, rankings + controls + the full log on the right. The seal still removes every control wherever it now lives, and the existing ending tests passed unchanged.

Task: a game you solve

id: CB-WP-0028-T06
status: done
priority: medium
state_hub_task_id: "df0a967e-636b-4fe5-9926-4d73030a87d7"

"'Game Over' is negative and should only be used for lost games. If the players won, let's use 'Game solved'."

Right, and it is more than tone. GROUND is a co-operative game about repairing a situation; group_success means the table solved something. "Game over" is arcade vocabulary for a failure state, and using it for a win tells the player the wrong thing about what they just did.

Controls:

• the heading is a function of outcome.group_success, asserted both ways — a test that only checks the win case passes for a page that always says "solved";
• the no-outcome case (a game that ended badly) keeps its own wording and claims neither.

Done 2026-08-06. game solved on a win, game over on a loss, the game stopped when there is no outcome — asserted all three ways, because a test checking only the win case passes for a page that always says "solved".

Task: who did what — rankings that are not invented

id: CB-WP-0028-T07
status: done
priority: medium
state_hub_task_id: "ad6e6f6f-e531-494c-b8a9-18dfdd14681c"

"A ranking of players in the final stats. 'Most valuable player', 'most problems solved' and other conditions that might apply."

OutcomeView already carries personal, winners, coalitions and mastery, and the aggregate knows who claimed each Problem.

The risk is inventing scoring the game does not have. "Most problems solved" is countable from claimed_by and is a fact. "Most valuable player" is a judgement, and any formula for it is a rule we made up.

So: report what the game counts, and mark anything derived as ours. A superlative computed from real data is fine; a superlative presented as if the rules defined it is the same defect as a provisional default silently canonised.

Controls:

• every ranking names its source: a rule id, or "clay-borg's reading";
• ties are handled and shown as ties, not broken arbitrarily — Modes.csv has a scoring_tiebreak column, so the game may already say how;
• a ranking that no seat leads (nobody solved anything) renders as that, not as an empty list.

Done 2026-08-06, and the interesting decision was not to rank.

Modes.csv gives scoring_tiebreak per mode, so the ordering is the game's own where it defines one — "Lower combined Stress, then fewer Blame tokens" for coalitions. For SHARED GROUND it says "Not applicable", because the table succeeds or fails together.

So in co-op the page shows contributions and refuses to order them, and says why. Drawing a leaderboard because a leaderboard is easy would have invented scoring the rules do not have — the same defect as canonising a provisional default.

Facts the game counts (Problems claimed, personal score, Stress, Blame) are shown plainly. The one derived superlative is labelled "clay-borg's reading, not a rule" and appears only where the mode ranks at all. Ties are shown as ties.

Task: evidence

id: CB-WP-0028-T08
status: done
priority: medium
state_hub_task_id: "a20a529e-0edf-49cc-9c80-7dba3b76cff5"

evidence/CB-EV-0026-*.md.

• Whether importing the card text changed what the maintainer understood — observation 2 is the acceptance test and it has a person attached to it.
• Whether any rules_text contradicted GroundRules.md. Nineteen files were hand-transcribed into 59 rules; if the import surfaces a disagreement, that is the most valuable output of this pass.
• Whether "click the deck to draw" was legal, and if not, what went to ground-game.
• What the overhead view cost the coverage gate, against CB-WP-0027's finding that a probe naming a fact survives a reflow.
• Quote CB-WP-0027's cost by re-running the instrument.

Done 2026-08-06. CB-EV-0026.

• Seven of nine observations were engine defects, one was a design finding, and 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, and nothing measures whether a player can read the game.
• Two gates were written for a smaller world — edition-check compared one digest across four files, and a cb-play test asserted the literal "game over" and went red when a won game said "solved". Neither was wrong when written.
• Chaos window 2 closes with zero overrides in eleven declarations. Third and final statement: d8 bought rarity by spending evidence, and a mechanism producing no data across a full window cannot be evaluated by it.