Commit 5280e4

2026-09-06 18:42:41 Anonymous: REF102 r5: repair sections 3a and 3b after PMR002 section 4. Achievement percentages are cross-sectional lifetime unlock shares, not completion, retention or abandonment rates; a ratio between two of them is a conditional unlock proportion only where the later marker's earners are demonstrably a subset of the earlier's, which sorting by share does not establish (Noita's parallel worlds are a concrete counterexample on this list). Headings corrected: 3a no longer says 'owners' or 'completion'; 3b is no longer 'the shape of the drop-off' and its column is 'share ratio', not step retention. Withdrawn: the 'never consumed' reading, the sentence pairing Noita reviewer playtime with an achievement share (different populations, no shared cohort), the drop-off/erosion/attrition language, r3's claim that 3b was firmer than 3a, and r3's answer to Q36 - which the data cannot answer, since that needs a longitudinal cohort no public Steam data describes. Every table, figure and query preserved unchanged per PM's instruction; no new survey run.

reference/survivorlike market data.md ..
@@ 1,14 1,17 @@
	# Reference — Survivorlike Market Data
-	**REF102 · r4 · 2026-09-06 · Reference Researcher · Reference, not a design proposal. Two claims on this page were wrong and are corrected in place.**
+	**REF102 · r5 · 2026-09-06 · Reference Researcher · Reference, not a design proposal. Four claims on this page were wrong and are corrected in place.**
+	r5 repairs §3a and §3b after [PMR002 §4](/Hold%20The%20Flood/Pm%20Reviews/Pmr002): achievement percentages are
+	**cross-sectional lifetime unlock shares, not retention or abandonment rates**, and a ratio between two of
+	them is a conditional unlock proportion **only where the later marker's earners are demonstrably a subset of
+	the earlier's** — which sorting by share does not establish. Every table and query is preserved unchanged;
+	the headings and conclusions around them are corrected. The audit is at [REF103 §7](/Reference/Length%20And%20Price%20Audit).
	r4 retracts §3's "order of magnitude" and §2's "two distinct bands" following PM's ruling, and points
	to [REF103](/Reference/Length%20And%20Price%20Audit), which carries the arithmetic, the separated
	measures and the completion-time data that was missing. The **open request is closed**: HowLongToBeat
	is reachable after all. Earlier revisions are page history.
-	r3 adds §3b, the shape of the drop-off, answering the structural half of
-	[Clarity Q36](/Hold%20The%20Flood/Clarity%20Questions). Step retention is a ratio over the same
-	population, so §3a's undocumented-denominator caveat cancels there. Adds to §3a; nothing already
-	published was changed. r1 and r2 are page history.
+	r3 added §3b as "the shape of the drop-off" and claimed the ratio cancelled §3a's denominator caveat;
+	**both of those are corrected at r5** — see the boxes in §3a and §3b. r1 and r2 are page history.
	r2 adds §3a, completion rates from Steam's public achievement API — the column
	[Clarity Q35](/Hold%20The%20Flood/Clarity%20Questions) and
	[External Critique X19/X20](/Hold%20The%20Flood/External%20Critique) are all missing. It adds to §3;
@@ 179,7 182,13 @@
	This does not make a three-to-four-hour game wrong. It makes it *unusual for this shelf*, which is a thing
	the owner should be told in section 17 rather than discover after launch.
-	## 3a. Completion: how many owners ever reach the end
+	## 3a. Lifetime unlock shares: what proportion has ever reached a marker
+
+	> **Heading and conclusions corrected at r5.** This section was titled *"Completion: how many owners ever
+	> reach the end"*. Two words in that were wrong. **"Owners"** — the denominator is undefined and cannot be
+	> called buyers or owners. **"Completion"** — these are cumulative lifetime unlock shares measured at one
+	> moment, not completion rates: a player counted as not having unlocked a marker may simply still be
+	> playing. See [REF103 §7](/Reference/Length%20And%20Price%20Audit).
