Final check: box and fixation in the revised 05 and 06#

Final check of 05-maynard-risk-and-ai-map.md (revised 27 September 2026, about 54,100 words) and 06-huang-and-late-lessons-through-maynard.md (rebuilt 27 September 2026, about 25,100 words). The standard is the checklist in perspective/portrait.md §10. The check was run after both revisions and their logs. Quotations were verified against working/maynard/corpus/, the FFTF text (working/maynard/book/), Resources/maynard-papers/web/ and Maynard stuff/substack article 1 draft 6.md. Line numbers refer to the files as of 27 September 2026, 10:30.

Verdict in brief#

The revisions fixed the root problem the earlier reviews found. Both documents now lead with his way of thinking. Layer language is gone. The private “hubris” remark, “flummox”, “the boat has left the harbor” and the toxicology parenthesis no longer carry any weight. Play is argued on risk grounds, and orphan risks and navigation have real standing. Recency looks acceptable by volume: in 06’s body, 27% of citations to his work date from 2005–2017, 20% from 2018–2022 and 22% from 2026.

What remains is subtler, and it concentrates at the points that carry the most weight.

1. 06 organises the lens around risk. Its lens is a set of risk concepts. His purpose (human flourishing, the future of being human) and his own umbrella frame (navigating advanced technology transitions) are absent from it.
2. Both documents demote the mindset in their summaries. Where they sum up, the change of mindset they describe becomes “a second lens beside” conventional safety again, and they judge it by the operational tools it does not supply.
3. Fixation has moved. It now sits in a handful of load-bearing phrases: “flip it” (one spoken line in an AI-drafted lecture), the “navigator” argument (one passage from January 2026, made “his strongest reason”), “fixed points” (the analysts’ term, presented as his), and a series introduction that is still a draft.
4. The corrections have overcorrected in places. “Never a midpoint” contradicts his own “the space between”. Navigation has been packaged into a component list, and the method into numbered moves. Both drift back toward procedure.

Ranked issues#

1. 06’s lens leaves out what his risk thinking is for: human flourishing and navigating advanced technology transitions (06; high)#

Evidence. - 06 never uses “advanced technology transitions” and never uses “flourish”. “Thriv” appears once. 05 uses them 12, 25 and 22 times. - 06 §3 has eleven subsections: mindset, value, landscape, navigating, orphan risks, learning across technologies, play, humility, the navigator, the public scholar, and the combination. None is about purpose. Being human appears as one sentence in §3’s “In brief” (“Underneath is a question about people”, line 137) and as one reading area (§4.9). - His own self-descriptions point the other way. His work is about “human flourishing and navigating complex advanced technology transitions” (2026-07-10). What drives it “more than anything” is whether technologies “begin to fundamentally change who we are” (2024-01-01). - 05 itself puts Purpose at the top of its architecture and says “His risk theory is one of the ways he pursues it, not the other way round” (05 §3, line 361). - In his draft series introduction, the thing he says a conventional first pass lost is “how I approach navigating advanced technology transitions”. - 06 never uses his own model for navigating AI transitions: three intersecting foci, “where we live”, “what we do” and “who we are” (2025-01-07; 2025-03-30 n.1). The interview maps onto it almost exactly: - Where we live: data centres, water, energy and the local veto [1:40:15]. - What we do: jobs, ambition and the purpose of a job [05:55, 11:29]. - Who we are: skills, “speak human” and formation [17:07, 22:26]. - 06 does use his four-ways model (2024-08-18), the lesser-known of the two.

Why it matters. This is the conventional box at the level of structure. Built this way, 06 reads him as a risk scholar with an unusually open mind, when he is a scholar of the future of being human for whom risk thinking is one instrument. The opportunity side of the moment stays thin: 06 flags that the analyses left it out (§6.5, §8.1) but does little with it itself.

Fix. - Add §3.0, “What it is all for”: flourishing, value created as well as threatened (05 C1), and the transitions frame. - Give the three foci a place in §4, as a map of the interview built on the model he uses for AI himself. - Open the §2 “In brief” with purpose, then mindset. - Add “flourishing” and “transition” to the questions in §11.

