1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
|
# Style contract (operative checklist for every section pass)
Adapted from the flossing project's contract. Sources: Shaib et al. 2026 slop
taxonomy (density/relevance/coherence are the load-bearing dimensions);
Buschek's four reviewer complaints; 2026 banned-pattern lists. Applied as a
per-section pass, not a vibe. The number/claim discipline this file enforces
is anchored in `notes/36_evidence_ledger.md` (§9 booby traps) and
`notes/41_paper_plan.md` (§3 structure, §7 rules).
## Hard bans (lexical)
delve, tapestry, landscape (figurative), testament, pivotal, showcase, intricate, vibrant,
seamless, elegant, dramatically, novel (self-describing), comprehensive (self-describing),
notably, crucially, interestingly, importantly, "It is worth noting", "Moreover," as sentence
opener, "In recent years", "has garnered attention", "paradigm shift".
Technical terms that overlap ban lists (e.g., "soft" in "soft ramp", "robust" conditioning)
stay.
## Punctuation & rhythm
- Em dash: ≤2 per section, never two in one sentence, never as paired parenthetical.
- No exclamation marks. Semicolons fine.
- Every paragraph contains at least one sentence under ~12 words.
- No uniform paragraph shapes: don't end three consecutive paragraphs with a summary clause.
## Structure
- The contrast template ("X, not Y" / "not X, but Y") has a total budget of ONE device:
the title ("What Does Random Feedback Cost?") resolves to one echo at the end of paragraph 1
("not free; irreducible but soft"). Zero elsewhere. Express other contrasts by stating the
positive finding and letting the numbers carry the negation.
- Rule-of-three closers only when the three items are an exhaustive measured list
(the three contributions qualify; nothing else does).
- No chronological narration of the investigation. No "we first tried hard-k, then...".
Structure follows note 41 §3.
- Results topic sentences contain a finding or a number, never a plan ("We then examine...").
## Integrity (Buschek's four, mapped to our risks)
- Marketing language: zero unevidenced evaluative adjectives. "essentially exact" only with
the MAE or correlation in the same sentence.
- Performative related work: every citation paired with the precise claim it supports; check
each against `notes/*.md` and the source paper before the citation lands (note 41 §6 lists
the per-cite rationale).
- Misrepresentation: prior-work characterizations quote verbatim with section; the
Song-Xu-Lafferty complementarity is stated as `E_B[H_FA]=0` in the residual direction, not
as a correction of their convergence result.
- Stretched summaries: no interpretive metaphors for results. One flourish per section maximum,
and it must be cuttable without losing content.
## Density (the actual anti-slop weapon)
- Every sentence must add a number, a definition, a caveat, or a logical step. Otherwise cut.
- Modal verbs (could/might/may) confined to Limitations and future work.
- Hedges appear only where the evidence concedes; one hedge per concession, not a seesaw.
## Honesty rails (project-specific)
- Closed-form ramp correlation: cite 0.933 (matched init) and 0.982 (5-init ensemble). The
retired 0.977 never appears. State which is which in the same sentence.
- "No fit" is defined once at first use: no post-hoc scalar tuned to the reported data;
predictions come from initialization or two early kernel snapshots. Never drift from it.
- The estimator is "snapshot-conditional, no fitted parameters" at first use, never
"architecture-only"; its lazy-regime scope (the MNIST eta-sensitivity) is stated where the
estimator is introduced.
- Hard-k is a null model and a reported failure (2x-250x overprediction), never a gap
predictor. `E[e]=k/P` (null model) and `rho=1-E_B[e_0]` (measured share) are distinct
objects; never conflate them.
- Every theorem states its assumptions inline: feedback independence and zero mean, the
output-layer-exact fact, A1 (lazy), A2 (scalar erosion, exact only in the residual direction).
- "Soft ramp / no phase transition" refutes one named hypothesis (redundancy-exhaustion
threshold); it is the analyticity of `gap_T` in the spectrum, stated as such, not a universal
claim about FA.
- MNIST estimator: report MAE first (0.00196, 3.9% relative); the correlation 0.911 appears
only with the gap range (0.040-0.063) that makes it the weaker statistic.
- The teacher test-gap sign flip is scoping evidence for the optimization-gap claim, not a
contribution.
- No promises (code release, future experiments) the authors have not decided. Code+config
release is planned (manifest note 40); state it only in the reproducibility section.
|