	§3 measured how long engaged players play. It could not measure the thing the length argument actually
	turns on: **how far into these games people get.** Steam publishes that, and it needs no key.
@@ 224,7 233,8 @@
	Worth knowing alongside it: reaching Room 46 is **not** the end of Blue Prince's content — a substantial
	amount of the game sits after the credits ([Game Rant](https://gamerant.com/blue-prince-how-reach-room-46/)).
-	**Almost everyone starts these games.** The most-earned achievement, for comparison:
+	**The most-unlocked marker in each game, for comparison** (a high share, not evidence about who started
+	or when):
	| Game | Most-earned marker | Share |
	|---|---|---:|
@@ 233,8 243,10 @@
	| Cult of the Lamb | "Deal with the Devil" (opening) | 94.9% |
	| Pathogenic | "Defeat the Skin boss" (first boss) | 93.9% |
-	So the drop is not people bouncing off the install. In Vampire Survivors, **97.8% survive one minute,
-	84.0% survive twenty minutes, and 27.0% ever defeat the final enemy.** Same population, three markers.
+	In Vampire Survivors the three markers read **97.8% (survive one minute), 84.0% (survive twenty minutes)
+	and 27.0% (defeat the final enemy)** against the same undefined denominator. *This does not establish
+	that the same people passed all three, nor that anyone stopped:* it is three lifetime shares measured at
+	one moment. r4 read a story of people "bouncing off" into these numbers; that reading is withdrawn.
	**And the middle of the distribution is low everywhere.** Median achievement share per game:
@@ 258,10 270,16 @@
	against **9.3–46.2 h of engaged-player playtime** (§3). Those measure different things, and the gap between
	them is what X19, X20 and Q35 are all pointing at.
-	This section adds the third number, and it changes the shape of the comparison: **in four of the five
-	comparables that have a definite ending, between 10% and 38% of the player population ever reaches it.**
-	The comparables' impressive lengths are, for most of the people who own them, *optional content that is
-	never consumed*. Noita's median player logs 46.2 hours and one in ten ever wins.
+	What this section adds, stated at its real strength: **in four of the five comparables with a definite
+	ending, the share of Valve's counted population that has ever unlocked the ending marker is between 10% and
+	38%.**
+
+	> **Three readings r4 made that r5 withdraws.** (1) That the remaining content is *"never consumed"* — an
+	> unlock share at one moment cannot show that; players may still be progressing.
+	> (2) That these are shares *"of the people who own them"* — the denominator is undefined.
+	> (3) *"Noita's median player logs 46.2 hours and one in ten ever wins"* — this pairs a Steam-reviewer
+	> statistic with a global achievement share. **They are different populations and there is no shared
+	> cohort**, so the sentence compares nothing. Withdrawn entirely.
	Two things follow, and neither is a recommendation:
@@ 275,21 293,33 @@
	Whether any of this changes decision 4 is entirely PM's and the owner's call. It is offered because Q35
	asked for exactly the context a reader without industry background cannot supply.
-	## 3b. The shape of the drop-off
-
-	§3a measured the endpoint: how many players ever reach an ending. It could not say **where along the way
-	they leave**, which is the question [Clarity Q36](/Hold%20The%20Flood/Clarity%20Questions) is actually
-	asking — *"what this game is for someone who plays four expeditions and stops. Do they get a complete
-	small thing, or two thirds of a setup?"*
-
-	Where a game's achievements form a **sequence**, that shape is measurable.
-
-	> **Why this section is on firmer ground than §3a.** §3a's warning was that Valve never documented the
-	> population these percentages are over. **Step retention is a ratio of two percentages over that same
-	> population, so the unknown denominator cancels exactly.** If 84.0% reach marker A and 27.0% reach
-	> marker B, then 32.1% of those who reached A went on to reach B — and that holds whatever N is. The
-	> per-step figures below are therefore the most reliable numbers on this page. The raw shares still carry
-	> §3a's caveat; the *steps* do not.