2. At summary points, the mindset is demoted to an add-on, on the strength of one [mixed] phrase (05, 06; high)#

Evidence. - 05 §2.9 closes: “These framings work best as a second lens beside capability-based safety and legal compliance, which is how he offers them” (line 303). 06 §3.11 repeats it (line 289). - 06 §9 opens by positioning every approach in the report: “The approaches add to engineering safety rather than replace it: he offers his frontier-AI framework ‘not as an alternative, but as an augmentation’ (2026-07-16 [mixed])” (line 686). That breaks 06’s own rule that a [mixed] text “is used as corroboration. None carries a position alone” (§1.4, line 68). - Elsewhere the same documents describe a frame, not a lens beside a frame: - “It names the stance within which management tools are used” (05 §2.3; 06 §3.4). - Gates belong “inside a larger stance” (06 §4.5). - “Without risk innovation, all we are left with is business as usual” (NN 2015-09 p.731). - Conventional categories are “the shavings off the tip of the AI iceberg” (2023-05-31). - The “augmentation” and “complementary” wording (2026-07-16; JLME 2024; Coronavirus 2020) is his, but it concerns how tools and frameworks relate to existing safety practice, not how the way of thinking relates to it.

Why it matters. The checklist’s first item says his concepts “are not procedures or add-ons to existing risk management”. A reader who reads only the verdicts will come away with the add-on reading.

Fix. Replace the “second lens” sentence in both places with a two-level statement. At the level of tools, his frameworks complement capability-based safety and compliance, which he says should not be loosened. At the level of thinking, they change the questions those tools are asked to serve (FFTF pp.22–23; NN 2015-09 p.731; 2016-01-11). In 06 §9, cite the tool-level point to JLME 2024 and Coronavirus 2020 first, with 2026-07-16 as corroboration only.

3. Weaknesses are judged almost entirely by operational yardsticks, and the same point recurs (05, 06; high)#

Evidence. The idea that he has not supplied thresholds, a rule, a bar or an operational bridge appears, by my count: - In 06: - §3.11: “no rule”, “no thresholds of the kind a regulator needs”, “without having assembled them for AI”; - §9.4: a six-row “Not supplied” table, plus the closing “that he has not yet assembled for AI”; - §10, three times: “A regulator who needs a threshold gets a framing”; “has not restated his transfer rule in operational form”; and “Mindset over tool”. - In 05: - §2.9: thresholds, aggregation, “an explicit rule for acting under uncertainty”, “have not been assembled for AI”; - the §3 unsettled list; - tensions 1, 3 and 16; - the “Gaps” section.

The table in 06 §9.4 pairs each gap with “what it supplies instead”, which helps. But nowhere is a mental model assessed as a mental model. Neither document asks: - whether it travels without him; - whether it changes what people notice (the Planner’s thirty-minute design brief, 2023-11-21; students’ reception, 2019-08-13); - whether firms can adopt the vocabulary of “navigation” without the change of mind, using it as a euphemism for going ahead; - whether it depends on facilitation and designed spaces (touched on only under “Access and reach”).

Tension 16 (05 line 1158; echoed in 06 §10, line 746) also says that reading the concepts as mental models “is partly a later emphasis on what was at the time also offered as a service”. Its own evidence points the other way: the Planner’s brief was to develop “a risk innovation mindset”. The founding column (NN 2015-09) and FFTF p.39 (“ways of thinking”) both predate the 2019–20 Nexus tools.

Why it matters. Judging mental models by the thresholds a regulator needs treats them as operational add-ons that are unfinished. That is the box, applied through evaluation.

Fix. - State the operational gap once in each document, as deliberate (“no easy guidelines or rules of thumb”, FFTF p.39) and with its honest cost. - Delete the repeats in 06 §3.11 and §10, and in 05 tensions 1 and 3; keep one statement in 05 tension 16 and 06 §9.4. - Add an assessment in mindset terms: how it travels, how it can be co-opted, what it needs in order to work, and what evidence of changed perception exists. - Revise tension 16. The tools were built to cultivate a mindset, framed in entrepreneurs’ language. That is not a later reinterpretation.

4. “Flip it”: one spoken line in an AI-drafted lecture has become a move, a lens, a sub-heading and a constant (05, 06; high, fixation)#

Evidence. - “Flip it” occurs once in the whole post corpus, in a single sentence of 2026-09-24, the King’s College lecture drafted into prose by an AI model (“so you can avoid it or flip it, and so get to the good”, corpus line 244). - It now appears in: - 06: the §2 “In brief” (“look for the flip”, line 127); §3.4 (line 180); a bolded element in §5.4, “The flip, in his sense” (line 525); and §9.1 move 6, “Look for the flip” (line 695); - 05: lens M5 (line 1197); the concept row “Navigating rather than managing” (line 689); and the endpoint of constant 16 (line 1069). - 05 §2.3 adds, without marking it as interpretation: “‘Flip it’ is the entrepreneur’s instinct inside risk innovation” (line 183). - 06 also uses Klein’s “the flip” (lines 99, 446) for a different idea, so one report carries two unrelated “flips”. - This breaks 06 §1.3: “No heading, alignment or divergence rests on a single … spoken aside”.

