Update d4050

Author: vinniefalco

source/d4050-failure-modes.md

@@ -37,7 +37,7 @@ The chair walks into a session with a paper, a room full of people, and a time s
 
 A decision that sets aside years of work gets the same process as a decision to add `[[nodiscard]]` to a function. The consequence varies by orders of magnitude. The process does not scale. The chair has no tools to scale it.
 
-The executor and networking arc is the case study. Over a decade, four locally reasonable decisions - each made by experienced practitioners under real constraints - produced twenty-one years without networking in the C++ standard. The failure modes extracted from that arc are general. The proposals in this paper are tools for the chair.
+The executor and networking arc is the case study. Over a decade, four locally reasonable decisions - each made by experienced practitioners under real constraints - produced twenty-one years without networking in the C++ standard, counting from [N1925](https://wg21.link/n1925)<sup>[15]</sup> (2005). The failure modes extracted from that arc are general. The proposals in this paper are tools for the chair.
 
 ---
 
@@ -59,7 +59,7 @@ Five retrospective papers ([P4094R0](https://wg21.link/p4094r0)<sup>[1]</sup>, [
 
 **A2. Assertions supported by hypothetical code only.** Six rationale assertions for unification were supported by one hypothetical code snippet as the only code-level evidence in the entire published record. *Source:* [P4094R0](https://wg21.link/p4094r0)<sup>[1]</sup> Section 4.
 
-**A3. Consensus poll on a claim without domain-specific evidence.** The October 2021 poll "including networking" achieved consensus. At the time, the published record contained no sender-based networking deployment, no prototype, and one hypothetical example. Published voter comments included "can't judge its suitability for networking" from voters who voted Weakly Favor. [P2430R0](https://wg21.link/p2430r0)<sup>[11]</sup>, documenting that compound I/O results cannot use `set_error` without information loss, was published two months before the poll. *Source:* [P4097R0](https://wg21.link/p4097r0)<sup>[4]</sup> Sections 2-3.
+**A3. Consensus poll on a claim without domain-specific evidence.** The October 2021 poll "including networking" achieved consensus ([P2453R0](https://wg21.link/p2453r0)<sup>[25]</sup>). At the time, the published record contained no sender-based networking deployment, no prototype, and one hypothetical example. Published voter comments included "can't judge its suitability for networking" from voters who voted Weakly Favor. [P2430R0](https://wg21.link/p2430r0)<sup>[11]</sup>, documenting that compound I/O results cannot use `set_error` without information loss, was published two months before the poll. *Source:* [P4097R0](https://wg21.link/p4097r0)<sup>[4]</sup> Sections 2-3.
 
 **A4. Predictions without follow-up.** [P2464R0](https://wg21.link/p2464r0)<sup>[12]</sup> stated "I don't know" whether Networking TS compositions scale. Five years later, no sender-based networking has shipped either. The constraint applied to one model was not applied symmetrically to the replacement. *Source:* [P4096R0](https://wg21.link/p4096r0)<sup>[3]</sup> Section 9.1.
 
@@ -79,7 +79,7 @@ Five retrospective papers ([P4094R0](https://wg21.link/p4094r0)<sup>[1]</sup>, [
 
 **D1. Pivotal analysis under one framing only.** [P1525R0](https://wg21.link/p1525r0)<sup>[16]</sup> identified four deficiencies in `execute(F&&)`. The analysis was under the work framing only. Under the continuation framing, three deficiencies do not arise and the fourth addresses a different question. *Source:* [P4095R0](https://wg21.link/p4095r0)<sup>[2]</sup> Section 4.
 
-**D2. Pivoting without evaluating cost to the blocked domain.** The Cologne pivot (July 2019) eliminated all interface-changing properties. No paper at Cologne analyzed the diagnosis under the continuation framing. No paper asked what networking loses. *Source:* [P4095R0](https://wg21.link/p4095r0)<sup>[2]</sup> Section 5.2.
+**D2. Pivoting without evaluating cost to the blocked domain.** The Cologne pivot (July 2019), driven by [P1658R0](https://wg21.link/p1658r0)<sup>[17]</sup> and [P1660R0](https://wg21.link/p1660r0)<sup>[18]</sup>, eliminated all interface-changing properties. No paper at Cologne analyzed the diagnosis under the continuation framing. No paper asked what networking loses. *Source:* [P4095R0](https://wg21.link/p4095r0)<sup>[2]</sup> Section 5.2.
 
