Hold the Flood — External Critique

EXT101 closed · EXT102 delivered · r6 · 2026-09-06 · External Critic (seated by owner, Brief D015) · In review Previous: r5 6a92c7 · r4 a8883a · r3 b3a54d · r2 280e75 · r1 319c975. Read against Design Document r7 / DES103 r1, Tasks PMU002, Brief D015–D018.

PM refuted X19 and PM is right. It is retracted, not withdrawn. I also made a process error worth recording: r5 audited Design Document r5 when r6 was already current, and r6 had answered X19's original form two minutes before I filed. I re-read my own page before saving, as the conventions require, and did not re-read the document I was criticising. That is my mistake and the convention I will follow from here is: re-read the target, not just the destination.

The lesson applied immediately: while composing this revision the target moved again, to DES103 r1 (document r7). I re-read it before saving. Both surviving findings are present in r7, verbatim, and one of them got heavier: the arc module has moved from optional into the minimum scope. EXT102 is delivered below, under criterion 11 (Brief D018).

X19 — retracted in full

PM's disposition (Tasks, PMU002): "the commercial conclusion is not established by the presented comparison… REF102 measures playtime at review from recent reviewers; the design states an untested completion-length target. Those are different measures and populations."

Accepted, on all three counts.

1. The measure mismatch is fatal and it was mine. Playtime-at-review is accumulated hours across a reviewer's whole ownership, replay included. A completion-length target is time to first finish. I compared one against the other and called the result a gap. Vampire Survivors' 22.0 h is not "how long it takes to finish Vampire Survivors."
2. "Order of magnitude" was arithmetically wrong. PM's number is correct: the lowest reported median over the proposed range is 9.3 ÷ 4 = 2.3 and 9.3 ÷ 3 = 3.1. I wrote "an order of magnitude" at r3 and r4 and "three to fifteen times longer" at r5. I repeated REF102's phrase without checking it, which is the same failure as r1's Bullet Heaven error: a secondary source taken on trust.
3. Eleven games chosen because they are the owner's inspirations are not price bands. PM: "prices of selected examples are not compulsory price bands for every competing game." Correct. They are a convenience sample of eleven, and I presented them as a constraint on the shelf.

Consequently the second district does not follow as a correction, and I withdraw the implication that it did. The commercial concern stays open, which is what PM ruled; the conclusion I drew from it is gone.

Running score: filed 22, retracted or withdrawn 19, narrowed 4. Two live.

X21 — live, re-verified against DES103 r1 (r7)

Both sentences are present in r7, unchanged (§4 clause 4 and §5 rack rules):

§4, clause 4 — "Motes fade in seconds; walking over one fills a battery in your rack; with no battery, the charge is lost."

§5, Rack rules — "A battery holds ten units. A discharge of the coil spends the whole battery. A kill mote is one unit; a stun sheds one large mote of five units per ten organisms stunned."

If a kill mote is one unit and a battery holds ten, walking over one does not fill it. If it does fill it, one kill funds a full coil discharge and four other systems stop meaning anything: the siphon (three units), the leak (one unit a minute per connected bundle), §5's own failed-arrangement teaching example ("the battery is empty in a little over three minutes"), and §6's finite-charge bound. §4's clause is inherited near-verbatim from DES101, written before units existed. Minimal change: "walking over one adds its charge to a battery in your rack."

X22 — live, re-verified against DES103 r1 (r7), and now heavier

§6 (r7, unchanged): "the electrician's loop, harvest a surge to fund the next discharge, pays only against a fresh surge standing in water." The printed electrician rack, C C B / S S I, holds one battery — §5 confirms "four of six slots on gear."

Spend 10, recover at most 10; motes "fade in seconds" and "with no battery, the charge is lost." Break-even, not paying. §6's own "two full batteries" needs two battery slots; the profitable rack C C B / B S I carries one bundle against the hauler's six — a real and interesting tension the document does not know it has.

The arc's bound is separately unsupported. §6: "bounded by the finite-charge rule so it cannot be sustained." The arc drains a battery over about twenty seconds = 0.5 units/second. A kill mote is one unit; the cutter kills at "about three hits… at roughly two hits a second" = 0.67 kills/second. Base output funds the arc with 34% to spare before chaining multiplies it. The binding constraint is the number of organisms — each "sheds once per expedition" — not charge. This matters more in r7 than it did in r6. The arc has moved from the owner's menu into the minimum scope — §6: "that is why the arc is in the minimum and not on a menu" — so its stated balance bound is now part of the recommended design rather than an option the owner might decline. The one module that makes this a survivorlike is bounded by a rule the document's own rates do not support.

Neither finding needs an outside source and neither offers a balance opinion. The stated rates do not produce the stated outcomes.

EXT102 — The product case, under criterion 11

Assignment: one bounded critique of the combined product proposition, with the strongest contrary case and the smallest credible correction. PM also asked EXT102 to report what the sample can and cannot support.

1. What the available data can and cannot support

Can (observed, re-queried by me against Steam's public endpoints today): list prices and discounts; review totals and score bands; Steam's refund terms as written. I independently re-ran six of REF102's price rows at r4 and all six matched exactly.

Cannot:

• Completion length. HowLongToBeat returns HTTP 403 from this environment — I tested it, confirming REF102's report. The measure that would settle length questions is unavailable, and no proxy should be substituted for it. That includes playtime-at-review.
• Sales, revenue or refund rates. Steam publishes none of these. Review counts are not sales.
• Price bands as constraints. A convenience sample shows what eleven named games chose, not what a twelfth must charge.
• Any minimum-length conclusion. Nothing available establishes one.

Everything below is labelled observed, predicted or testable accordingly.

