Bichitr — Jahangir Preferring a Sufi Shaikh to Kingsجهانگیر · Mughal, c.1615–18 · opaque watercolour, gold & ink on paper, 25.3×18.1 cm · Freer Gallery of Art F1942.15a

Entry 02 — Eye-first reading & the four pre-registered hypotheses

Words before numbers, and four bets the instrument could lose

Structural précis at the plate: an enormous gold disc upper-centre; the emperor enthroned before it at centre-right as the dominant single figure; a descending column of four diminishing subordinates down the left; a flat jewel-patterned carpet ground; a vertical hourglass spine to putti at the base; a pale, quiet sky above. Centred dominance with a left-weighted subordinate column. The scale gradient is the message. Four hypotheses were pre-registered before measuring:

Entry 03 — Run 1: the raw spine & the trust gate

The instrument reads the picture faithfully, and the finding is already visible

Raw read, no persistence, the painting crop (runs/run_read.py, core untouched). Kernel delta (+0.064, +0.226): the mass centroid sits low, not up at the halo; mu 0.884; mask_mode REGION_FIELD; tonal tau 0.630; Notan field_void 0.045, mass_fraction 0.955. Raw and edge-aware agree to three decimals (the crop is clean, I-1 not triggered); PM verdict "meaningful" but the kernel is robust to it.

Entry 04 — H1: hierarchy-as-mass (the World-Seizer, measured)

The rank is confirmed; the way it is built is the finding

Per-figure Notan mass (painter's mass) and spectral saliency integrated over region masks (runs/run_h1_hierarchy.py); massDens is a region's mass-share divided by its area-share (1.0 = frame-average density). Regions are approximate; the claims rest on orderings and the halo's role, and the density result was re-checked on figure-only pixels.

Entry 05 — Adversarial re-check (the rug, the whites, the dissolve)

Pushed on the reading, three of RP's eye-first calls beat the aggregate fields

RP pushed back: the rug may be a strong pull; the whites may be the structural mover; the emperor tonally disappears into the gold; and the accent/island layer may not understand these interactions. Each was tested (runs/run_h1_adversarial.py).

Entry 06 — The tonal-armature lens (candidate C-13)

Building the reading the instrument was missing, then testing that it discriminates

The accent lens returned accents=1: it had no grip on this picture's tonal organisation. So a bolt-on lens was written (toolkit/tonal_armature.py, core untouched, feeds nothing downstream) to read three things the accent lens does not: the near-neutral white armature (lean, elongation, anchor), the signed left/right tension between the white-centroid and the mass-centroid, and the dissolution plateau (the largest light region of low figure-ground contrast, a mass that melts into its ground).

It discriminates on the axes that matter: the light anchor is in a different place for each (a left figure, an upper-right void, a centred figure), and the tension is high for the two off-balance works but low for Klimt (whose whites and mass coincide). On dissolution Bichitr's emperor is the clear dissolver (0.008); Jeong Seon pops (0.035); Klimt's 0.023 leans pop but is only two-thirds of the way there (the dissolve/pop boundary sits near ~0.02), a clean separation from Bichitr on anchor and tension, a weak one on dissolve. The third row (Klimt) was run later as a deliberate predicted-null control; see the null-control entry. Two operationalizations, one phenomenon: this C-13 plateau measure (0.008) and the Entry-05 H1 region-mask ring (0.010) both report the emperor melting into the gold, at different scopes; the number should always name its subsystem.

Entry 07 — H2: the gold halo, reserved or worked?

Worked by coherence, flat by contrast, the two halves of the dissolve

void_typology (C-11), its first non-paper pole. Crops of the halo's gold band against a clean open-sky control. The gold disc is ~84% flat burnished field carrying a coherent worked sun-burst (coherence 0.74 over the full band, 0.993 when the sun-burst is isolated, so the coherence is concentrated in the rays); open sky is 99.8% flat with zero coherent structure (coherence 0.000). The tool discriminates: two large uniform-ish fields, but only the halo carries coherent worked form.

Entry 08 — H3: the Holbein-inverse (modelled faces, flat carpet)

The phenomenon holds; the metric I predicted was wrong

Per-region construction split, faces vs carpet. Confirmed: faces carry ~1.6× the smooth modelling relief of the carpet (0.094 vs 0.058), and the carpet carries ~2.6× the high-frequency chroma pattern of the faces (0.057 vs 0.022). Smoothly modelled figures on a busy, non-recessive ornamental ground, the Holbein-inverse Weekes describes, measured.

Entry 09 — H4: the strata (figure vs ornament), and a declined date

The instrument confirms the regime split and refuses the provenance call

The 5/3/1 crop ladder. Cumulative (whole-frame reads): as the decorative strata are added, cohesion collapses (mu 0.884 → 0.734 → 0.313) and off-centre mass symmetrises (|delta| 0.235 → 0.069). Isolated (A/B/C as separate zones): Layer A is focal (angular isotropy 0.62, low edge density), the ornamental bands are ring-like and edge-dense (B 0.85 / C 0.94 isotropy, ~1.4× A's edge density). Two structural regimes on one folio: figure vs ornament.

Entry 10 — Null control: is any of this just the image's statistics?

The core findings are compositional, not distributional

Every metric above is measured on the intact arrangement, so a hostile reader can ask whether any of it is just the picture's statistics. Two arrangement-destroying nulls answer it: block-shuffle (64px blocks permuted, preserves the histogram exactly) and phase-scramble (preserves the power spectrum). The spine re-run on each (n=3 seeds each, ±SD).

The C-13 lens's arrangement claim was checked the same spirit, by prediction rather than scramble: run on a third work chosen to look different (Klimt's centred, popping Mäda Primavesi), it landed where predicted, centred anchor and low tension, moving C-13 to n=3 (Entry 06 table).

Entry 11 — The meta-finding

Every hypothesis confirmed its phenomenon and corrected its metric

Four pre-registered hypotheses, all run and pressure-tested. On this one folio, all four bets resolved the same way, and that consistency is the study's most transferable observation, a description of these four bets, not a law about the instrument.

In every one the phenomenon survived and the metric was corrected. Read narrowly, and that is the right way to read it, that is the mark of an honest eye-first reading on this work: the eye named real structure, the first metric for it was too tidy, and holding the instrument to measurement rather than to confirming the reading traded a clean claim for a true one. That all four resolved this way here is evidence the reading was honest, not a regularity the ISR guarantees, a study fitted to its own numbers would have no corrections to show. Whether the same pattern recurs on the next work is a question for the next work. The critique is to be written from this record; nothing here is promoted to the instrument's numbered findings yet.