### BIEA26 — Eagle Island
**Date:** 2015-03-17 (dive window 14:38–15:39, from frame EXIF — **the second dive of the 17th and the last Eagle Island site of the mission**, entering the water 1 h 19 min after BIEA25 surfaced at 13:19; P3170439 is the handover frame BIEA25's author flagged)
**GPS:** 6.17221°S, 71.33191°E
**Frames reviewed:** 32 of 32 (100% of `BiEa26-finished`; 0 evicted, all 32 CoralNet-annotated and all 32 matched to Coralscapes masks — the same complete CV basis BIEA25 had, and the reason this site's result cannot be waved away as BIEA24.5's n=5 was)
**Coral cover:** 21.0% (formal transect surveys) · **% bleached:** 8.61% (`% bleaching (2nd training)`, Mayfield & Dempsey supplemental file S1; the manifest's `pct_bleached: null` is a manifest gap, not a contradiction)
**Reef type:** fringing, intermediate exposure · **Temperature:** 29.2 °C
**Live hard coral cover across methods** (n=32 photos): transect 21.0% · CoralNet points 14.9% · Coralscapes 8.9%. Deviation vs transect **−12.1 to −6.1 pp**. **Mean of 3: 14.9% ± 6.1.** **BIEA26 inverts the batch.** At BIEA22, BIEA23 and BIEA25 both photo methods read HIGH against the transect; **here both read LOW**, and Coralscapes sits **6.0 pp below CoralNet** — its second-widest model-vs-model split of the Eagle Island series. Blue-water dilution was the obvious suspect and does not survive testing: see flags 1–3.
**Coral genera identified:** 6 scleractinian groups — ***Porites*** in three guises, smooth massive/domal boulders (P3170444, P3170454, P3170455), knobby-columnar olive colonies (P3170460, P3170467) and **a pale mauve-cream nodular plating form dominating the mid-dive frames** (P3170456, P3170458, P3170459); ***Pocillopora***-type knobby branching colonies, the commonest small coral here (P3170446, P3170447, P3170461, P3170468, P3170469); ***Acropora*** as small corymbose and digitate colonies and low tables (P3170440, P3170442, P3170461, P3170464); meandroid and cerioid brains compatible with ***Platygyra*/*Favites*/*Goniastrea*** (P3170442, P3170444, P3170461, P3170469); thick brown laminar plates compatible with ***Turbinaria*** or a plating ***Montipora*** (P3170445, P3170477); encrusting sheets compatible with ***Montipora*** (P3170442, P3170467). Plus abundant crustose coralline algae, ***Halimeda***, encrusting sponges, and **black feathery arms compatible with crinoids, possibly antipatharians among them** (P3170458, P3170459). Growth-form calls from photographs, given as "compatible with"; **no species is asserted.**
**Representative photos:** P3170442.jpg, P3170447.jpg, P3170461.jpg, P3170469.jpg, P3170480.jpg

**BIEA26 is two reefs photographed in one hour, and the camera spends the second half on the poorer one.** From 14:38 to 14:56 (P3170439–P3170462) the frames are over **a dense, structurally complex reef top** — a mosaic of small, tightly packed colonies on CCA-crusted pavement under clouds of anthias and fusiliers. P3170440, P3170441, P3170442, P3170461, P3170462 and P3170469 are the highest-cover; **P3170461 and P3170469 read to the eye as 40–50% live coral in the benthic half of the frame.** Then from 15:00 the dive crosses onto **a flat, pale, low-relief plain of consolidated pavement, rubble and coarse sand** carrying only scattered colonies — P3170464, P3170465, P3170477, P3170478 and P3170479–482 are all the same barren ground, and it occupies the last twenty-five minutes. **The reef-top eighteen frames average CoralNet 15.9% / Coralscapes 12.6%; the plain's fourteen average 13.6% / 4.2%.** A one-hour swim samples that plain heavily; a transect run once does not. Same structural problem as BIEA25 with the sign reversed — BIEA25's camera finished in *Acropora* thickets, BIEA26's on rubble.

