Corrections Log

We published a hard cap that limited any project near a documented flood hotspot to a maximum score of 5.4, and described it as applying where there was "no engineered mitigation". Three things were wrong with it. We had no mitigation data for any project, so the second half of the rule was never tested and every project was treated as unmitigated. The Location & Climate axis was already reduced to 5.4 on 91 of these 93 projects — a value that appears on none of the other 6,810 — so the cap charged the same hazard a second time. And 31 of the 93 were flagged not by measuring the site but by matching the project’s locality name against a hotspot list.

The cap was retired. Flood-hotspot proximity now does two things and no more: it lowers the Location & Climate axis, which carries 14 of the 100 weight points, and it withholds a Buy verdict regardless of the total score. That Buy restriction was already in force and is unchanged; it is now stated publicly, which it had not been. Where a developer evidences engineered mitigation for a specific project, we record it and lift the restriction. Withholding a recommendation on incomplete evidence is fair; publishing a two-point downgrade on it is not.

No published verdict changed — none of these projects could carry a Buy before or after. 82 projects keep the score they had rather than dropping by an average of 2.2 points.

Axis-6 distribution across all 6,903 scored projects; the flood basis string on each of the 93; and the Buy-suppression path in the verdict gate, which was left untouched.

Three different versions of the 11-axis weightings were live at the same time. The scoring page, the methodology page and the generator that produces paid reports each used a different set, so a weight a reader saw was not necessarily the weight the arithmetic used.

All nine places where the weights are declared were aligned to the canonical set held by the scoring engine. A dated entry was added to the scoring changelog, which until then carried no record of the 2026-06-02 reweight and so contradicted the methodology page on 7 of 11 axes.

A script that reads all nine declarations and asserts each set sums to 100 and matches the canonical order.

A flood-risk cap was labelled "CWC Category-A flood zone". Central Water Commission Category-A is a major-river flood-forecast classification. Bengaluru has no CWC river category — it floods from lake overflow and storm-water drain failure, which is a different hazard from the one the label named.

The cap was renamed "BBMP flood-hotspot proximity, no engineered mitigation" and its wording brought in line with what is actually measured. The threshold itself was not changed.

Scores fell on pages where the cap applies. No project gained a Buy from this change, and the underlying flag had been visible on those pages throughout.

CWC flood-forecasting scope; BBMP flood-vulnerability records for the affected wards.

The page carried another phase's records: the RERA registration number, escrow bank and survey number belonging to Phase V, plus a parking ratio and floor-space figure taken from a different block altogether.

Corrected in place to the Phase-2 record — RERA PRM/KA/RERA/1251/309/PR/160124/006550, promoter Bhartiya Urban, Survey 53 Kogilu, Standard Chartered escrow — and the mis-attributed parking and floor-space figures were deleted rather than replaced, because no verified figure existed for them.

Removing the mis-attributed figures dropped the count of independently verified documents on this page to three, below the five we require before publishing a verdict. The page had previously looked close to a Buy on the strength of evidence that belonged to a different phase. It now publishes no verdict at all.

The project's own 19-document KRERA filing set, including the registered joint-development agreement, encumbrance certificate, khata, mutation record and the portal's bank-details record.

Our change monitor read the word "withdrawn" out of the sentence "Amount withdrawn from the bank for the purpose of the project" — a routine escrow drawdown line — and reported the registration as withdrawn. It also read "100% complete" out of a page-layout style value and an escrow account label. Separately, the approving authority was recorded as Mangaluru UDA, and the project start date was wrong.

The monitor now reads only the labelled status field instead of matching text anywhere on the page. The authority was corrected to the Bangalore Development Authority and the start date corrected against the filing.

A second correction on the same page ran against us: the land record had been described as a BBMP A-Khata when it is a Gram Panchayat e-Khata on converted land. That correction lowered the page's legal-cleanliness score rather than raising it.

The project's full KRERA detail filing (235 documents), where the live status read "New Project Launch, 24% complete" — matching the promoter's own quarterly progress report.

A bare "category-A" flood token had been applied to 349 projects with no source behind it, and it triggered a hard ceiling of 5.4 on the overall score. 151 of those projects had nothing more specific than "Bangalore" recorded as their locality, and the flag was firing on demonstrably elevated ground, including the Whitefield ridge and Yelahanka.

We built a sourced hotspot dataset — 84 locations with coordinates, radius and a stated basis, drawn from BBMP vulnerable storm-water-drain listings and published flood-prone-area records — then reclassified every flagged project by measured distance to it.

184 of 290 flags — 63% — were false and were corrected, releasing scores that had been artificially capped. 106 were kept where the flood risk is documented. Two of the twelve project pages published here retain the classification: it was re-examined and stood.

BBMP vulnerable storm-water-drain listings, published Bengaluru flood-prone-area records, and per-project coordinates.

One project was found carrying a RERA registration number belonging to a different project. An audit of every page with a committed number found 240 that needed review.

23 wrong registration numbers were removed, together with the escrow, approving-authority, progress-report and document records that had been attached on the strength of them. 32 pages filed under the wrong city were corrected. Where a correct number could be established from the public register it was recorded separately as a candidate rather than promoted automatically, because promoting it is an editorial decision.

A guard was added upstream: a close name match in a district outside Bengaluru now routes to human review instead of being committed automatically, which is the specific path that produced the original error.

State RERA registers, checked per project.

The encumbrance-certificate field on 437 pages had been populated with escrow bank text. The effect was that a page could present an encumbrance certificate as verified when no encumbrance certificate had been read.

All 437 were cleared, and the field is now written only from a document that has actually been retrieved and read.

Direct inspection of the stored values against the documents they claimed to represent.