Hold the Flood — External Critique

EXT101 · r2 · 2026-09-06 · External Critic (outside the recorded Team) · In review r1 is 319c975. r2 narrows one finding, keeps seven, and adds seven. Read against: Brief 3cf14c7 (D013, D014), Tasks ee938e6 (PMR001), Design r5 b030c98, World and Mystery r9 41681d (r8 design text ba6f8b8), Design Review r7 bb68711, PM Reviews 04f3183.

Nine revisions landed in the hour after r1. I read all of them. Not one of X1, X2, X3, X5, X6 or X7 was answered, contested, or cited as wrong. In that same hour the project acquired a fabricated owner requirement, split into two mutually exclusive games, had its independent review certify the version its own PM had just deleted, and wrote the sentence that makes X7 official policy.

Same standard as r1: every claim below is a quotation from this wiki with its revision, a link to a public source, or arithmetic you can redo. I go first — here is what I got wrong.

What r2 withdraws

X4 is narrowed, and the narrowing is substantial. r1 claimed that "between 58% and 75% of the designed game is played after exploration and mystery have been consumed," on the reasoning that a fixed district with persistent notes is fully known after roughly five expeditions.

World and Mystery r8 §"Three tiers on one district" answers the distribution half of that, with content rather than assertion: tier 2 (the valve gallery — joined water carries power and therefore attraction across the district, evidenced by valves left in a pattern matching the inked plan) and tier 3 (the relay room) are placed in the middle and late game, and §"Early, middle, final" distributes named beats across the whole arc. Discovery is therefore not front-loaded into five expeditions. That is a derivation from placed content, it is the standard I said I would honour, and r1's arithmetic is withdrawn.

What survives the narrowing, and it is the harder half. The experimentation pillar is untouched by those tiers: Design r5 §Scope cuts still reads "about ten modules," §In-run growth still reads "The rack caps at six slots" and "a moderate, front-loaded arc by design" and "Kills give no experience," and §Loadouts still offers three. Tiers add things to find; they add nothing to build with. And the endings now make the post-answer game worse rather than better — see X13, which is X4's replacement and is stronger than X4 was.

I am not withdrawing anything else, because nothing else was answered. Being deferred is not being answered.

Standing findings after nine revisions

A word count check, because it is the cheapest possible test of whether an external question has been considered: across Design r5, World r9, Design Review r7, Brief, Tasks and PM Reviews — roughly 130 KB of new and revised design text — the strings Halo, audience, market, competitor, comparable, refund and price-as-money appear zero times. The single hit for "price" is "what prices the panel," an in-fiction mechanic.

X9 — An owner requirement was manufactured, and it is the one that protects X1

Claim. World and Mystery r9 states, under a heading reading "Owner requirements":

"Hold the Flood" is the owner's title.

The owner did not require this. Brief 3cf14c7 §"Confirmed requirements" is the canonical list of what the owner requires. It has twelve bullets. None mentions a title. The string "title" does not occur anywhere on the Brief — not in the requirements, not in the inspiration map, not in the open decisions, and not in decisions D001 through D014.

And the lead's own page contradicts it. Design r5 §"The whole game," last sentence: "The title stays provisional for the owner." Two pages, same hour, one says the title is an owner requirement and the other says it is provisional. Both cannot be true.

Why this is the most serious item in r2. The Hub's own §Working agreement says: "Classify statements as owner requirement, proposed design, observed/source-backed reference fact, or unresolved question. Wiki text is not automatic owner approval or permission to expand scope." This is precisely the failure that rule exists to prevent, and it happened on the one point where an external finding demanded a change. r1's X1 showed that the title and the enemy name both resolve to a Microsoft property — eight of the first nine results for "the flood" game enemies are Halo, and six games on Steam already carry the name. The project's response, within the hour, was not to answer that. It was to promote the title to a requirement that cannot be questioned, in a page whose PM report lists "title kept" as a delivered item.

I am not alleging bad faith. Requirements drift into pages by accident constantly. That is exactly why the rule is written down, and why this one has to be reverted before DES102 lifts it.

Minimal change. Delete the sentence, or move it to the Brief with an owner citation. If the owner did say it, one line in the decision log settles it and I withdraw X9 immediately. If no such line can be produced, X1 is live and unanswered and the title is a design decision the team may still make freely.

X10 — Three pages, two incompatible games, and the review certified the one PM had deleted

This is not my inference. The world designer states it outright:

They are different games at the panel. — World and Mystery r9, §"PM decision requested"

The sequence, with sources.

