From 7eced7eadf41010e88a6014886dd0a8d570b2a6f Mon Sep 17 00:00:00 2001
From: Marcin Pawlowski <marcin@logos.co>
Date: Thu, 6 Aug 2026 17:36:03 +0200
Subject: [PATCH] reports/blend: record which evidence predates the
 seed-derivation change

data/README promised that re-running with an unchanged config reproduces exactly.
That is no longer true for three of the eight runs: adding failure domains put
n_regions and region_locality into the topology seed, which changes the graph
drawn for every config, including those leaving both at their defaults.

default, percolation and redundancy predate that change. A spot-checked cell moves
by 0.3%, about 1.5 standard errors -- ordinary variation between independent
realisations, not a change in behaviour, and within the error bars the report
already states. Closed-form quantities are identical either way.

The README now says which runs are current, which are not, and why.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---
 reports/blend/data/README.md | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/reports/blend/data/README.md b/reports/blend/data/README.md
index acfd7b3..801d2ff 100644
--- a/reports/blend/data/README.md
+++ b/reports/blend/data/README.md
@@ -39,9 +39,19 @@ runs instead.
 
 From [`tools/simulators/blend`](../../../tools/simulators/blend): `make sweep`,
 `make redundancy`, `make percolation`, `make correlated-churn`, `make sweep-fullscale`. Results
-land in that simulator's `runs/<timestamp>_<label>/`. Note that the seed streams depend on the
-configuration, so re-running reproduces the *statistics*, not bit-identical numbers, unless the
-config is unchanged — in which case it does reproduce exactly.
+land in that simulator's `runs/<timestamp>_<label>/`.
+
+**On exact reproduction.** Seed streams are derived from the configuration, so re-running a study
+reproduces its *statistics* rather than bit-identical numbers whenever that derivation has moved.
+It has moved once since these runs: adding failure domains (§3.9) put `n_regions` and
+`region_locality` into the topology seed, which changes the graph drawn for every config, including
+those leaving both at their defaults. `default/`, `percolation/` and `redundancy/` were produced
+before that change and so no longer reproduce bit-exactly — a spot-checked cell moved by 0.3 %,
+about 1.5 standard errors, which is ordinary variation between independent realisations rather than
+a change in behaviour. `correlated-churn/`, `cover-traffic/`, `fullscale/`, `timing/` and
+`attribution/` were produced by the current derivation. Quantities that are closed-form
+(`deanon_rate`, the quota ceiling, the confidence formulae) depend only on counts and are identical
+either way; the Monte-Carlo ones scatter within their stated error bars.
 
 Two runs from the same session are deliberately **not** kept: the smoke runs (throwaway, far too
 noisy to interpret) and an earlier 144-rounds/cell redundancy grid that was superseded because its