2. The intended player

The document proposes (§1): "players of automatic-weapon crowd games who want to think, and players of exploration mysteries who will accept pressure." That is two audiences joined by "and", which is the weakest form of an audience statement — it describes who might tolerate the game rather than who is underserved without it.

Sharper, and offered as a hypothesis, not a finding. The player is someone who liked a crowd game for its route decisions rather than its power curve — the Deep Rock Galactic: Survivor player who found the extraction timer the interesting part, not the part to be endured — and who liked Noita for learning what the rule does rather than for the chaos. That is a real and identifiable taste. It is also narrower than either parent audience, and the document should say so rather than claim both.

Testable: whether a player who bounces off the mowing payoff stays for a routing decision is exactly the thing the arc module (§6) is a hedge against, and it is answerable in a later authorised phase with a single build, two loadout sets and no market research at all.

3. Why that player might choose it

The distinctive claim is real and I have never disputed it: one rule with four faces — what fights, what opens, where the mass goes, how the pack is packed — with the district fixed so the rule can be learned rather than survived. §1's differentiator paragraph makes the argument properly, concedes Noita by name, and does not overclaim. That is the product's actual asset and it should lead every owner-facing sentence.

4. The strongest contrary case — against my own critique

I spent four revisions implying that three to four hours is a commercial problem. The observed record says short authored games are a well-established, well-priced category. Queried today against Steam's public endpoints:

Limitations, stated plainly and applying to my table exactly as PM applied them to REF102's. These are not survivorlikes; they are a different genre neighbourhood and the owner requires a survivorlike. They are a convenience sample I selected, not a survey. Review counts are not sales. Release dates span fifteen years of different storefront conditions. And critically: their short lengths are widely reported but I could not verify them here, because HowLongToBeat is 403 — so treat the prices and reviews as measured and the "short" characterisation as unverified.

What it does support: the price range \(9.99–\)24.99 is demonstrably live for authored single-playthrough experiences, and every one of these carries more reviews than Blue Prince at $29.99. Length is not the binding constraint I implied it was.

5. What section 17 already does, and the one thing it is missing

Read r7 before reading me. §17 is now a single recommendation rather than a menu, it states a positioning hypothesis, and it separates observed from predicted with every limitation PM asked for, including "playtime at review is not completion length, so these figures cannot set a minimum length for this game, and the price of eleven chosen examples is not a rule for every game" and "content is not to be added merely to sit beyond a refund threshold." That is criterion 11 met. I am not going to ask for it again, and I am recording that three of the four things I was going to recommend were delivered before I filed.

The one thing genuinely missing is a second observed comparable set, and it bears directly on the hypothesis §17 states. §17 proposes positioning "as a short authored game on the shelf your crowd-game references sit on, at their price band" — the crowd band, about $5 to $13. That band was measured from eleven games selected because they are the owner's mechanical inspirations. It was never measured from games selected because they are short and authored, which is what §17 calls this game in the same sentence.

Queried today, that set prices differently (§4 above): $9.99 to $24.99, with Portal at $9.99/99% and 198,905 reviews, INSIDE and Stanley Parable at $24.99, Firewatch and Edith Finch at $19.99. Every one carries more reviews than Blue Prince at $29.99.

Observed: two different comparable sets, chosen on two different criteria, give two different price ranges that overlap only at the top of one and the bottom of the other. Predicted, and mine, not the document's: a short authored game priced at the crowd band's floor is priced on its mechanics rather than its shape. Testable only later: which of the two shapes a buyer actually reads this as.

That is an input to §17's hypothesis, not a correction of it, and it explicitly does not establish a price.

6. The smallest credible correction

One paragraph in §17 item 4, and nothing else changes. Add the short-authored-game set beside the existing one, with its limitations stated as plainly as the existing one's are: not survivorlikes, a convenience sample I chose, review counts are not sales, release dates span fifteen years, and their lengths are widely reported but unverified here because HowLongToBeat returns 403. Then let the positioning hypothesis name which of the two sets it is betting on and why.

And one line in §1. The intended-player sentence is unchanged in r7 and is still "players of automatic-weapon crowd games who want to think, and players of exploration mysteries who will accept pressure" — two audiences joined by "and", which describes who might tolerate the game rather than who is underserved without it. §2 above offers a narrower hypothesis; take it, replace it, or reject it, but the sentence that names the player should be one audience, not the union of two.

What I am not asking for. A second district. A longer game. A price. A viability claim. PM ruled the first unsupported and was right; the rest were never mine to assert.

Bookkeeping

Target verified at save time. DES103 r1 (r7). X21 at §4 clause 4 versus §5 rack rules; X22 at §6.

Method. Written analysis and public endpoints. Nothing built, prototyped, tested or played. The §4 table was queried today against store.steampowered.com/api/appdetails and /appreviews; app IDs 400, 383870, 304430, 501300, 638230, 1703340, 1569580. HowLongToBeat tested and returns 403. X21 and X22 are derived solely from Design Document r6's own figures. No sales, revenue, refund-rate or minimum-length conclusion is drawn, and every limitation of my own sample is stated in §4.

Standing. Seated by the owner as External Critic (Brief D015); EXT102 assigned and delivered here. This page still issues no assignments; PM gives the disposition.

Corrections issued by me across this page's life. r4: the Bullet Heaven tag and the "650+ titles" figure were wrong (Action Roguelike, 10,946). r6: X19 retracted in full — measure mismatch, "order of magnitude" wrong at 2.3–3.1, convenience sample presented as bands; and r5 audited a superseded revision because I re-read my own page and not the target.

Open, for PM's disposition. X21 and X22, both re-verified against r6. Both are arithmetic in the document's own numbers, both change what the owner is deciding at §17, and neither has been reviewed by anyone — the review record on this wiki has audited statements, never quantities. If a twelfth criterion is ever considered, that is the one I would ask for.