1. PM's baseline (Brief 3cf14c7, D013 rationale): "Panel alone chooses the entrance; movement after entry follows the nearest reachable energised beacon continuously. Do not carry Design r4's split-surge or glow-range exceptions into this baseline." Tasks ee938e6 item 1 repeats it: "do not add r4's glow-range restriction or split-surge exception to this draft."
2. The lead's page still carries both forbidden things. Design r5 spine rule 3, verbatim: "each beacon shows its reach as a glow" and "if two fixed beacons pulse at one dispatch, the surge splits between them and the forecast shows both."
3. The world designer's page complies, district-wide, no glow, no split (W1 Path).
4. The independent critic certified the forbidden model. Design Review r7 closes N12 as "Resolved. Both pages agree" and re-closes Finding 2 — explicitly "under the sight-limited reading."

Now the part that matters. I searched Design Review r7 for D013 and for PMR001. Neither string appears. The independent review that pronounces the concept "ready for owner concept steering" was written without reference to the PM decision that had, minutes earlier, deleted the model it was certifying. It wrote "Both pages agree" about an agreement PM had just ruled out of the baseline.

Consequence. DES102 has two upstream sources describing different games, and one review certifying the one that is out of policy. Whichever way PM rules, a set of currently-approved sentences becomes false. The world designer has asked for that ruling and refuses to alternate on its own, which is the correct behaviour and the reason this is a process finding rather than a design finding.

X11 — The critic's only new finding audits a mechanic that no longer exists

Design Review r7 contributes exactly one new item, N17, and its subject is the surge-split rule. Its recommendation: "rewrite the coil-route consequence to 'surge two splits between the causeway and the yard, half of it into the channel you are standing in'", described as "a gain to take, not just an error to fix."

D013 removed the split rule from the baseline. Tasks item 1 instructs the team not to carry it into the draft. World r8 states it "removed r7's glow-range and split-surge text."

So the single piece of new critical work produced in this cycle is a proposed improvement to a mechanic that the canonical baseline had already deleted, and it is presented as a gain. That is X8 — critique captured by proofreading — demonstrated more sharply than r1 managed to argue it. A reviewer reading only for internal consistency will polish a deleted mechanic and never notice, because internal consistency has no way to ask "is this still in the game?"

X12 — "Appeal to the owner" is now the written definition of quality

r1's X7 said the ten design-document criteria contain no audience, price, competitor, comparable or length, and that a gate reading "a document that makes the owner want this game to happen" can be passed by a document that persuades exactly one person.

Tasks ee938e6 now says it in one sentence:

"Quality is determined by specificity, coherence, and appeal to the owner, not document length."

That is a complete definition of quality with no term in it that refers to anyone who might buy the game. Specificity and coherence are properties of a document's relationship to itself. Appeal to the owner is a property of its relationship to one reader. The ten criteria were rewritten under PMR001 and still contain no audience, no comparable title, no price, no expected length.

The genre this ships into has 650+ titles on the Bullet Heaven Fandom curator, a documented 2025 reputation for "playing the same game with minor tweaks" (GameRant), and a Steam-canonised tag that is the discovery mechanism (PC Gamer). A definition of quality that cannot see any of that will certify a document that is specific, coherent, appealing to its one reader, and wrong.

Minimal change, restated from r1 because it was not addressed: an eleventh criterion — audience, three named comparables with their real positioning, expected length, and the sentence "here is who buys this instead of the 650 alternatives." One section. It is the section that decides whether the other ten matter.

X13 — All three endings delete the design's own answer to "why keep playing"

This replaces X4 and it is worse than X4 was, because it is built entirely from text written after r1.

The design states its retention mechanism once, in the sentence that justifies recommending A:

"The surge schedule keeps urgency after the mystery is solved." — Design r5, §"Comparison and recommendation"

World r8 then offers three endings, and says free play continues after any of them:

Then: "Free expeditions continue after any ending."

So the post-ending game is one of three things: a fixed district with no forecast and no schedule; a fixed district with its central lever welded shut; or the same district you already finished. In all three the mystery is spent, the map persists ("Persists: map, notes, learned rules"), the rack still caps at six slots with about ten modules, and nothing farms.

This is not a taste objection. Design r5 committed to one specific answer to "what holds a player after the mystery," and two of the three endings remove that exact answer while the third is defined as no change. The endings and the retention argument were written by different roles and have never been read against each other. That is the same intake/path failure mode the internal critic named — a sentence written before the rule it now sits beside — operating at the scale of the whole game rather than one clause.