**The macrofauna is the best of the Eagle Island series and nearly all of it is in the second half.** **Four frames hold sharks.** P3170479, P3170480 and P3170481 were shot in fourteen seconds and are one animal: **broad-headed, dorsals set back over the pelvics, caudal long and low — compatible with a tawny nurse shark (*Nebrius ferrugineus*)**; P3170480 is the clearest. **P3170482, forty-four seconds later, shows a slimmer animal with a more forward first dorsal — plausibly a whitetip reef shark (*Triaenodon obesus*) and probably a second individual, but neither call is safe from these frames**, and the BIEG04 precedent argues for leaving both hedged. **A large olive-yellow, densely speckled moray compatible with the giant moray (*Gymnothorax javanicus*)** appears in P3170446 and again eight minutes later in P3170454/P3170455; one animal or two is not resolvable. **P3170464 and P3170465 are portraits of a single mauve-pink, orange-eyed snapper compatible with *Lutjanus gibbus*.** Also a jack or trevally (Carangidae) in P3170478, **raccoon-type butterflyfish (*Chaetodon lunula*-type)** in P3170467 and P3170469, a **powder-blue surgeonfish (*Acanthurus leucosternon*)** in P3170477, and **dense *Pseudanthias*-type anthias** over almost every reef-top frame — thick enough in P3170459, P3170461 and P3170469 that Coralscapes gives 18–20% of their area to fish.

**Two frames stand outside the survey.** **P3170447** has a bright green **St Patrick's Day hat, shamrock hatband and all, set down on top of a living *Pocillopora*-type colony** against white sand — BIEA25's flag (9) predicted this frame from a green woolly hat worn that morning, and it is confirmed. A prop, not equipment, but **it is resting directly on live tissue**, which is why it is recorded. **P3170483 is the cleanest single-diver-with-transect-pole frame in the redo**, a clean overhead of a diver carrying the white banded pole across a *Porites*-boulder pavement; P3170485 repeats it with three divers, and P3170484 shows two more in a heavy selective-desaturation edit. Coralscapes scores those three at 0.4%, 0.0% and 0.0% live, so **the dive's final frames contribute nothing but denominator** (flag 6).

<b>Assemblage composition.</b> Across 32 analysed frames the Coralscapes masks put live hard coral at 8.9% of image area, of which massive and meandroid is by far the largest component at 49.6%; the rest breaks down as other hard coral 25.3%, tabular <i>Acropora</i> 9.2%, <i>Pocillopora</i> 9.1% of live cover. The independent 30-point annotation of the same frames (960 points) gives live hard coral as 14.9% of all points and apportions it <i>Porites</i> 54.5%, other hard coral 23.8%, <i>Acropora</i> 18.9%, <i>Pocillopora</i> 2.8% of live coral — **the most <i>Porites</i>-weighted assemblage of the Eagle Island series**, against BIEA25's <i>Acropora</i>-led 49.2%. Non-coral cover is dominated by crustose coralline algae 4.4%, turf 25.7% of all points. Structural loss registers as dead coral 1.8% of image area and rubble 11.4% — **more than double BIEA25's 5.2%**, consistent with the rubble plain. Bleaching stands at 1.35% of annotated points — **13 points in total** — concentrated in <i>Acropora</i> 1.15%, other hard coral 0.21% of points. Frame-to-frame live cover runs 0% to 30.0% by point count (0% to 28.4% by mask area). Conspicuous fauna and sessile invertebrates appear as sponges in 21 frames, debris in 5 frames, urchins in 1 frame, anemones in 1 frame, and **crown-of-thorns in 1 frame — a call I could not confirm and believe is spurious (flag 7).**