+	## 3b. Ratios between ordered markers, and what they do not mean
+
+	> **This section was titled "The shape of the drop-off" and is corrected at r5.** It described these
+	> ratios as retention, attrition and players leaving. **They are none of those.** The audit is at
+	> [REF103 §7](/Reference/Length%20And%20Price%20Audit); the short version is in the box below. The tables
+	> are unchanged — only the claims made from them.
+
+	[Clarity Q36](/Hold%20The%20Flood/Clarity%20Questions) asks *"what this game is for someone who plays four
+	expeditions and stops."* **The measurements below do not answer that**, and r4 was wrong to present them as
+	if they did. What they give is the relative size of unlock shares along a sequence of markers.
+
+	> **What the ratio is, corrected at r5.** r4 argued that dividing two shares cancels the unknown
+	> denominator, making the result the most reliable figure on the page. The algebra is right and the
+	> conclusion was not. **The cancellation yields a conditional unlock proportion only if the later marker's
+	> earners are demonstrably a subset of the earlier marker's** — and sorting markers by share, or by a
+	> plausible in-game order, does not establish that. Achievements can be added by later updates (earnable
+	> only by players active after the patch), and alternative routes can reach a later marker without the
+	> earlier one: Noita's [parallel worlds](https://noita.wiki.gg/wiki/Parallel_Worlds), reachable by digging
+	> through the world edge and carrying their own Holy Mountains, are a concrete example on this very list.
+	>
+	> **And even where nesting does hold, the result is still not a retention or abandonment rate.** It is a
+	> cumulative lifetime proportion at one instant. Someone who has not unlocked the later marker may be
+	> mid-playthrough, not gone. Nothing here shows that earlier content was unfinished, unseen or unenjoyed.
+	>
+	> The column below is therefore labelled **share ratio**, not step retention, and should be read as *"the
+	> later marker's share is X% of the earlier marker's"* — an arithmetic relation between two published
+	> numbers, nothing more.
	### The sequences
@@ 297,7 327,7 @@
	**Vampire Survivors** — markers are ordinal by definition (minutes survived), and the last is the ending:
-	| Marker (the game's words) | Share | Of those who reached the step before |
+	| Marker (the game's words) | Share | Share ratio to the row above |
	|---|---:|---:|
	| Survive 1 minute with any character | 97.8% | — |
	| Survive 10 minutes with any character | 93.5% | 95.6% |
@@ 307,7 337,7 @@
	**Cult of the Lamb** — the four bishops are fought in a fixed order, Leshy → Heket → Kallamar → Shamura
	([Cult of the Lamb Wiki](https://cult-of-the-lamb.fandom.com/wiki/Bishops_of_the_Old_Faith)):
-	| Marker | Share | Step |
+	| Marker | Share | Share ratio |
	|---|---:|---:|
	| Defeat Leshy | 77.4% | — |
	| Defeat Heket | 54.3% | 70.2% |
@@ 318,7 348,7 @@
	→ Temple of the Art ([Noita Wiki: Biomes](https://noita.wiki.gg/wiki/Biomes)). The path continues to the
	Laboratory, which has no achievement of its own; "Victory" is the run's completion.
-	| Marker | Share | Step |
+	| Marker | Share | Share ratio |
	|---|---:|---:|
	| Reached Coal Pits | 72.5% | — |
	| Reached Snowy Depths | 59.3% | 81.8% |
@@ 331,49 361,51 @@
	**Pathogenic** — bosses. *The order here is **inferred from the shares alone**; I found no source stating
	the designed sequence, so treat this table as the weakest of the four:*
-	| Marker | Share | Step |
+	| Marker | Share | Share ratio |
	|---|---:|---:|
	| Defeat the Skin boss | 93.9% | — |
	| Defeat the Lung boss | 40.2% | **42.8%** |
	| Defeat the Heart boss | 27.5% | 68.4% |
	| Brain | 5.6% | 20.4% |
-	### Three different shapes, and they are not the same product
-
-	- **A complete beat early, then a cliff — Vampire Survivors.** Losses are trivial through the twenty-minute
-	run (95.6%, then 89.8% per step), and then **two-thirds of the remaining players never see the ending**
-	(32.1%). The whole satisfying unit is the run, and 84% get it. The ending is optional long-tail content.
