### BIEA24 — Eagle Island
**Date:** 2015-03-16 (dive window 14:32–15:12, from frame EXIF; Olympus TG-2)
**GPS:** 6.20961°S, 71.30426°E (the manifest's `gps` field is malformed — see flag 6)
**Frames reviewed:** 15 of 15 (all of `BiEa24-finished`) — **but only 11 carry CoralNet annotation, and only those same 11 were run through Coralscapes; every cover number below rests on those 11.**
**Coral cover:** 12.7% (formal transect surveys) · **% bleached:** **20.6%** of coral (CoralNet analysis of informal survey images; Mayfield & Dempsey 2026, Oceans) — the manifest records `pct_bleached: null`, so the sources disagree; see the flags
**Reef type:** fringing, intermediate exposure · **Temperature:** 29.3 °C
**Live hard coral cover across methods** (n=11 of 15 photos): transect 12.7% · CoralNet points 21.5% · Coralscapes 35.3%. Deviation vs transect +8.8 to +22.6 pp. **Mean of 3: 23.2% ± 11.4.** **The two photo methods disagree by 13.8 pp — the widest model-vs-model split of the five sites in this redo — and that split is entirely bleached tissue.** Coralscapes assigns **0.00% bleached area in all 11 frames**; add the human's 11.5% bleached points back into human live and the two read **33.0% vs 35.3%.** See flag 1.
**Coral genera identified:** 6 scleractinian groups — *Porites* (massive/domal, lobate-columnar and knobby digitate — the dominant builder, 64.8% of live-coral points), *Acropora* (**golden-tan corymbose with bright white branch tips**, plus cream plating/tabular forms), foliose plates with large well-spaced conical corallites compatible with *Turbinaria*, ridged rugose plates compatible with *Echinopora*/*Merulina*/*Pachyseris*, encrusting sheets compatible with *Montipora*, and unresolvable submassives; plus **antipatharian whips** and a bushy octocoral (P3160320), crinoids, ascidians and sponges (8 frames). Photo-based calls, hedged.
**Representative photos:** P3160321.jpg, P3160337.jpg, P3160344.jpg, P3160348.jpg, P3160350.jpg

**BIEA24 is a wall-then-terrace dive, and the camera changes job halfway through.** It opens at 14:32 on a **near-vertical pale carbonate face falling into deep blue** (P3160320): a brown ridged foliose plate and a yellow-green tabular colony on the lip, antipatharian whips trailing off it, surgeonfish and white-tailed damselfish alongside. Both read it low and agree (human 6.7% live, Coralscapes 14.7%, 47% of points on water) — the only genuinely water-dominated frame at the site. Thirteen minutes on, the dive is on the reef proper: **P3160321, the best wide benthic frame**, a bommie of olive-green foliose and scrolling plates with orange anthias across it. Human 36.7% live, Coralscapes 28.3% — the model reading *below* the human, worth holding onto.

**A burst of five frames in 44 s (14:58:54–14:59:38) works one overhang.** P3160334 establishes it: a ridge of brown digitate colonies over a maroon foliose plate, anthias, a dark-blue triggerfish compatible with *Pseudobalistes fuscus*, and an adult **emperor angelfish (*Pomacanthus imperator*)** at the base. It carries the next two frames: **P3160336 is the observation worth keeping: it hangs stationary with a small striped wrasse compatible with *Labroides dimidiatus* at its flank — a cleaning station mid-service**, over a **dense school of small golden fish** compatible with apogonid cardinalfishes or sweepers. **P3160337 is the highest-cover benthic frame at the site** (human 46.7% live, Coralscapes 43.3%): a broad pale grey-tan massive dome compatible with *Porites*, knobby digitate colonies, feather stars, squirrelfish, and — the detail that matters — **sharply-edged pale patches of recent partial mortality, with small pink-red lesions among them.**

**The last nine minutes are flash macro, and this is where the models part company.** P3160340 is a frame-filling close-up of a foliose plate margin — **brown-red tissue grading to teal-green, conical corallites with pale white oral discs**, compatible with *Turbinaria*. The annotator scored **16 of 30 points bleached (53.3%)**; Coralscapes read it 81.7% live, 0% bleached. **The tissue is plainly pigmented — the white is corallite mouths, not bare skeleton — so here the model is closer to right than the annotator.** P3160343 and P3160344 are the finest images at the site and **they are misidentified in the master entry**, which calls them "a pair of red-and-cream mottled moray eels". They are not eels. They are a **single coral guard crab**, front-on among its host colony's branches: cream-white carapace and limbs blotched with crimson, stalked yellow-green eyes, orange-tipped dactyls — a ***Trapezia rufopunctata*-type red-spotted guard crab**, and **the sharpest invertebrate portrait in the Chagos redo so far.**

**And the closing sequence is a moray, not a puffer.** The master entry calls P3160347–351 a puffer compatible with *Arothron*. The animal in P3160348–351 is an eel — elongate head, small forward-set eye, tubular anterior nostril, jaw sloping back past the eye, dark ground densely spotted white — **compatible with a whitemouth moray, *Gymnothorax meleagris*-type.** P3160347, 26 s earlier, is a blurred close-up of the same head, **mouth agape and pale lining showing** — almost certainly the same animal. It sits between a **large pink-cream lobate-columnar *Porites*-type colony** and a **golden-tan corymbose *Acropora* with brilliant white branch tips**, over pavement carrying pink coralline crusts and **inky black patches compatible with an excavating sponge**.

