# NOTES_2026-08-21.md — The Unknown Face (jung_psyche) The standing reference. Read this before touching the paper again. --- ## 1. Where the project stands | Deliverable | File | Status | |---|---|---| | Research base | `research_notes.md` | Phase 0 + four sweeps, UNVERIFIED flags live | | Number/quote ledger | `data.py` | every prose and figure number, sourced + statused | | Figures | `generate_figures.py` → `figures/*.png` | 6 figures, rebuilt at v2 | | Draft v1 | `build_paper_v1.py` → `unknown_face_v1_{core,full}.pdf` | superseded; kept, never edited | | Trial record | `ATTACK_NOTES.md`, `DEFENSE_NOTES.md`, `VERDICT.md` | 24 charges → 18 orders | | **Draft v2 (current)** | `build_paper_v2.py` → `unknown_face_v2_{core,full}.pdf` | revised under all 18 orders | Rebuild from scratch, two commands: ``` python3 generate_figures.py python3 build_paper_v2.py ``` Word counts. Two measurements, because the bench's target was set on raw whole-PDF extraction and that number includes captions, the table, the reference block, and page furniture: | Edition | v1 raw | v2 raw | v2 body (before Sources, footers stripped) | bench target (order 13) | |---|---|---|---|---| | core | 6,636 | **6,442** | 5,774 | ≤ 5,200 | | full | 8,695 | **10,599** | 9,806 | 9,000–9,500 | Core is now 61% of full (was 76%). See §5 for why the numeric limb of order 13 is recorded as partial rather than met. **Archive warning (verification finding, 2026-08-21).** `build_paper_v1.py` still builds and exits clean, but it is not *reproducible*: both build scripts read one mutable `figures/` directory and one mutable `data.py`, neither pinned. Rebuilding v1 today regenerates it against v2's figures and corrected numbers — producing, for example, a v1 caption describing "bars... drawn at 2×" above the v2 dot plot. **The shipped `unknown_face_v1_*.pdf` files are the archival artifacts; do not overwrite them by rerunning v1.** A v3 that cares about this should snapshot `figures/` and `data.py` per draft (e.g. `figures_v1/`, `data_v1.py`) rather than editing the old build script, which house rules forbid. --- ## 2. The thesis, stated precisely Three claims of declared and different strengths: 1. **Descriptive (strong evidence).** Jung's architecture — small conscious ego, large active unconscious, a disposal mechanism relocating the unowned into other people — survives, once stripped of inherited archetypal images, discrete types, dream compensation, and synchronicity. Roughly half the map is dead and the paper says so in its own text. 2. **Diagnostic (argued as best of three named candidates, NOT established).** Public life runs on measured misperception at scale — the ×2 pattern — amplified by platforms that price out-group animosity and left uncontained by emptied institutions. Whether the engine is *defensive projection*, *egocentric inference*, or *information diet / representativeness* is open; the paper argues projection and prints the discriminators and disconfirmers. 3. **Prescriptive (a specified wager, not a finding).** Sustained, voluntary projection-withdrawal practice at civic scale is the missing durable intervention. Endpoints, dose floor, and retention threshold are stated; the mechanism claim additionally requires an inward-only trial arm. **The v2 headline change:** claim 2 was demoted from identity ("the ×2 pattern *is* projection") to argued candidacy. That single sentence in v1 §1 was the root of all four FATAL charges at trial. --- ## 3. What is established, and how well | Claim | Standing | Load-bearing evidence | |---|---|---| | Vast nonconscious processing | Strong, but architecture only | Zheng & Meister 2024 (10 bits/s vs ~10⁹); Wilson 2002; Friston 2010 | | Projection exists as a mechanism | Strong, in reformed (suppression→accessibility) form | Newman, Duff & Baumeister 1997; Schimel 2003; Govorun 2006 | | The ×2 misperception pattern | Strong as measurement | Levendusky & Malhotra 2016; More in Common 2019 (advocacy, graded); Moore-Berg 2020; Waytz 2014 | | Projection *causes* the ×2 pattern | **Weak — candidate only** | valence asymmetry + defensive-function lab results, both graded suggestive | | Misperception → animosity | Supported | Landry 2023; Ahler & Sood 2018 (−6.4 pp) | | Animosity → democratic erosion | **Contested — printed as contested** | Broockman, Kalla & Westwood 2023; Voelkel et al. NHB 2023 | | Sustained practice beats one-shot | Untested; the wager | megastudy decay (7–53% at 2 weeks) is the motivating gap | | Jungian analysis works | Effectiveness-level only, no RCTs | Roesler 2013 | --- ## 4. What the paper will be attacked on, and the answers 1. **"Misperception isn't projection."** Answer in the paper: correct, not established — §4 names egocentric inference and information diet, prints Table 1 showing the ×2 pattern and the engagement gradient discriminate nothing, and states two disconfirmers in advance. 2. **"Ahler & Sood kills it."