### BITB32 — South Brother
**CRD page:** [Three Brothers & Central Chagos Coral Surveys (Mar 17-20, 2015)](https://coralreefdiagnostics.com/three-brotherscentral-chagos-20150317-to-0320)
**Date:** 2015-03-19 (dive window 08:18:17–09:06:06, 47 min 49 s, from `DateTimeOriginal` EXIF on all 19 files. The counter runs P3190394→P3190441, so the edited set keeps **19 stills of ~48 exposures**, plus two movies shot inside the lobster encounter. The frames are badly skewed in time: **five cover the first 41 minutes, fourteen cover the last six and a half.**)
**GPS:** 6.17152°S, 71.59340°E
**Frames reviewed:** 19 of 19 (100% of the site folder, `BiTb32-finished`; all on disk, 0 evicted, no EPERM, and all 19 carry CV rows — complete coverage)
**Coral cover:** 30.25% (formal transect surveys; algal cover 47.92%) · **% bleached:** 7.56% (`% bleaching (2nd training)`, Mayfield & Dempsey 2026 supplement)
**Reef type:** submerged lagoonal bank reef, protected, inside the lagoon · **Temperature:** 29.4 °C. Three depth strata were surveyed and they disagree — **D2 36.50% coral, D3 26.73%, D4 27.52%** — and 30.25% is their mean.
**Live hard coral cover across methods** (n=19 of 19 photos): transect 30.25% · CoralNet points 13.3% · Coralscapes 12.8%. Deviation vs transect **−17.0 to −17.5 pp**, the largest shortfall of the Three Brothers series on complete coverage. **Mean of 3: 18.8% ± 9.9.** The two photo methods land **within 0.5 pp of each other** while sitting ~17 pp below the transect, and their per-frame correlation (r = 0.831) is the highest of the redo — they agree with each other unusually well and disagree with the transect almost equally. **The published paper already knew:** the supplement's own informal estimate is 16.65%, a **58.0% formal-vs-informal difference — 10th largest of 105 sites against a corpus median of 10.9%**. See flags 1–3.
**Coral genera identified:** the transect recorded **30 scleractinian genera — 15th richest of 109 Chagos sites (median 25)** — and this is a faviid/merulinid reef, not an *Acropora* reef: ***Porites* 4.91%, *Pocillopora* 4.30, *Favia* 4.20, *Platygyra* 3.75, *Montipora* 2.79, *Echinopora* 2.17, *Favites* 1.34, *Acropora* only 1.03**, then twenty-two more at ≤0.8% each (*Montastrea*, *Acanthastrea*, *Echinophyllia*, *Astreopora*, *Pavona*, *Symphyllia*, *Hydnophora*, *Stylophora*, *Fungia*, *Goniastrea*, *Cyphastrea*, *Galaxea*, *Ctenella*, *Leptastrea*, *Psammocora*, *Turbinaria*, *Leptoseris*, *Coscinaraea*, *Cantharellus*, *Mycedium*, *Pachyseris*). **The photographs corroborate the habits but not the names**: fine-valleyed meandroid brains compatible with *Platygyra*/*Leptoria* (P3190437, P3190438, P3190441), tan knobby-columnar and massive colonies compatible with *Porites* (P3190417, P3190437, P3190441), coarse-corallite domes compatible with *Favia*/*Favites*/*Astreopora* throughout, ridged plates compatible with *Hydnophora*/*Merulina* (P3190438, P3190441), grey foliose sheets compatible with *Echinopora*/*Pavona* (P3190401), encrusting sheets compatible with *Montipora*, mauve clumps compatible with *Pocillopora*, and one branching colony compatible with *Seriatopora*/*Stylophora* (P3190394); plus *Halimeda*, sponges, ascidians and heavy pink CCA. **There is not a single tabular *Acropora* in the 19 frames**, and Coralscapes assigns 0% of live cover to its tabular class — the opposite of BITB27, BITB29 and BITB30. Growth-form calls from photographs are given as "compatible with"; the percentages are transect IDs.
