diff --git a/CONTEXT.md b/CONTEXT.md
index cb977c4ff..aa2199323 100644
--- a/CONTEXT.md
+++ b/CONTEXT.md
@@ -86,7 +86,7 @@ One certificate row from the final data dump of the shut-down old EPC register,
 _Avoid_: old EPC (ambiguous with a pre-SAP10 cert from the new API), historical EPC API (the API is gone; only the backup exists)
 
 **Expired-Enhanced Prediction**:
-**EPC Prediction** for a Property whose only certificate predates 2012: the expired **Historic EPC**, found by exact UPRN in its postcode shard, **conditions** cohort selection with its *stable* attributes (property type, built form, wall material, roof construction, age band, main fuel, and a ±5% floor-area band) exactly as a Landlord Override would. Volatile attributes (heating, hot water, glazing, PV, insulation, lighting) are excluded — 14+ years stale — and stay neighbour-predicted; the historic cert is **never copied into the Effective EPC as current state**. Persisted to the predicted slot with `source="expired"`. Scoped to the historic backup only; post-2012 expired certs from the new API keep their existing treatment (ADR-0054).
+**EPC Prediction** for a Property whose only certificate predates 2012: the expired **Historic EPC**, found by exact UPRN in its postcode shard, **conditions** cohort selection with its *stable* attributes (property type, built form, wall material, roof construction, the age band's ±1-band neighbourhood, main fuel, and a ±20% floor-area band) exactly as a Landlord Override would. Band widths are harness-evidenced, not guessed (ADR-0054 amendment: age band agrees with a relodged cert 52% exactly / 90% within one band; TFA 45% within ±5% / 82% within ±20%). Volatile attributes (heating, hot water, glazing, PV, insulation, lighting) are excluded — 14+ years stale — and stay neighbour-predicted; the historic cert is **never copied into the Effective EPC as current state**. Persisted to the predicted slot with `source="expired"`. Scoped to the historic backup only; post-2012 expired certs from the new API keep their existing treatment (ADR-0054).
 _Avoid_: historic override (it is conditioning, not an override the effective picture trusts), expired EPC path (names the input, not the operation)
 
 ### Survey documents
diff --git a/docs/adr/0054-expired-historic-epc-conditions-prediction-with-stable-attributes.md b/docs/adr/0054-expired-historic-epc-conditions-prediction-with-stable-attributes.md
index 58c869bc8..ba0067ce3 100644
--- a/docs/adr/0054-expired-historic-epc-conditions-prediction-with-stable-attributes.md
+++ b/docs/adr/0054-expired-historic-epc-conditions-prediction-with-stable-attributes.md
@@ -41,8 +41,10 @@ into the Effective EPC as current state**.
    — heating system, hot water, glazing, PV, insulation states, lighting — are
    excluded and stay neighbour-predicted.
 2. **Floor area is a tolerance band, not an override.** Comparables are
-   soft-filtered to within **±5%** of the historic certificate's total floor
-   area. The predicted floor area remains the cohort's geo-weighted median.
+   soft-filtered to within **±20%** of the historic certificate's total floor
+   area — a coarse dwelling-size filter, not a precision match. The predicted
+   floor area remains the cohort's geo-weighted median. *(Amended from ±5% —
+   see Amendment below.)*
 3. **Every historic filter rides the existing filter-then-relax ladder**
    (ADR-0029): an attribute that cannot be resolved into the cohort's code
    space maps to `None` and its filter is simply inactive; a filter that would
@@ -84,3 +86,26 @@ into the Effective EPC as current state**.
   predicted slot; `source_path` is unchanged).
 - The whitelist is a hypothesis until the pairs harness reports; attribute
   membership changes are cheap (one resolver each).
+
+## Amendment (2026-07-06): band widths set by the pairs harness
+
+The harness ran at scale (2,000 stride-sampled postcodes → **439 pairs**, 419
+scored per arm; tables on PR #1466) and two parameters changed with evidence:
+
+- **Age band conditions on the ±1-band neighbourhood, not equality.** The
+  historic band agrees with the newly lodged one only **52%** exactly but
+  **90%** within one band — assessors re-band constantly (skewing newer: 31%
+  of new certs band later). Equality-conditioning steered engaged cohorts
+  toward a coin flip.
+- **The floor-area band widens ±5% → ±20%.** Historic-vs-new TFA agreement:
+  45% within ±5%, 63% within ±10%, 72% within ±15%, **82% within ±20%**
+  (remeasurement + extensions). At ±5% the filter engaged on only 19% of
+  pairs and worsened the floor-area residual.
+
+Also established by the same evidence: main fuel is **95%** stable (after
+resolving the register's legacy "backwards compatibility" fuel descriptions),
+property type **91%**, wall construction **79%** — the whitelist's core holds.
+Known collateral to watch: `roof_insulation_thickness` transfers slightly
+worse from age-conditioned cohorts (negative in all three harness rounds) —
+candidate follow-up: exclude the roof-insulation mode when age conditioning
+engaged.