 **D3. Absence of the blocked domain's use cases from the pivotal analysis.** [P1525R0](https://wg21.link/p1525r0)<sup>[16]</sup>'s examples are thread pools, deadline executors, GPU contexts, and `when_all`. No example involves an I/O reactor or sockets. *Source:* [P4095R0](https://wg21.link/p4095r0)<sup>[2]</sup> Section 3.3.
 
@@ -141,11 +141,13 @@ The documentation requirement must scale with the consequence. Scaling requires
 | Reversibility               | Can the decision be undone in a future standard?                      | Irreversible decisions need more scrutiny     |
 | Domain breadth              | Number of distinct domains affected                                   | More domains = more stakeholders to consult   |
 
+These criteria combine into four tiers, each with its own documentation requirement.
+
 | Tier | Name        | Criteria                                                                                    | Required artifacts                                                                                                                 | Example                                              |
 | ---- | ----------- | ------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------- |
 | 1    | Routine     | One paper, no deployed code, reversible, single domain                                      | The paper itself                                                                                                                   | Adding `[[nodiscard]]` to a function                 |
 | 2    | Significant | Multiple papers or one deployed codebase, single domain                                     | Stakeholder brief + evidence summary + reflector discussion                                                                        | Adopting a new library feature with deployment history |
-| 3    | Major       | Years of prior work, multiple deployed codebases, or multiple domains; partially reversible | Full artifact set (5.3); domain coverage report (Section 6)                                                                        | Adopting `std::execution` for C++26                  |
+| 3    | Major       | Years of prior work, multiple deployed codebases, or multiple domains; partially reversible | Full artifact set (5.3); domain coverage report (Section 6)                                                                        | Adopting `std::execution` for C++26 ([N4985](https://wg21.link/n4985)<sup>[27]</sup>) |
 | 4    | Structural  | Sets aside or replaces years of prior work; multiple domains; effectively irreversible      | Full artifact set + mandatory stakeholder presence or written position + decision rationale + follow-up criteria with revisit date | Setting aside the Networking TS; the Cologne pivot   |
 
 Four design choices keep the sizing system from becoming a political instrument.
@@ -167,6 +169,8 @@ By the time the meeting starts, the chair already knows the decision's size, the
 
 ### 5.3 The Missing Artifacts
 
+**F1. Missing artifacts disproportionate to the decision.** A Tier 4 decision proceeds with the same documentation as a Tier 1 decision. The artifacts that would have surfaced objections, alternatives, or gaps in evidence do not exist because nobody was required to produce them.
+
 Eight artifacts WG21 does not systematically produce. The table maps each artifact to the tier at which it becomes required.
 
 | Artifact                         | What it contains                                                                         | When produced            | Who produces it                                | WG21 today                                          | Required at tier |
@@ -197,7 +201,7 @@ Decision sizing: 13 years of prior work (Tier 4). 100+ papers in the chain (Tier
 
 ### 5.5 Hypothetical: Sofia Task Adoption (P3552R3)
 
-Decision sizing: P3552R3 is new (Tier 1-2 on prior work). But it affects `std::execution` (years of prior work, Tier 3). Multiple domains (Tier 3). Partially reversible via NB comments (Tier 2-3). **Tier 3 by highest criterion.**
+Decision sizing: [P3552R3](https://wg21.link/p3552r3)<sup>[28]</sup> is new (Tier 1-2 on prior work). But it affects `std::execution` (years of prior work, Tier 3). Multiple domains (Tier 3). Partially reversible via NB comments (Tier 2-3). **Tier 3 by highest criterion.**
 
 Jonathan M&uuml;ller's [P3801R0](https://wg21.link/p3801r0)<sup>[24]</sup> arrived after the Sofia vote. Under Tier 3 requirements, his concerns would have been a stakeholder brief before the vote. The 29 abstentions might have been fewer - people who abstained because they did not understand the issues would have had written briefs to read. The plenary vote passed 77-11. The record does not document why 29 members abstained.
 
@@ -419,3 +423,5 @@ The author thanks Christopher Kohlhoff for the executor model, [N3747](https://w
 26. [P4100R0](https://wg21.link/p4100r0) - "The Network Endeavor: Coroutine-Native I/O for C++29" (Vinnie Falco, Steve Gerbino, Michael Vandeberg, Mungo Gill, Mohammad Nejati, 2026). https://wg21.link/p4100r0
 
 27. [N4985](https://wg21.link/n4985) - "WG21 2024-06 St Louis Minutes of Meeting" (Nina Ranns, 2024). https://wg21.link/n4985
+
+28. [P3552R3](https://wg21.link/p3552r3) - "task - An Asynchronous Coroutine Task Type" (Lewis Baker, 2025). https://wg21.link/p3552r3