**Flags:** (1) **BLUE-WATER DILUTION IS PRESENT AND FAR TOO SMALL TO EXPLAIN THIS SITE.** Nine frames are ≥40% water (P3170441, 443, 457, 460, 462, 478, 483, 484, 485). Dropping them moves CoralNet 14.9→**16.5%** and Coralscapes 8.9→**9.9%** against a 21.0% transect — **the gap narrows by only 1.6 and 1.0 pp and stays firmly negative, at −4.5 and −11.1 pp.** A <30%-water filter (19 frames) gives 17.4% and 9.6%; dropping the five diver frames too changes almost nothing. **BIEA20's flag-(1) mechanism operates in the expected direction and accounts for about a sixth of the CoralNet deficit and a twelfth of the Coralscapes deficit. It is not the explanation.** (2) **THE CORALNET DEFICIT IS ARITHMETIC; THE CORALSCAPES DEFICIT IS REAL.** As a fraction of points that landed on substrate rather than open water, the point set gives **19.4% live — within 1.6 pp of the transect** (29.6% if the 26.5% "Unknown" points go too). The same renormalisation for Coralscapes, excluding its background/dark class, gives only **15.8%**. **Fix the denominator and CoralNet barely disagrees with the transect; Coralscapes still does.** (3) **CORALSCAPES FAILS ON THE FINE-GRAINED MOSAIC, NOT ON THE BIG COLONIES — THIS IS THE MAIN FINDING.** **16 of 32 frames get under half the point set's live cover from the masks**, and the within-site correlation of the two per-frame estimates is only **r = 0.63**. Where coral is large, discrete and close-range the masks match or beat the points — P3170454 26.2 vs 20.0, P3170455 26.6 vs 23.3, P3170462 28.4 vs 30.0. Where it is a dense wide-angle mosaic of small, turf-fringed, low-contrast colonies they collapse — **P3170461 9.6 vs 23.3, P3170469 12.8 vs 30.0, P3170468 10.4 vs 20.0, P3170465 1.2 vs 16.7, P3170480 4.0 vs 23.3.** Those first two are the frames I read as the highest-cover at the site. **A resolution/contrast failure mode, and the single biggest contributor to the −12.1 pp.** (4) **THE BIEA24 BLEACHING MECHANISM DOES NOT OPERATE HERE, WHICH IS ITSELF NOTABLE.** BIEA24 found Coralscapes folding pale tissue into live coral, inflating cover. Here it reports **0.01% bleached site-wide** at a site published at **8.61% bleached** — and its live estimate is deflated, not inflated, so pale tissue is going somewhere other than live coral. **The exported columns cannot show where**: the CSV omits Coralscapes' rubble/rock/CCA/turf classes and its nine columns sum to only **63.5%** of image area, while its background/dark class averages **43.3%** against 23.1% water in the point set (84–89% on the diver frames). **Check the mask files before treating either as settled.** (5) **P3170484 IS A HEAVY SELECTIVE-DESATURATION EDIT** — monochrome background, colour only on the divers — and both CV methods scored it anyway: Coralscapes returned 0.0% for every benthic class, the point annotation returned 9 macroalgae points on a greyscale frame. **A direct cost of the edited-photo basis and the first clear instance in the redo. Screen the edited sets for selective-colour and B&W conversions before scoring.** (6) **METHODOLOGY IMAGERY, FIVE FRAMES** — P3170456, P3170458, P3170483, P3170484, P3170485; the last three carry the transect pole and score 0.4%, 0.0% and 0.0% live. Belongs with BIEA20's eight, BIEA25's six, BIDI08/P3120298, BIEG04/P3100401, BIEA18/P3150341 and BIEA19/P3150369. (7) **THE CROWN-OF-THORNS CALL AT P3170479 IS ALMOST CERTAINLY SPURIOUS.** `chagos_site_composition.json` lists `cots_frames: [["P3170479.jpg", 0.01]]` — **0.01% of image area, a few dozen pixels.** P3170479 is a shark-over-rubble frame and **there is no COTS in it.** No COTS, no disease and no predation scars in any of the 32 frames. **Do not let this reach the COTS log as a sighting.** (8) **THE MANIFEST `gps` FIELD IS MALFORMED — THE THIRD OCCURRENCE, AFTER BIEA24 AND BIEA25.** It stores `"6.17221°S, 71.33191°E · **In-situ temperature:** 29.2°C"`. Both values are correct and used above, but **any parser splitting `gps` on a comma returns a garbage longitude.** Three consecutive sites: **a systematic build fault, not a typo. Fix at the manifest-builder, then re-scan all 109 Chagos entries.** (9) **THE POINT SET RECORDS ZERO SAND POINTS HERE — 0 of 960**, on a site where P3170445, P3170454, P3170455 and P3170480–482 are substantially white sand (Coralscapes gives them 34.9%, 21.3%, 27.8%, 3.6%); the points call that ground **Turf**. Chagos-wide only **38 of 2,571 frames** carry any sand point — **the SAND label is effectively unused in this source.** An annotation-provenance artefact; it inflates turf but does not bias live coral. (10) **FOUR OF THE 13 BLEACHED POINTS SIT WHERE NOTHING SUPPORTS THEM** — **P3170445 (2 bleached-*Acropora* points on a frame that is one brown laminar colony on white sand, no *Acropora* in it)** and **P3170485 (2 on a 70%-water diver frame)**; P3170481 (2) and P3170482 (1) are on turf pavement. Concerns the point set only; **the published 8.61% comes from the supplement and stands.** (11) **FIFTEEN FRAME NUMBERS FROM THIS DIVE ARE ABSENT FROM THE FOLDER** — P3170448–453, 463, 466, 470–476. The n=32 rests on a curated subset of roughly 47 exposures. (12) No per-frame depth data.

**Benthic data:** [Download BIEA26 benthic survey data (CSV)](../benthic_data/BIEA26_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).