Why it matters. This is exactly the fixation the checklist warns against. The underlying idea is well documented in his own prose; the name is not.

Fix. Name the idea in his long-standing words: - risk turned “from a barrier to progress into a way of supporting beneficial and sustainable progress” (2016-01-11); - risks to “circumnavigate” or “strategically absorb” (2023-11-21); - responsible practice as “the competitive edge” (2018-09-03).

Retitle 06 move 6 (for example, “Look for ways through, around and beyond a threat”) and the §5.4 element. Cite the lecture as corroboration only. Drop Maynard’s “flip” wherever Klein’s also appears.

5. The “navigator” argument is backdated and made “his strongest reason”, from one 2026 passage (05, 06; medium-high, fixation and recency)#

Evidence. - The argument rests on one passage in 2026-01-10: “But what if the mismatch impacts the very cognitive abilities we rely on to navigate differences between what we experience, and what we’ve evolved to live with?” There is corroboration in 2026-01-17 (“how do I know I’m not an unwitting victim here?”). - The documents quote it cut off after “navigate” (06 §4.5; 05 §4 connections, which add “technological change”). - 06 calls it “his sharpest point” (§2, line 92), and says in §3.9, where claims are “[Stated] unless marked”: “This is his strongest reason why AI makes a change of mindset unavoidable” (line 264). 05 C15 says the same without an interpretation marker (line 620). - 05 §2.2 and 06 §3.1 say the argument was “argued since 2018” (05 line 153; 06 line 147). The 2018 quotation offered is FFTF p.174, “we’re not thinking creatively enough”, which is about failures of imagination. - The nearest 2018 text is FFTF p.177, “guarding against AIs that learn how to use our cognitive vulnerabilities against us”. That is first-order manipulation, not the second-order claim. - 05’s own firmness line in C15 is accurate: “the navigator argument is set out in 2026”. - The reason he himself gives for a changed mindset is different: AI fits no earlier type of risk (FFTF pp.22–23; “shavings off the tip of the AI iceberg”, 2023-05-31; “defies analogy”, 2026-01-22).

Why it matters. - It over-weights recent work. - It replaces his stated reason with the analysts’ favourite synthesis. - It mislabels an interpretation as [Stated].

The phrase “risk to the navigator” is correctly marked † in 05’s tables but not in the prose.

Fix. - Date the second-order argument to 2026, rooted in FFTF p.177 (2018) and FR p.56 (“increasingly poorly equipped”, 2020). - Quote the full sentence. - Mark “sharpest” and “strongest reason” as [Inferred], or drop them. - Lead 06 §3.1’s “Why AI makes the change unavoidable” with his own reason, AI fitting no earlier category, and then add the navigator argument as a 2026 extension.

6. The analysts’ vocabulary is presented as his, and it turns navigation back into a procedure (05, 06; medium-high)#

Evidence. - “Fixed point(s)” appears nowhere in the post corpus, the FFTF text or the supplementary text files. It appears 9 times in 05 and 12 in 06. In 05 it is part of a commitment title: C7, “keep fixed points where harm cannot be undone” (line 546). In 06 it appears in: - §2 (“with fixed points where harm cannot be undone”); - §3.4 (a bullet); - §4.1 and §4.5; - §9.1 move 4 (“Keep a few fixed points”). - It entered through the portrait’s independent assessment (“navigation needs fixed points”, portrait §8; value.md guardrails). The analysts’ prescription has been written back into his stance. - “A map, not a forecast” is also absent from his corpus. - Neither phrase carries 05’s † marker, although 05’s conventions require one for “a descriptive label of the map’s” (§1, line 96). 06 §3 presents such phrasing as [Stated]. - The result is a four- or five-part apparatus of navigation in both documents: map, fixed points, course correction, openings, direction. The method in 06 §9.1 appears as eleven numbered “moves”. 05 has grown from 29 to 37 lenses. value.md warned: “Do not replace one checklist with another.”

Why it matters. The portrait says “His method is not a procedure”. A component list presented in his name is the operational box coming back through the side door.

Fix. - Mark “fixed points” and “map, not a forecast” as the map’s terms (†), or replace them with his own wording: “turn the clock back” (2025-03-02 n.2), “trigger points” (Nature 2011), “rapid course correction” (2025-05-18) and “quick to question, and slow to respond” (NN 2016-03 p.212). - Describe navigation as a stance, illustrated by examples, not as a list of parts. - In 06 §9.1, compress the eleven moves into three or four questions and show one worked example in prose.