**Representative photos:** P3190433.jpg, P3190417.jpg, P3190437.jpg, P3190426.jpg, P3190439.jpg

**BITB32 is the lobster dive, and that governs the whole photographic record.** From 09:02:45 the camera finds **two painted spiny lobsters (*Panulirus*-type — white-reticulated carapace, olive-green abdomen with white transverse banding, black-and-white ringed legs, pink antennular peduncles; compatible with *P. versicolor* but not callable to species)** walking in the open over pavement in broad daylight, and stays with them for the rest of the dive. **Nine of the nineteen edited frames are composed on lobsters** — P3190429 and P3190430 wide with both animals, P3190432 closer, **P3190433 the best of them**, P3190434 and P3190435 from above with a mottled brown five-armed sea star below, then P3190437, P3190438 and P3190441 on a single animal. Two movies were shot mid-sequence. Add the two giant-clam macros and the colony macro and **twelve of nineteen frames are composed on an animal or a single colony rather than on the reef.**

**The reef those frames do show is a low-relief, high-diversity, small-colony bank.** **P3190401 is the best statement of it** — a slope of dark massive domes, grey foliose plates, brown digitate and encrusting colonies and mauve *Pocillopora*-type clumps packed shoulder to shoulder over purple-pink CCA pavement, with a checkerboard wrasse (*Halichoeres hortulanus*-type) and a yellow sponge. **P3190417 is the highest-cover frame** — tan lobate *Porites*-type, pink-margined encrusting *Montipora*-type, corymbose *Acropora*, coarse-corallite faviids and *Pocillopora*, scored 53.3% live. **P3190437 and P3190441 are the best genus frames**, each carrying a big tan knobby-columnar colony, a fine-valleyed meandroid brain, *Astreopora*-type domes and a ridged *Hydnophora*/*Merulina*-type plate in one view. Against these, **P3190416, P3190425 and P3190428 are genuinely thin** — broad pale consolidated pavement and rubble carrying small, well-spaced colonies, Moorish idols and a parrotfish over them in P3190416, a fusilier school over P3190425. P3190428 reads 3.3% live and looks it. **That spread is real, not an artefact, and the depth strata say so independently** (D3 26.7% against D2 36.5%) — but even the poorest stratum is twice the photo estimate.

**Structural loss is old and unremarkable, and there is no acute damage anywhere.** Coralscapes puts dead coral at 2.79% of image area and rubble at 4.92%; the pavement in P3190430, P3190432, P3190435 and P3190441 is maroon-and-pink CCA plated over old framework, ordinary turnover. **No crown-of-thorns starfish in any of the 19 frames. No disease lesions. No predation scars.** The only mortality worth a name is a maroon dead patch on the meandroid brain in P3190441. **The invertebrate record, by contrast, is exceptional.** Beyond the lobsters, **P3190426 is a large live *Tridacna* with its mantle fully open**, olive-and-black patterned, wedged into CCA rock — the best clam frame of the redo — and **P3190427 appears to be a second, dead and CCA-overgrown clam shell** a minute later, its fluted margins still legible under pink crust. **P3190439 puts a requiem shark over the reef in open blue water**; the dark-margined caudal and plain first dorsal are **compatible with grey reef (*Carcharhinus amblyrhynchos*)**, but the BIEG04 lesson applies and it is **not callable from this frame**. Blue encrusting sponges (*Haliclona*-type) recur in P3190433/434/435/441. Fish are otherwise sparse and small. **No survey apparatus appears in any frame** — no tape, quadrat, diver, slate or collecting gear — and **P3190394 is the one specimen macro**: a small olive-green branching colony with pale tips on CCA-crusted pavement. The CRD site page records that **one *Pocillopora acuta* (ChPa21) and one *Seriatopora* sp. (ChSe13) were sampled here** — both target pocilloporid groups — and the transect's **0.00% for *Seriatopora*, despite a colony being collected**, is itself an under-detection note. P3190394 is the only candidate frame for either sample and is not resolvable at this resolution. The point annotation on it calls it *Porites* and *Acropora*, neither defensible.

