Log real-world road test results and correct/extend the CO-trim theory

New research/ROAD_TEST_LOG.md: the 2026-09-15 flash of
20262-arrow-delete-lowthrottle-composed.hex was a major real-world
success (stalling fully resolved, decel backfire much reduced but not
eliminated). Captures the live diagnostic reasoning that followed:
correcting an overly-lean-leaning framing mid-session (residual decel pop
is more likely a rich-unburned-mixture-with-no-SAI-assist problem, not a
lean one), a structural hypothesis that the Idle Fuel Trim (CO) curve
likely applies whenever the throttle is closed at any RPM rather than
only at a literal stop (only 2 of its 32 points are currently corrected,
and the untouched middle range lines up with the residual pop's RPM
window), and the reasoning for rejecting airbox-baffle removal as a fix
for decel popping specifically (wrong mechanism -- closed throttle isn't
airbox-limited -- and risks worsening the separately-flagged, still
unaddressed main-VE-table leanness gap).

Also updates TABLES.md with the CO-trim applicability open question, and
brings TUNING_GUIDE.md's "local research assets" section up to date (it
still said maps were gitignored/unflashable, both stale).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FP2GaxS9HkUdL5sLBnjKje

tunie/research/ROAD_TEST_LOG.md

@@ -0,0 +1,121 @@
+# Road test log — real-world tune results
+
+Empirical outcomes from actually flashing and riding, kept separate from the
+static reverse-engineering docs (`TABLES.md`, `DEVICES.md`) since this is
+field data, not derived facts about the ROM format.
+
+## 2026-09-15 — `20262-arrow-delete-lowthrottle-composed.hex`, first real ride
+
+Flashed after several days running plain stock `20262Map.hex` (SAI/O2 flags
+still enabled despite SAI being physically removed, snorkel already removed
+with zero main-table correction). Result: **major, immediate improvement.**
+
+- **Stalling at lights / low-speed choking: fully resolved.** Not just
+  fixed — noticeably *better* than expected: "a lot more low rev and slow
+  speed control," easier to get going, smooth at slow speeds. Attributed to
+  the Low-throttle fuel table correction (under ~10% throttle, both
+  cylinders, +7% to +24%) plus the two corrected Idle Fuel Trim (CO) points.
+- **Decel backfire (3-4.5k RPM, 2nd-4th gear): dramatically better, not
+  eliminated.** Before: "backfires like crazy," reliably reproducible.
+  After: 1-3 noticeable pops on a 5k→3k decel sweep in 2nd-3rd gear, and
+  *not* reliably reproducible anymore (sometimes it doesn't happen at all).
+- **New/newly-noticeable, minor**: holding steady revs at 5k, the engine
+  can "break up and get a tiny bit throaty." Described as minor, not
+  concerning, bike is very rideable as-is.
+
+**User's own assessment**: "In the current state I could drive it as is and
+be happy" — this is a real, working baseline, not a stopgap.
+
+## Diagnosing the residual decel pop — reasoning, 2026-09-15/16
+
+Two mechanisms were on the table for residual exhaust-side popping on
+deceleration:
+
+1. **Rich unburned mixture reaching the hot exhaust with no SAI assist to
+   burn it cleanly** — the classic mechanism for popping *from the exhaust*
+   on decel (as opposed to backfiring through the intake, which is the
+   lean-misfire pattern). SAI's actual job was adding oxygen downstream to
+   assist combustion of that charge; a fuel table can't replicate an air
+   *supply* by adjusting how much fuel goes in. **This is the mechanism the
+   user's own reasoning correctly pointed at** (session correction:
+   "I mean it's still rich, since that's what caused the popping" — right
+   call, corrected an overly-lean-leaning framing from earlier in the same
+   conversation).
+2. Unaddressed leanness in areas the current composed file doesn't touch.
+
+**Important structural correction, made mid-session**: the Idle Fuel Trim
+(CO) table (`0x53690`-`0x536AF`, 32 RPM-indexed points, same RPM axis as
+the main tables — see `TABLES.md`) was initially assumed to apply only at a