-	- **Steady erosion, no plateau — Cult of the Lamb and Noita.** Every step sheds 15–40%, with no stretch
-	where the curve flattens. Nothing is delivered whole and early; the game is the descent.
-	- **An early cliff — Pathogenic** (order caveat above): most of the loss happens immediately after the
-	first boss, 42.8% carrying through.
+	### Three different profiles
-	**Front-loading, measured without needing any order.** Two comparables put a great deal above the waterline
-	early. Halls of Torment's fourteen most-earned achievements all sit between **83.8% and 99.1%**; Deep Rock
-	Galactic: Survivor's between **61.3% and 99.1%**. Both games hand most of their players a lot of distinct
-	completed things before any attrition bites.
+	r4 called these "shapes of drop-off" and described losses, erosion, cliffs and attrition. **All of that is
+	withdrawn** — the numbers are share ratios, not movements of people. What remains is a description of how
+	the published shares sit relative to one another:
-	### Bearing on the question Q36 raises (inference, for the lead designer and PM)
+	- **Vampire Survivors** — near-level across the run markers (ratios 95.6%, 89.8%), then a large gap to the
+	ending marker (32.1%).
+	- **Cult of the Lamb and Noita** — each successive marker's share is 60–85% of the one before, with no
+	stretch where the ratios approach 100%.
+	- **Pathogenic** — the largest gap is at the first step (42.8%), on the weakest-sourced ordering here.
-	Q36 notes that the design's tier one lands at expeditions one to three and calls that *"a genuinely good
-	structural choice which most of these comparables do not make."* The sequences above let that be checked,
-	and they refine it in a way that is worth having before the sentence goes to the owner:
+	**A distribution fact that needs no ordering at all**, and is therefore unaffected by everything above:
+	Halls of Torment's fourteen most-unlocked achievements all sit between **83.8% and 99.1%**; Deep Rock
+	Galactic: Survivor's between **61.3% and 99.1%**. Many distinct markers in those two games carry high
+	lifetime unlock shares. That is a statement about shares, and it implies nothing about when, in what order,
+	or whether anyone stopped.
-	**The early complete beat is not rare among these comparables — it is what the most successful one does.**
-	Vampire Survivors delivers a whole, finishable unit to 84% of its players and puts its ending far out on a
-	tail only 27% reach. Halls of Torment and Deep Rock Galactic: Survivor front-load similarly. The games that
-	*don't* do it — Noita, Cult of the Lamb — are the ones with a smooth erosion curve and no plateau.
-
-	So the structural point in Q36 stands, and the supporting clause is the part to drop: an early tier one is
-	a good choice **because it matches what works**, not because it is unusual. And the second half of the
-	finding is the one that bears on tier three: in every sequence measured here, **the last thing built is
-	seen by a minority** — 10.0% in Noita, 27.0% in Vampire Survivors, 33.9% in Cult of the Lamb, 38.0% in
-	Blue Prince (§3a). A design that puts its final choice about the Flood at expeditions eight to eleven is
-	making a normal choice for this shelf; what the comparables suggest is that the *earlier* tiers carry the
-	product for most buyers, and that is a claim the document can make with evidence rather than hope.
+	### Bearing on the question Q36 raises (inference, for the lead designer and PM)
-	None of this is a recommendation about where tier three should sit. It is the measurement Q36 said a
-	reader without industry background could not supply.
+	**r4's answer to Q36 is withdrawn.** It read: *"the early complete beat is not rare among these
+	comparables — it is what the most successful one does,"* and it went on to say that earlier tiers "carry
+	the product for most buyers." Both sentences rest on reading these shares as a record of players moving
+	through content and stopping, which [PMR002 §4](/Hold%20The%20Flood/Pm%20Reviews/Pmr002) correctly rejects.