<b>Assemblage composition.</b> Across all 19 analysed frames the Coralscapes masks put live hard coral at 12.84% of image area, of which massive and meandroid is much the largest component at 48.5%; the rest breaks down as other hard coral 13.6%, <i>Acropora</i> (non-tabular) 11.9%, <i>Pocillopora</i> 9.6%, meandroid 7.4% of live cover — <b>and tabular <i>Acropora</i> 0.0%</b>. The independent 30-point annotation of the same frames (570 points) gives live hard coral as 13.3% of all points and apportions it <i>Porites</i> 40.8%, other hard coral 30.3%, <i>Acropora</i> 21.1%, <i>Pocillopora</i> 7.9% of live coral — <b>but the transect puts <i>Acropora</i> at 3.4% of live cover, so the point set over-calls it roughly six-fold</b>. Non-coral cover is dominated by turf 32.3% and CCA 17.2% of all points, sponge 0.5%, and <b>sand, rock and dead coral at 0.0% — not one point in 570</b>. Structural loss registers as dead coral 2.79% of image area and rubble 4.92%. Bleaching stands at 1.4% of annotated points — <b>8 points in total</b> — as other hard coral 1.05% and <i>Acropora</i> 0.35%, against a 7.56% transect figure; Coralscapes reads bleached at 0.03% of image area, maximum 0.19% on any frame. Frame-to-frame live cover runs 0.0% (P3190430, P3190439) to 53.3% (P3190417). Conspicuous fauna appear as sponges in 17 frames, sea cucumbers in 9, urchins in 6, giant clams in 2, anemones in 1.

**Flags:** (1) **THE UNKNOWN-DENOMINATOR CORRECTION FAILS AT BITB32 — THE FIRST LOW-READING SITE WHERE IT DOES.** All 19 frames are scored, so there is no subset excuse. But the Unknown pool here is **22.28% (127 of 570 points)**, well short of BITB29's 32.24%, and **dropping Unknown from the denominator lifts the point set only 13.33% → 17.16% (18.96% with bleached added) against a 30.25% transect — residual −13.1 pp, about a fifth of the gap.** At BIEA26, BITB27, BITB28 and BITB29 the same correction landed *on* the transect. **BITB32 does not conform; the correction is not general.** (2) **THE DEFICIT IS A CLASSIFICATION PROBLEM, NOT A DENOMINATOR PROBLEM, AND THE ALGAE PROVE IT.** On the raw denominator the point set's turf + CCA is **50.2% against a 47.92% transect algal cover — nearly exact.** Restricted to the 373 genuinely benthic points, turf + CCA rises to **76.7% vs 47.92%**, coral falls to **20.4% vs 30.25%**, and bare substrate and invertebrates — 11.57% and 10.26% by transect — read **0.0% and 0.8%**. The label set is absorbing bare pavement, rubble and sessile invertebrates into TURF, and some coral with them. **Rescaling a denominator cannot fix that; only relabelling can.** This is the cleanest counter-example in the redo to the Unknown-pool explanation and belongs in the corpus-wide re-derivation. (3) **EVERY SUBSET STILL READS LOW.** Excluding the 2 high-water frames: 14.31/13.87. Excluding the 3 specimen macros: 13.75/12.70. Excluding the 6 close lobster frames: **12.05/10.96 — lower.** Survey-only 8 frames: 14.17/11.69. Full substrate renormalisation: 20.38%. Mask renormalised over non-background pixels: 18.69%. Nothing reaches 30.25%. (4) **BLUE-WATER DILUTION REJECTED, harder than anywhere.** Site mean water is **10.5%, the lowest of any site checked**, and only 2 of 19 frames are ≥40% water (P3190425, P3190439). Dropping both moves the point set 13.33 → 14.31 and the mask 12.84 → 13.87. No BIEA20-style sign flip. (5) ***ACROPORA* IS OVER-CALLED SIX-FOLD** — 21.1% of live points against 3.4% of live cover on a reef with **no tabular *Acropora* in any frame**. This is the direct source of the phantom bleached-*Acropora* points in flag 6. (6) **BLEACHING — THE PUBLISHED 7.56% IS NOT CORROBORATED, AND THIS IS A MARCH SITE.