diff --git a/continue.md b/continue.md
index fef8c73..af45712 100644
--- a/continue.md
+++ b/continue.md
@@ -5,7 +5,7 @@
 ### 1. Structure
 - One `SCENARIO("description")` per logical test group, with `[tag1][tag2]` annotations
 - Shared fixture: all constants, structs, and variables declared between `SCENARIO` opening and first `SECTION`
-- Precompute expected values analytically at fixture level (use `scripts/*.py` for complex simulations)
+- Run `scripts/precalc_*.py` to generate expected values — these Python scripts are the authoritative source, providing an independent check against the C++ simulation. Embed the values directly in `WithinAbs()` calls (no need to declare named constants unless the value is reused).
 
 ### 2. Duplication Elimination
 - Use lambdas that capture the fixture for repeated setup→call→assert patterns
@@ -13,14 +13,18 @@
 - Single-line `SECTION`s when body is one statement: `SECTION("name") { helper(arg); }`
 
 ### 3. Assertions
+- Include `src/test_utilities.h` for tolerance constants and test utilities
+- `#include <catch2/matchers/catch_matchers_floating_point.hpp>` (required for `WithinAbs` matcher)
 - `using Catch::Matchers::WithinAbs;` after includes
 - `REQUIRE_THAT(value, WithinAbs(expected, tolerance))` — never `Approx()`
 - **Always use named tolerance constants** — never hardcode raw numbers for tolerance in `WithinAbs()`.
 
 #### Tolerance Reference
+All constants defined in `src/test_utilities.h` — use those, do not redefine locally.
+
 | Constant | Value | Use for |
 |----------|-------|---------|
-| `A_TOL` | `1e-6` | Semi-major axis (meters), magnitude < 1e10 (use `A_TOL_LARGE` for larger) |
+| `A_TOL` | `1e-6` | Semi-major axis (meters), magnitude < 1e10 |
 | `E_TOL` | `1e-12` | Eccentricity, round-trip conversion |
 | `ANG_TOL` | `1e-12` | Angles in radians (nu, inc, Ω, ω) |
 | `ANG_TOL_COARSE` | `1e-4` | Angles, degenerate cases (polar/retrograde) |
@@ -30,7 +34,7 @@
 | `REL_TOL` | `1e-8` | Relative / percentage errors (dimensionless) |
 | `DRIFT_TOL` | `1e-12` | Energy drift percent (parabolic orbit) |
 
-- Declare tolerance constants in the fixture (between `SCENARIO` opening and first `SECTION`)
+
 - Tighten aggressively: if observed error is `1e-8`, use `1e-6` (two orders of margin)
 - Replace qualitative checks (`a > b`) with quantitative (`WithinAbs(expected, tol)`)
 - `INFO("label: " << value)` for debugging context
@@ -42,14 +46,14 @@
   - C++ tests typically use local coordinates (e.g., `vec3_distance(craft->local_position, (Vec3){0,0,0})`).
   - Global distances are dominated by parent body positions (e.g., Earth-Sun distance swamps LEO orbit).
 - **Always output SI units** (meters, m/s, seconds) — C++ tests use SI internally.
-- Output C++-style comments with precalculated constants for embedding in the test.
+- Output C++-style comments with precalculated expected values for embedding in the test. Tolerances are chosen separately by the test writer using the Tolerance Reference table — the precalc script should not output tolerance values.
 - Run with: `python3 scripts/precalc_<test_name>.py`
 - If sim_engine.py lacks a feature, use analytical formulas instead (**but notify the user what feature was missing**)
 
 ## Refactoring Procedure
 
 ### Step 1: Refactor
-- Verify the test file has a TOML config in `old_tests/`. If it doesn't, skip — it's likely hardcoded.
+- Verify the test file has a TOML config in `old_tests/`. If it doesn't, skip the TOML rewrite step — the test is likely hardcoded.
 - Check if the test is already in `tests/` (already refactored). Skip if so.
 - Check the capability matrix in Section 5 — if the test needs SOI, maneuvers, rendezvous, etc., flag this before starting.
 - Process **one test file at a time**.