** Answer: cited in the paper's own text with its magnitudes, its in-party finding, and its correction result — the strongest rival, presented as such. The narrow ×2 construct (attitude/hostility) is distinguished from composition errors (far larger). 3. **"Implicit-bias training already failed."** Answer: what failed was one-shot, coerced, cognitive; what durably works is sustained, voluntary, relational — with Paluck, Green & Green 2019 (d = 0.39, CI 0.23–0.55) cited alongside Pettigrew & Tropp, caveats included. 4. **"Jung was an antisemite in 1933–39."** Answer: conceded in full, and explicitly *not* scored as evidence for the mechanism. Samuels' genre charge is answered by the essentialist/aggregative register commitment. 5. **"Unfalsifiable."** Answer: the wager now has population, dose floor (≥8 sessions / ≥3 months), primary endpoints, and a 6-month retention threshold; the mechanism claim requires an inward-only arm. --- ## 5. What changed from v1 → v2 (all 18 orders) REQUIRED orders 1–14, RECOMMENDED 15–18. Status of each: | # | Order | Done | |---|---|---| | 1 | Demote central inference to candidacy (§1, abstract, §4) | yes | | 2 | Three-way discrimination table replacing "two details" | yes — Table 1 | | 3 | State diagnostic disconfirmers; fix Bail comparison class | yes | | 4 | Enter omitted literature in the paper (Ahler & Sood; egocentric/attributive; L&M media account; decoupling; Paluck 2019; Klar et al.; MiC graded) | yes | | 5 | Figure fixes: fig2 dots/bounds; fig6 rows 1 and 8 | yes | | 6 | Number corrections (Boxell 12; megastudy base rates; Paluck; Moore-Berg hedge) | yes | | 7 | Wager specification + mechanism arm | yes | | 8 | 1930s: non-evidence framing; answer Samuels | yes | | 9 | Marriage statistic caveats | yes | | 10 | §6 hydraulic axiom deleted; functional substitution | yes | | 11 | §9 two-track recast (A: outside, B: inside) | yes | | 12 | AA harmonization (adherence, not outcomes) | yes | | 13 | Scope reconciliation | **partial — see below** | | 14 | Evidence hygiene (hate-crime caveats; "measurable share"; sexism citations) | yes | | 15 | Figure typography for approximate points | yes | | 16 | "internal, not independent" on the trial record | yes | | 17 | §7 credit sentence disclaimed | yes | | 18 | Epigram hygiene rule | applied | **Order 13, partial compliance — recorded rather than papered over.** The structural half is done: §7 is out of the core entirely, §2's biography is compressed, §5/§6 asides are edition-specific, the core's reference list no longer cites works the core never uses, and the two-tier design is stated in both front matters. Core fell from 76% to 61% of full. The numeric half is not met: core 6,442 raw vs ≤5,200 (24% over), full 10,599 vs 9,500 (12% over), and the full edition moved *away* from its target. Reason, stated plainly: orders 2–4 and 7–8 required roughly 1,800 words of new material the paper did not previously contain — Table 1, the Ahler & Sood account, the stated disconfirmers, the two separated causal chains, the Samuels answer, and the fully specified wager — while the same order 13 defines the core as "the argument entire: every claim, every objection, every commitment." Those two instructions pull in opposite directions, and a late verification pass confirmed the cost of resolving it the other way: four commitments (the AA retention claim, the individuation-trade concession, the §10 wager sentence, and the overdose qualifier that cuts against the paper's own direction) had been squeezed out of the core and had to be restored. Judgment: keep the commitments, report the overage. A v3 that wants the number should cut §2's map paragraph and §6's data recital — which Figures 1 and 5 already carry — and not §4 or §8. --- ## 6. ERROR LOG Every number that was wrong in an earlier draft, and how it was corrected. Corrections here are often in a non-obvious direction — without this log they silently revert. 1. **Boxell country count.** v1 prose and `data.py` said "nine OECD democracies" while citing the REStat 2024 publication, which covers **twelve**. `research_notes.md` had recorded the uncertainty as "9–12" and the draft silently resolved it the wrong way. v2: twelve, with the nine-country 2020 working paper noted. (VERDICT order 6a.) 2. **Megastudy base rates missing from prose.** `data.py` held antidem_worked = 6/25, violence_worked = 5/25, violence_best = −2.79 pp; v1's prose said only "the single best lever," which is true of the ranking and misleading about the magnitude. v2 prints all three in the same sentence. (Order 6b.) 3. **Paluck cited only where it convicted the rival.** v1 used Paluck et al. 2021 to deflate implicit-bias training but leaned on Pettigrew & Tropp 2006 for contact without the same auditors' 2019 re-evaluation. v2 adds Paluck, Green & Green 2019: **d = 0.39, CI 0.23–0.55** in the credible subset, with the age and domain caveats. Note the direction: the re-audit *sobers* contact, it does not gut it. (Order 4e / 9.) 4. **Moore-Berg hedge drift.