** Anderson recorded **no bleaching in March or April, only May**, and the half-month means agree (6.1/8.2/8.4/10.0% through March–April against **31.8% in early May**). Both photo methods are far below the published figure: the point set gives **1.4% (8 of 570 points)** and Coralscapes **0.03% of image area, maximum 0.19% on any frame** — effectively nothing. The bleached points are also on the wrong frames: **5 of 8 fall on frames scored ≤13.3% live**, and **the only bleached-*Acropora* pair sits on P3190418**, a wide view of dark massive coral and pale sand that is **43.3% Unknown, 20% water and shows no *Acropora* at all**; another sits on **P3190426, the giant-clam macro**, where the pale shell margins are the brightest thing in frame. Report 7.56% as the published transect figure and state that the diver who made the surveys recorded no bleaching before May and that the photographic record does not corroborate it. (7) **BOTH METHODS CLASSIFY THE SPINY LOBSTERS AS FISH — a new failure mode.** Coralscapes' five highest "fish" frames are **P3190433 (7.75%), P3190434 (4.93%), P3190438 (4.10%), P3190432 (2.36%), P3190441 (2.26%)** — every one a close lobster frame; the point set does the same, with **3 FISH points on P3190426 (the clam macro, no fish present)**, 3 on P3190434, 2 on P3190433. At BITB29 the mask read the same animals' leg banding as *bleached coral*; here they go to the fish class. (8) **P3190430 IS SCORED 0.0% LIVE ON A FRAME PLAINLY CARRYING 15–20% CORAL** (faviid domes, *Pocillopora* clumps, encrusting colonies; mask 6.1%). **P3190439 is worse** — 16 WATER + 12 Unknown + 2 CCA exhausts the whole point budget for 0.0% live, on a frame whose lower third is a coral-dominated slope. Annotation-provenance artefacts; ~76% of Chagos point annotations are robot predictions. (9) **P3190427 IS THE ZERO-UNKNOWN CONTROL** — the only frame of 19 with no Unknown and no water points, and it is a flat, evenly-lit, full-frame benthic macro. **The 22.3% site-wide Unknown pool is a property of frame geometry and lighting, not of the reef.** (10) **NON-SURVEY IMAGERY IS 63% OF THE SET (12 of 19)** — nine lobster frames, two clam macros, one colony macro — **but it is not what drives the shortfall** (flag 3), unlike BITB27, where measurement shots were 38% of the CV sample and mattered. **No survey apparatus appears at all.** (11) **LOBSTER SEQUENCE P3190429–P3190441 (nine frames plus two movies), ≥2 *Panulirus*-type animals in the open in daylight** — not callable to species; flagged for illustrative use with **P3190433**. With BITB29's den aggregation this is the second exceptional lobster record of the redo. (12) **P3190426: large live *Tridacna*, mantle fully open** — flagged for illustrative use. **P3190427 is probably a second, dead, CCA-overgrown clam; Anderson's judgement needed** on whether it is the same animal. (13) ***SERIATOPORA* AND *P. ACUTA* WERE BOTH SAMPLED HERE (ChSe13, ChPa21) BUT NEITHER IS IDENTIFIABLE IN THE FRAMES**, and the transect records *Seriatopora* at **0.00% despite a colony being collected**. P3190394 is the only candidate; **Anderson's judgement needed.** (14) **P3190439's shark is NOT callable** to species; the BIEG04 lesson applies. (15) **THE EDITED SET IS 19 STILLS OF ~48 EXPOSURES AND IS SEVERELY TIME-SKEWED** — five frames cover 08:18–08:59, fourteen cover 08:59–09:06. Every photo-derived number here rests on six and a half minutes of a forty-eight-minute swim, most of it spent on two lobsters. **The strongest frame-selection confound of the redo so far.** (16) No per-frame depth data; the supplement's site-depth field reads 28.4 and cover was surveyed at three strata.

**Benthic data:** [Download BITB32 benthic survey data (CSV)](../benthic_data/BITB32_benthic.csv) — scleractinian cover by genus with full names, plus substrate and algal categories (bare substrate, CCA, turf, macroalgae, invertebrates), from the LOF BIOT-1 survey (Mayfield & Dempsey 2026, *Oceans*, supplemental file S1).