7. 06 cites a series introduction that is still a draft, and paraphrases it selectively (06; medium-high, fixation)#

Evidence. - Maynard stuff/substack article 1 draft 6.md carries the front matter “draft: 5 (Claude light edit of draft 4 …)”, placeholder links (“(insert-url)”), and describes the companion articles as “forthcoming”. - The value.md guardrail recorded that “The series intro is an unpublished draft.” - 06 nonetheless cites it as published (“The Future of Being Human, 27 September 2026”, §1.2 and Appendix B), about ten times. It uses it for positions: how he reads a leader (§4.1), his reservation about 01–04 (§1.1, §6.5), the reports’ “particular frame” (§6.4) and the article (§7). - The paraphrase in §4.1 omits the other half of his first reaction. His first thought was a quick post on claims by Huang “that felt naive and misguided”, before he checked that impulse. That half is exactly his combination of judgement and curiosity. - §6.5 turns his reservation into a “narrower frame”. His draft words are that the analysis did not position AI “within a broader landscape of emergent AI characteristics, capabilities, threats, risks, and benefits”. The capability, emergence and benefit side drops out. - §1.1 says the later essay is one “in which Maynard responds in his own voice”. The draft says it is “also authored by Claude, but carefully checked and edited by me”.

Fix. - Check whether the introduction has been published. If it has, cite the final text and restore both halves of each paraphrase. - If it has not, mark it as forthcoming and remove it as evidence in §4.1; 2024-08-04 and 2023-05-15 carry that reading on their own. - Correct the description of the later article.

8. Identity box: the defining noun is “measurer” or “quantitative risk scientist”, and only the foundations are “identity-defining” (05, 06; medium)#

Evidence. - 06’s first substantive sentence about him: “Maynard is a quantitative risk scientist who argues…” (§2, line 92). §3’s “In brief”: “Maynard is a measurer who argues…” (line 137). - 05 §3 opens: “came to AI as a risk scientist who built on his discipline”. - 05 labels only two concepts “Core (identity-defining)”: quantitative risk assessment (line 682) and hazard versus risk (line 692). C4’s firmness line says “identity-defining”, and “professional home” appears twice (lines 151, 520). - His own self-descriptions are broader: - a physicist for whom physics was “all about the sheer delight of putting ideas together in different ways” (TechTrends 2023 p.2); - “very un-disciplinary” (same interview); - “So much of how I explore new ideas … is grounded in play” (2024-03-17); - someone who works on flourishing and transitions (2026-07-10). - His September 2026 account puts quantitative risk science as “part of his foundations”.

Fix. - Replace “identity-defining” with “foundational”. - Open 06 §2 and §3 with the combination the portrait uses: the physicist’s delight, the risk scientist who learned where the numbers stop helping, and the un-disciplinary scholar of the future of being human. The insider’s reframing stays as a distinctive element, not as his identity.

9. Overcorrection: “never a midpoint” contradicts “the space between” (05, 06; medium)#

Evidence. - 05 §2.9 item 6 and 06 §3.11 item 6 say “Each refusal comes with a reframing, so none is a midpoint”. 05 §2.10 says “he is not at their midpoint either”. - Both documents quote the 30Y 2026 sentence as stating “the thesis exactly”. They stop before the next sentence: “The interesting and difficult work is in the space between” (2026_What-Thirty-Years...md, line 80). - 05 C7 itself records that “On precaution he has sought a middle ground since 2007” (AOH 2007 pp.9–10). The portrait records “a middle course” (2026-09-24). - Separately, a single April 2026 retrospective line is promoted to the thesis statement (05 §2.2, C3; 06 §3.1). His 2016 statement of the same idea (2016-01-11, “a way of supporting beneficial and sustainable progress”) is older and should come first.

Fix. Say what he says. He refuses the camps, and he locates the work in the ground between them, understood as terrain to be navigated, not a point of compromise. Quote the “space between” sentence. Cite 2016-01-11 before 30Y 2026 for the thesis.

10. In 06, imagination is applied mainly as a stance on forecasting, not as a way of seeing (06; medium)#

Evidence. - 06 §3.7 argues play and creativity on risk grounds well. - In the reading of Huang, §4.6 frames the difference as “Enough predictions” against “better speculation: labelled, plausible, open to other voices”. That is the conventional debate about speculation and evidence. - The generative work that his record credits to imagination never gets done on the material: juxtaposition, story, hands-on play that produces concepts, and creativity as a skill of risk perception. - “What are we failing to imagine?” appears as a lens (05 M3) and as open question 11.6, but it is not attempted anywhere in 06. - Films are used well: Ex Machina in §4.5 and §8.3, The Man in the White Suit in §4.8, and “wow” to “meh” in §4.1. That shows it can be done.