** v1 rendered the same finding three ways ("double", "more than twice", "at least 70 percent"). Canonical phrasing is now fixed in `data.py` as `CANON['moore_berg_hedge'] = "more than double"` and reused. (Order 6d.) 5. **fig2 manufactured alignment.** v1 drew More in Common ("almost twice", i.e. **below** 2) and Moore-Berg ("more than double", i.e. **above** 2) as identical 2.0 bars, producing the visual uniformity the "×2 pattern" name traded on. v2 plots them as open markers at 1.85 and 2.2 with direction whiskers and the studies' own wording. The computed Levendusky–Malhotra value (0.38/0.18 = 2.1) is the only filled point. 6. **fig6 scorecard row 1** graded a compression statistic (10 bits/s) as if it certified Jung's purposive unconscious. Now "KEPT (in part) — as scale and automaticity of processing." Row 8 cited the paper's own contested reading of the ×2 pattern as the audit's verification — circular; regraded to PARALLELS on enantiodromia-as-description alone. 7. **Marriage statistic served raw.** The 5/4% → 49/33% comparison spans two different instruments and partly measures aversion to partisan *intensity* (Klar, Krupnikov & Ryan 2018). v2 reports it at reduced strength with both caveats attached. 8. **"A measurable share" of hate/racism/sexism.** No share is measured anywhere in the literature. v2: "an unmeasured but experimentally demonstrated contribution," plus an explicit sentence that no partition exists and the paper claims none. 9. **Quote hygiene (caught during the v1 sweep, recorded here so it stays caught).** "Until you make the unconscious conscious, it will direct your life and you will call it fate" is **not Jung's sentence** — no citation exists. The genuine passage is *Aion* CW 9ii par. 126. Likewise "projection makes the whole world a replica..." is a paraphrase; the verbatim is par. 17. Both are flagged in `data.py` (`PARAPHRASES_DO_NOT_USE`) and the paper debunks the first in its own text. 10. **Popper scope.** *Conjectures and Refutations* (1963) names Marx, Freud, and Adler — **not Jung**. Verified against the chapter text. The paper states this rather than repeating the common misattribution. 11. **Dangling cross-reference in a shared figure (caught at final verification).** `fig6_scorecard.png` row 8 cited "(§7)" as its evidence. The figure ships in both editions, but §7 is full-only — so in the core edition a scorecard verdict pointed at a section that does not exist, and after order 5(c) struck its other support, cited nothing at all. Rule learned: **a shared figure may not cross-reference an edition-specific section.** Fixed by making the cell self-contained. 12. **Commitments lost to compression (same pass).** Cutting the core toward order 13's word target silently removed four things the front matter promises every core reader: the AA retention claim with its no-clinical-claim disclaimer, the individuation-trade concession, the §10 wager sentence (in a section titled "The wager"), and the "even counting 2024's welcome drop in overdoses" qualifier — the last being the charge-8 pattern recurring, counter-evidence sitting in `data.py` and absent from prose. All four restored. Rule learned: **compress history and depth, never commitments and qualifiers.** 13. **fig2 plot positions were not in the ledger.** The two qualitative studies are drawn at 1.85 and 2.2 to show direction; those offsets were hard-coded in `generate_figures.py` against the file's own rule that every drawn number lives in `data.py`. Now `FIG2_POSITIONS`, explicitly statused as "drawing positions, not data." 14. **Table 1's caption skipped its own inconvenient row.** The caption accounted for rows 1–2 and 4–5 and silently omitted row 3 — the own-party-composition-error row, which cuts *against* defensive projection. Caption now states what row 3 does to the thesis. --- ## 7. Standing UNVERIFIED items (do not promote without a source) Carried from `research_notes.md` §10: Memories, Dreams, Reflections page numbers; Baldwin Library of America page numbers; Hodgdon IFS pilot N and remission percentage; Jungian session-fee range; "most therapists per capita" (avoid entirely); a Jung quote warning against rigid classification; CW 11 shadow paragraph number; IAAP's own analyst headcount; whether Jung's formal severance from the society was 1939 or 1940; Fernbach 2013 (failed preregistered replications — used only with the caveat, or dropped). --- ## 8. If there is a v3 The verdict's own bottom line is the standing instruction: the diagnostic claim must remain an adjudicable contest, not a verdict. Beyond that: - run the numeric half of order 13 (cut §2 and §6, not §4 or §8); - the paper still has no *positive* discriminating evidence for projection over egocentric inference — only two suggestive results. If a study appears that measures own-trait/own-affect against out-party attribution with media diet controlled, it decides the paper's central question either way; - external human review has still not happened. The trial record is internal.