+literal stop. That's likely wrong. The table has real entries all the way
+out to 10,000 RPM — no reason for that range to exist if it only ever
+applied near idle. **More likely: this curve applies whenever the throttle
+is closed/near-closed, at whatever RPM the engine happens to be at** — i.e.
+it's the relevant table during an active decel sweep through elevated RPM,
+not just idle-at-a-stop.
+
+Given that, the composed file's coverage of this curve matters directly for
+the decel-popping symptom:
+
+| Curve region | RPM range | Correction status |
+|---|---|---|
+| Point 12 | 2400 | **Corrected** (see `compose_arrow_delete.py`) |
+| Points 13-26 | 2600-6500 | **Untouched** — still Arrow's original values |
+| Point 27 | 7000 | **Corrected** |
+
+The user's actual decel sweep (5000→3000 RPM) runs almost entirely through
+the **untouched middle section** of this curve. That's a more precise,
+evidence-grounded candidate for the residual pop than either "lean the
+idle" (don't — that's the low-throttle fix that's working great, protect
+it) or "richen the higher range" broadly (that's a different symptom,
+different table — see below).
+
+**Two symptoms, two different tables — don't conflate them:**
+
+| Symptom | Condition | Likely table |
+|---|---|---|
+| Decel pop | closed throttle, elevated RPM (falling through 3-5k) | Idle Fuel Trim (CO) curve, untouched middle points |
+| "Breaks up/throaty" at held 5k revs | open throttle, sustained load | Main VE fuel tables — still 100% stock-Arrow, unaddressed |
+
+The second one is the known, previously-flagged gap (no main-table
+correction exists for the snorkel/baffle removal). It is a *separate*
+project from the decel-pop investigation, not the same fix.
+
+## Baffle removal considered and rejected as a fix for decel pop (2026-09-16)
+
+User proposed removing the airbox internal baffle (on top of already-removed
+snorkel), reasoning more airflow would "even out" the remaining popping.
+**Rejected, reasoning below** — not because baffle removal is bad in
+general, but because it's the wrong tool for this specific symptom:
+
+- Decel popping happens at **closed throttle**, where the throttle plate
+  itself (not the airbox) is the dominant restriction. Baffle removal
+  doesn't meaningfully change airflow in that condition.
+- Baffle removal *does* increase airflow at every **open**-throttle
+  condition — exactly the untouched main VE tables. This widens the
+  existing, already-flagged leanness gap rather than closing it, and is
+  likely to make the "breaks up/throaty at 5k" symptom *worse*, not better.
+- Per `COMMUNITY_TUNING.md`'s TTP dyno data, baffle-removal+2-1 already
+  lands close to *full* airbox removal in airflow/power terms. Pulling the
+  baffle on top of an already-removed snorkel pushes past what either
+  reference data point in this project actually characterizes.
+
+**Recommendation given**: don't pursue baffle removal as a decel-pop fix.
+If pursued later for its own sake (more performance), it should come
+*with* main-table correction work, not before it.
+
+## Where this leaves things, if picked up again
+
+Current file (`derived/20262-arrow-delete-lowthrottle-composed.hex`) is a
+good, real, working baseline — user is happy running it as-is. If further
+work is wanted:
+
+1. **For the decel pop specifically**: pull real reference deltas for the
+   untouched CO-trim points in the 2600-6500 RPM range (same source
+   methodology as the rest of this project — `20184dynoTuneSteveO2-Disable.hex`
+   vs `20186Map.hex`), test in isolation, don't touch the low-throttle
+   fix that's already working.
+2. **For the main-table gap**: the bigger, separately-tracked project —
+   full VE-table composition for the snorkel/airbox-baffle removal, still
+   not started (`COMMUNITY_TUNING.md`'s "~1340-byte main VE overlap").
+3. Neither is urgent — current state is rideable and the user is satisfied
+   with it. This is optimization, not a fix for something broken.