Fix. - Lead §4.6 with creativity as risk perception (FFTF p.174; 2016-01-11), and keep forecasting discipline second. - Add one explicit, labelled exercise, for example: what would a playground rather than a playpen reveal about “we’re going to discover new ones” [22:26]?

11. Residues of the fixation fixes in 05 §8 (05; medium-low)#

Evidence. - The phase table’s row 8, “Signature new ideas”, still lists “safety message first” (line 992). The revision re-weighted it everywhere else as “a single 2026 wording” of a long-standing stance. - The same row lists “the gap and the early window”, which 05 anchors elsewhere in 2008 (§6.4, “Recurring (from 2008)”). - The signalled changes-of-mind table keeps two single-line 2026 items: - “We’re not on a plateau” (S3 2026, line 1101), removed from 06 for recency; - “Early GPT a ‘toy’” → “I was wrong” (line 1103), a spoken aside in the [mixed] lecture.

Fix. - Remove “safety message first” from row 8. - Move “the early window” to phase 1 (with the 2026 retrospective as a restatement). - Drop the two table rows, or mark them as single-source.

12. Curiosity, the first of his principles, has no conceptual standing in 05 (05; medium-low)#

Evidence. - “Obsessive curiosity” leads the initiative’s principles. “Something catches him” is step 1 of the method. - “Social curiosity” (FFTF p.222) is his remedy for myopic benevolence, which makes it a risk concept. 06 uses it well (§4.8). - 05 has no §6 row for curiosity or social curiosity, no commitment, and no lens. D2 (“Who is certifying that this is responsible?”) never asks whether anyone has asked the people affected “what they think, and what they want”. - Plausibility, by contrast, has a Core row, a named filter and the label “his most consistent epistemic habit” (C8 firmness). The §3 paragraph headed “Imagination disciplined by plausibility” (line 341) gives two sentences to imagination and the rest to plausibility and complexity.

Fix. - Add a §6.2 row, “Social curiosity (the missing quality in myopic benevolence)”, and a §6.5 row for curiosity as method. - Add the social-curiosity question to D2. - Retitle the §3 paragraph “Imagination and plausibility together”, to match C8.

13. “His most valuable framing for AI” leans on the part of orphan risks most likely to have come from the model (05, 06; medium-low)#

Evidence. - The weighting (05 §2.9 and the §6.1 row, “Value for AI: high”; 06 §3.5) and 06’s “three orphaned regions” (§5.3) rest on the institutional formulation of 2026-07-16 [mixed]: “by what process …”, “organized themselves not to see”, “incentive field”. - Of that paper, he said the model applied his work to frontier AI “in a way that hadn’t previously occurred to me” (2026-07-04). - There is secure own-prose support: 2018-12-13; 2020-10-15; and NANO 2026 (April, before the paper: AI’s human-side risks, which “no existing institution owns”).

Fix. Anchor the weighting on those texts. Say once that the question of how a risk becomes nobody’s responsibility may be partly the model’s framing. In 06 §5.3, label the three regions as the report’s application, with the paper cited for the facts about the frameworks only.

14. A known misreading is still in 05 (05; low)#

Evidence. - 05 §6.9 describes the slide “from ‘wow’ to ‘meh’” as “a real loss” (line 850). - FFTF p.285 calls it “a really important survival mechanism”, while noting that everyday life can swamp the significance of discoveries, and later that a “meh” response “probably isn’t the best strategy” for handling powerful technologies. - The 06 revision log flagged this and left 05 unchanged.

Fix. Describe it as double-edged, as 06 §4.1 already does.

15. Wording that dates, or rests on the unpublished account (05, 06; low)#

Evidence. - 05’s header and §1 Method say the map was “reviewed by him”. 06 says “for his review”, which is accurate for an edition he has not yet seen. - 05 §1 frames its note as “responding to an earlier analysis of his work”, and 06 §1.3 explains that “This edition replaces an earlier one”. Both are process history that will date. - The organising phrase “mental models that open up possibilities” (13 times in 05, 6 in 06) follows the unpublished September account closely. - In his published prose “mental model” appears three times, never about his own concepts. - His published vocabulary is “mindset” (in 29 posts), “ways of thinking” (FFTF p.39), and “designed to open up new ideas and possibilities” (2016-01-11).

Fix. - Align 05 with 06’s “for his review” until he has reviewed this edition. - Cut the process history, or move it to a dated note. - Anchor the key descriptor in his published wording, with “mental models” given as his September 2026 gloss.

Checklist status (portrait §10)#