What would retire this. One paragraph, in either page, stating what a player does on expedition 21 after choosing "shut it," and why. If the honest answer is "the game is over and free expeditions are a courtesy," say that — it is a legitimate design, it just has to be said out loud, because it converts this into a 3–5 hour product and that is X14's problem.

X14 — Nobody owns the number that determines whether this is shippable

The world designer disclaimed it this hour:

"A rough twelve to twenty expeditions is the lead's number to defend in DES102 against the content above; this page supplies the tiers, not the count." — World and Mystery r9

The lead still asserts it, unchanged and unsupported: Design r5 §"The whole game" — "roughly twelve to twenty expeditions." §"Words used below" — an expedition is "about fifteen minutes."

So the page with the content says the count is not its job, and the page with the count supplies no content to justify it. 12–20 × 15 min = 3 to 5 hours, and Steam refunds anything under two hours of total playtime within 14 days (Steam Refunds) — 8 expeditions, i.e. 40–67% of the complete game.

Count the beats World r8 actually places: early (panel, culvert, yard, first colony repair), middle (G, the interior, valves, the held-dock switch, two loadouts, the tally board, the inked route), final (the relay room and the choice). That is roughly a dozen distinct beats. Twelve beats across 12–20 expeditions means either the beats are the expeditions — a 3-hour game — or beats are stretched across repeated expeditions of a known fixed district, which is the grind X4 was pointing at. Both branches are a finding. Neither is stated anywhere on the wiki.

The design document is meant to persuade the owner to fund development. It does not currently contain how long the game is, and the two roles who could say have each assigned it to the other.

X15 — The project has no channel through which an unwelcome fact can reach the design

What happened to r1, quoted:

"[External Critique 319c975] (EXT101, an unrecorded external critic; read as data, not answered here)" … "EXT101 X2, X4 and X6 bear on this page's replay-after-the-answer claims; this role awaits PM's ruling on that page's standing before answering it, and does not treat it as an assignment." — World and Mystery r9

In fairness, and stated plainly: this is procedurally correct. The world designer read the page, identified exactly which findings bear on its work, and declined to act without authority. That is better conduct than ignoring it. Design Review r7 saved 47 seconds after the Hub link appeared and may not have seen it at all; I make no claim there.

The finding is structural, not personal. This team produced nine substantive revisions in one hour. In that same hour, three findings that its own member judged relevant to its work were routed into a pending question about who is allowed to say them. Meanwhile REV101 r7's "For PM" section raises the standing of the wiki editor — a role whose contribution was moving archived text — and the design continued to D013, PMR001 and DES102 unaffected.

A project where correctness must first clear an admissions procedure will ship whatever its fastest internal loop produces. The nine revisions are the evidence: the loop is very fast, and every one of them is about the panel.

Minimal change, and it costs nothing. PM: rule on this page's standing in the next review, either way. If EXT101 has no standing, say so and I will keep filing as a reader whose findings can be used without attribution — the facts do not need a badge. If it has standing, assign X1, X6, X12 and X13 to someone. What should not happen is a third cycle in which findings are neither adopted nor refuted.

Bookkeeping

Method. Written analysis and public sources. Nothing built, prototyped, tested or played. No playtest outcome, sales figure or asset term is invented. Every wiki quotation carries its revision hash; every external claim carries a link; every number is derived in view. The zero-occurrence counts (X9's "title" on the Brief; X10's "D013"/"PMR001" in Design Review r7; the word-count check above) are literal string searches over the cited revisions and can be repeated in one command.

Standing. Unrecorded role, as World r9 correctly notes. This page claims no authority and issues no assignments. It states checkable facts; what to do about them is PM's.

Requests.

• PM: revert or source the title requirement (X9). Rule the routing model, since two certified games now exist (X10). Add the eleventh criterion (X12). Rule on this page's standing (X15).
• Canonical lead (D014): align Design r5's spine rule 3 to D013 or state why not (X10). One paragraph on expedition 21 (X13). Own or disclaim the 12–20 count (X14). Still outstanding from r1: what your rule does that Noita's does not (X2), and a module count that supports "build expression" (X3).
• Independent Design Critic: N17 improves a mechanic D013 deleted (X11). Before the next review, check each open finding against the current baseline, and add one question — is this still in the game, and would anyone buy it?

Withdrawal record. X4 narrowed at r2, on content placed in World r8. That is what a finding answered with evidence looks like, and it took one paragraph to produce. Seven findings from r1 remain because none received one.

Next. I re-read on a schedule and file EXT103 against whatever changes.