tunie/research/TUNING_GUIDE.md

@@ -487,22 +487,27 @@ odometer").** Wrong-generation map IDs that came up and got ruled out:
   variants (`20262`-`20316`), the real SAI/O2 delete map
   (`20188Map2009AIRBOXBONNY.hex`), the America dyno tune
   (`20184dynoTuneSteveO2-Disable.hex`) plus its stock baseline
-  (`20185`/`20186`). Gitignored (`*.hex` at the samplez repo root) — these
-  are working files, not committed, so they won't survive a fresh clone;
-  re-download commands are in each research doc's history if needed again.
-- `reference-maps/derived/` — our own composed candidates (SAI-only,
-  SAI+O2, the trim-composed Arrow variant). **None are known-flashable**
-  — the ECU flash checksum isn't solved (`docs/ROADMAP.md` Phase 3),
-  separate from everything in this document.
+  (`20185`/`20186`). **Committed to the repo as of 2026-09 (`.gitignore`
+  rule removed, explicit user call)** — no longer working-files-only,
+  reachable from a phone browser for TuneECU use.
+- `reference-maps/derived/` — our own composed candidates. **Flashable —
+  the checksum problem is solved** (`reference-maps/checksum.py`, see
+  `docs/ROADMAP.md`). `20262-arrow-delete-lowthrottle-composed.hex` has
+  been flashed on the real bike and road-tested successfully — see
+  `ROAD_TEST_LOG.md` for actual outcomes, not just derivation.
 - `reference-maps/{decode_map,reconstruct_rom,table_map,toggle_devices,
-  compose_arrow_delete}.py` — the working toolchain: decrypt → flat ROM →
-  named tables → device-flag edit → trim-byte compose.
+  compose_arrow_delete,compose_lowthrottle,checksum}.py` — the working
+  toolchain: decrypt → flat ROM → named tables → device-flag edit →
+  trim-byte compose → low-throttle table compose → checksum patch.
 - `tuneecu_site_mirror/` — full local mirror of tuneecu.net's
   documentation (built by `tuneecu_crawl.py`), including the German
   road-tuning primer. The live site says it won't be updated after
   September 2024 — this mirror is insurance against link rot.
 - `PERMUTATIONS.md`, `TABLES.md`, `DEVICES.md`, `COMMUNITY_TUNING.md` — the
-  detailed research trail; this file is the index, those are the evidence.
+  detailed research trail (how the ROM format works); this file is the
+  index. **`ROAD_TEST_LOG.md`** — the separate, real-world log: what
+  actually happened flashing and riding on it, distinct from derived facts
+  about the ROM.
 
 ## Open questions, if this gets picked up again
 

tunie/research/reference-maps/TABLES.md

@@ -133,6 +133,25 @@ Composing two independent tunes' curves here needs the same per-cell
 arithmetic treatment as the main VE tables (`PERMUTATIONS.md`), not a
 handful of spot fixes.
 
+**Open question, raised from real-world testing (2026-09-15/16, see
+`../ROAD_TEST_LOG.md`): does "idle" in this table's name mean literal
+engine-stopped idle, or does it apply whenever the throttle is
+closed/near-closed regardless of RPM?** TuneECU's own docs say "at idle,"
+but the table has real, non-trivial entries all the way out to 10,000 RPM
+— hard to explain if it only ever applied near a stop. Circumstantial
+support for "applies whenever throttle is closed": a real-world decel
+sweep through the *uncorrected* middle of this curve (RPM 2600-6500,
+composed only touched points 2400 and 7000) produced residual popping in
+exactly that RPM range, while the low-throttle table fix (a genuinely
+different, confirmed-idle-adjacent table) independently and completely
+fixed literal stopped-idle stalling. That's consistent with this curve
+being live during the whole closed-throttle deceleration, not just at a
+stop — but it's inference from one road test, not confirmed in the
+decompile. Worth designing a deliberate test for if this gets revisited:
+does the CO Trim value visibly change on TuneECU's live-diagnostics screen
+while decelerating at 4000 RPM with the throttle shut, or only once RPM
+actually reaches idle?
+
 ## SAI / O2 / lambda flags
 
 Now findable via the localised flat-ROM diff (20187 vs 20188), which isolates small