+	Neither is supported.
+
+	**What survives, stated narrowly.** Across the four sequences, the marker corresponding to each game's
+	ending carries a lifetime unlock share of 10.0% (Noita), 27.0% (Vampire Survivors), 33.9% (Cult of the
+	Lamb) and 38.0% (Blue Prince, §3a), while markers earlier in each game carry higher shares. **That is a
+	description of published unlock shares and nothing else.** It does not establish that the earlier content
+	carries the product, that anyone abandoned the later content, or that a minority "sees" the ending —
+	players may still be progressing, and the denominator is undefined either way.
+
+	**And it does not answer Q36.** Q36 asks what the game is for a player who stops early. That is a question
+	about a cohort followed over time, and **no public Steam data describes one** — see
+	[REF103 §7](/Reference/Length%20And%20Price%20Audit) for what would be needed and why it is not obtainable.
+	As PMR002 puts it, the question *"remains useful independently of the statistical premise"*; this page
+	should not be cited as having answered it.
	## 4. The refund policy, stated exactly
@@ 483,7 515,7 @@
	**Drift.** §3a's figures were taken across a few minutes; two calls minutes apart already differed in the
	last decimal (Noita's most-earned achievement read 72.4% and 72.5%). Treat the final digit as noise.
-	**What r3 changed.** Added §3b, the shape of the drop-off, after
+	**What r3 changed** (its conclusions are superseded by r5; recorded here as history). Added §3b, then titled "the shape of the drop-off", after
	[Clarity Q36](/Hold%20The%20Flood/Clarity%20Questions) asked what the game is for a player who stops
	early — a question §3a's endpoints could not answer. Four ordered achievement sequences with per-step
	retention. Two orders are sourced (Noita's biomes, Cult of the Lamb's bishops), one is ordinal by
@@ 491,11 523,21 @@
	shares alone and labelled in place as the weakest table in the section**. Nothing previously published
	was changed.
-	**A note on §3b's reliability versus §3a's.** §3a rests on percentages over a population Valve has never
-	defined. §3b's per-step figures are ratios of two such percentages over that same population, so the
-	unknown cancels exactly and the step numbers are sound regardless. Where §3b quotes raw shares it
-	inherits §3a's caveat; where it quotes steps it does not. This is stated in the section itself so nobody
-	has to take it from the bookkeeping.
+	**A note on §3b's reliability, corrected at r5.** r3's bookkeeping claimed §3b's per-step figures were
+	"sound regardless" of the undefined denominator because the ratio cancels it. The cancellation is real but
+	the conclusion was wrong: the ratio means something only where the later marker's earners are
+	demonstrably a subset of the earlier's, and even then it is a cumulative lifetime proportion rather than a
+	retention rate. §3b is **not** firmer than §3a; it was over-claimed in a different way.
+
+	**What r5 changed.** Corrections only, no new measurement, per
+	[PMR002 §4](/Hold%20The%20Flood/Pm%20Reviews/Pmr002)'s instruction to repair the current inference rather
+	than run another survey. Every table, figure and query on this page is **unchanged**; the headings and the
+	conclusions drawn from them are corrected. Withdrawn: §3a's "owners"/"completion" framing, its
+	"never consumed" reading, and its sentence pairing Noita's reviewer playtime with an achievement share
+	(different populations, no shared cohort); §3b's "drop-off", "retention", "erosion", "attrition" and
+	"cliff" language, its claim to be firmer than §3a, and its answer to Q36. The audit that drove this is
+	[REF103 §7](/Reference/Length%20And%20Price%20Audit). Old wording is quoted beside each correction so
+	anyone holding a stale sentence can see what replaced it.
	**What r4 changed.** Corrections only, no new measurement. PM ruled that this page's length and price
	conclusions rested on mismatched measures and an over-read sample, and assigned