<b>Assemblage composition.</b> Across 11 analysed frames the Coralscapes masks put live hard coral at 35.3% of image area, of which massive and meandroid is the largest component at 47.7%; the rest is other hard coral, 46.0% of live cover. The independent 30-point human annotation of the same frames (330 points) puts live hard coral at 21.5% of points and apportions it <i>Porites</i> 64.8%, <i>Acropora</i> 21.1%, other hard coral 14.1% of live coral. Non-coral cover is dominated by crustose coralline algae 3.3%, turf 13.6%, macroalgae 4.2% of points. Structural loss registers as dead coral 2.3% of image area and rubble 3.7%. Bleaching stands at 11.5% of annotated points — **38 points** — concentrated in other hard coral 9.1%, <i>Acropora</i> 2.4% of points. Frame-to-frame live cover runs 0% to 46.7% by point count and 4.1% to 81.7% by mask. Conspicuous fauna and sessile invertebrates appear as sponges in 8 frames, sea cucumbers in 2, giant clams and anemones in 1 each.

**Flags:** (1) **THE 13.8 pp MODEL-VS-MODEL SPLIT IS ENTIRELY BLEACHED TISSUE — a label-definition problem, not a segmentation one.** Coralscapes returns 0.00% bleached everywhere and folds pale tissue into live coral. **On the 7 frames without a bleached signal it reads 2.1 pp BELOW the human; on the 4 that carry bleached points — P3160340, P3160347, P3160348, P3160350 — it reads +41.4 pp ABOVE (+68.3, +31.0, +26.1, +40.2).** **Do not report BIEA24 as evidence the photo methods disagree about coral; they disagree about bleaching**, and run this split at every site. (2) **P3160340 IS A PROBABLE HUMAN OVER-CALL, NOT A MODEL FAILURE** — 16/30 points bleached on a macro whose tissue is visibly brown-red and teal-green. **All four high-bleached frames fired the flash (EXIF flash=9); all seven wide frames did not (flash=16).** Flash-lit close-ups appear to inflate the human bleached score — the mirror image of the blue-water dilution bias at BIEA18/20. (3) **THE WHITE-TIPPED *ACROPORA* IN P3160349/350/351 IS UNRESOLVED** — white tips fit both normal growing axial corallites and early tip bleaching; the annotator scored 5 B_Acro points there. **Needs Anderson's call.** (4) **TWO ANIMAL MISIDENTIFICATIONS IN THE MASTER ENTRY, APPARENTLY SWAPPED** — P3160343/344 are one ***Trapezia rufopunctata*-type guard crab**, not moray eels; P3160347–351 are a **whitemouth moray, *Gymnothorax meleagris*-type**, not an *Arothron* puffer. (5) **15 FRAMES ON DISK, 11 SCORED — the four unscored are missing from BOTH methods.** P3160343/344/349/351 have no CoralNet points and no Coralscapes `_pred.png` under `regions/Chagos/`. 349/351 near-duplicate the annotated 348/350 and would only reinforce the signal, but **343/344 are flash macros in which coral fills most of the frame, and the one comparable macro that IS scored returned 81.7%. Adding all 15 would push Coralscapes higher and WIDEN the split; n=11 is the conservative subset.** (6) **MALFORMED MANIFEST FIELD** — `chagos_work_manifest.json` stores `gps` as `6.20961°S, 71.30426°E · **In-situ temperature:** 29.3°C`: a coordinate with temperature markdown concatenated on, which breaks any automated GPS parse. Temperature is correctly duplicated in `temp_c`. Sweep the manifest for this pattern. (7) **"% BLEACHED" DOES NOT RECONSTRUCT** — master 20.6% of coral, manifest `null`, and neither obvious denominator gives 20.6 from these 11 frames (11.5% of all points; 34.9% of live+bleached), so it must come from a larger CoralNet run. Same unresolved question as BIEA24.5/BITB34.5. (8) **MORTALITY AND BIOEROSION** — P3160337's pale sharply-edged patches and pink-red lesions on a *Porites*-type dome (3.9% dead) read as a pigmentation response to injury or boring organisms, **not** a named disease; P3160348 holds the site's highest dead-coral value (11.7%); P3160350/351 show inky black pavement patches (excavating sponge?). (9) **HIGH UNKNOWN FRACTION — 25.5% of human points, peaking at 56.7% on P3160334, which scores 0.0% live despite plainly showing a coral-covered ridge.** Renormalise without Unknown and the human read rises to **28.3%** live, or **41.8%** counting bleached tissue as coral — *above* Coralscapes. **The CoralNet column is depressed by unlabelled points, not by the reef.** (10) **No methodology imagery** in any of the 15 frames — no diver, tape, quadrat or instrument — worth recording, given BIEA20 holds eight. As a curated highlight set, none of the three cover numbers is an unbiased areal estimate. (11) No per-frame depth data; no COTS.

**Benthic data:** [Download BIEA24 benthic survey data (CSV)](../benthic_data/BIEA24_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).