On your entries, as a first import
The 392 entries whose ledger is known (your choices and the matches they made), replayed with no earlier choices — the worst case.
Where the 392 entries end up
Each bar splits the same 392 entries by outcome. Hover or Tab onto a segment for its count.
Show as a table
| Outcome | Before | After |
|---|
Problem 1 — the 40% bar decided the step
A suggestion under 40% likely was hidden as “Possible” and the entry waited in Suspense. Now the first suggestion always fills the box, and its “% likely” stays visible in the list.
Before Routing
Now Routing
The server now sends preselect: true for every suggested entry, so installed apps follow the new rule as soon as the server is deployed.
Why a full name showed “17% likely”
The % comes from a replay against older Miracle books. For “ledger name found in the narration”, most of its “wrong” rows were not wrong names:
What made it 13%
- 16 rows: the company’s own name (MARUTI ENERGY) printed on its own receipts. Now the own name is never suggested.
- 7 rows: MOLIYA DHANANJAY / SUSHIL — the old books posted them to LABOUR WAGES EXP.; you now use their own ledgers.
- 6 rows: INTEX CAD/CAM — two real ledgers (creditor and unsecured loan).
Measured again
9 of 69 right before → 103 of 129 right now (more rows, because glued names are now read).
- MOLIYA, GAFAR, INTEX, R P, J P, UCAM: 17% → 80%
- Alphatech (name cut short): 83% · Priyankkumar (cut): 95%
Problem 2 — names the matcher could not read
These narrations name the ledger, but the old reader missed them completely.
| Narration | What blocked it | Now suggests |
|---|---|---|
| 168033ALPHATECH ENTERPRIS ICICI | Cheque number glued to the name | Alphatech Enterprise Part of the name · 83% |
| 168037MULTITECH ENTERPRIS KOTAK… | Cheque number glued to the name | MULTITECH ENTERPRISE |
| 168039ARGUS CARBIDE PRIVA KOTAK… | Cheque number glued; PRIVATE cut short | Argus Carbide Private Limited |
| 168042J P INDUSTRIES H D F C BANK LTD | Cheque number glued to an initial | J.P. INDUSTRIES |
| RTGSOSKO VALVES PRIVATE | RTGS glued to the name | Osko Valves Private Limited Now saved by itself |
| NEFT OSKO VALVES PRIVAT | “PRIVAT” not read as PRIVATE | Osko Valves Private Limited Now saved by itself |
| NEFT UCAM PVT LTD | One-word name too short to search | UCAM PVT. LTD. Only when PVT / LTD follows · 80% |
| RTGSPRIYANKKUMAR POPATBHKKBKR52026091700871825 | Cut name glued to Kotak’s RTGS reference (KKBK + R + number) | Priyankkumar Popatbhai Bhuva (Capital Acc.) Name cut short · 95% |
| NEFT ROHIT CHOTHABHAI B | Only the father’s name matches, spelled CHOTHABHAI; surname is just “B” | CHOTHABHAI BIJALBHAI BHUNIDYA (ADVANCE), then CHOTABHAI BHUNDIYA One name of the ledger · as written first · ~38% likely |
| LOANCHARGES · CASHWITHDRAWCHR · INSPECTIONCHARGERECOVERY | Charge word glued to others | BANK CHARGES |
| 0000000/DR: ATM ANNUAL FEE | ATM read as cash | BANK CHARGES Was wrongly Cash Account |
| Paid to SELF | No rule for a cheque to SELF | Cash Account A NEFT/UPI to SELF is not cash |
Every entry, before and after
Your 392 entries grouped by narration (reference numbers shown as #). “Before” and “Now” show the step the entry lands in on a first import.
| Rows | Narration | Your ledger | Before | Now |
|---|
Check against Miracle books (10 companies, 1,611 rows)
The replay test compares matching with what the books show. “Fresh” is a first import; “prior” and “full” learn from earlier choices.
| Outcome | fresh: before | now | prior: before | now | full: before | now |
|---|---|---|---|---|---|---|
| Saved by itself, right | 196 | 220 (+24) | 497 | 516 (+19) | 632 | 646 (+14) |
| Saved by itself, WRONG | 0 | 0 | 1 | 1 | 1 | 1 |
| First suggestion right | 417 | 624 (+207) | 396 | 441 (+45) | 563 | 585 (+22) |
| Right ledger only in Suspense | 160 | 0 (-160) | 53 | 0 (-53) | 17 | 0 (-17) |
| No suggestion at all | 697 | 584 (-113) | 566 | 523 (-43) | 330 | 302 (-28) |
| First suggestion wrong | 141 | 183 (+42) | 98 | 130 (+32) | 68 | 77 (+9) |
| Decisions left for you | 998 | 767 (-231) | 718 | 654 (-64) | 416 | 380 (-36) |
Read it like this
- Wrong automatic saves did not move (0 / 1 / 1).
- 231 more entries are right on a first import; 231 fewer decisions.
- No entry that was right before is wrong now. Each new wrong first suggestion was, before, a weak guess hidden in Suspense or no guess at all.
Why “First suggestion wrong” still rises
Showing every guess means weak guesses can be wrong in plain sight. After the first fix it rose 141 → 297 (fresh). Tracing those rows found two causes, now fixed:
- 114 rows: “Interest” suggested Miracle’s built-in Interest Expense A/c.(Default) first, with INTEREST ON TERM LOAN second. Built-in “(Default)” ledgers now come last. Your Interest rows: 2% → 80%.
- 4 rows (full): a UPI ID used only for money in suggested GOODS SALES on a payment. Sales, purchase, income and expense ledgers are no longer suggested against their direction.
What is left (+42 fresh, +9 full) is mostly salary rows with several salary ledgers, people who share a first name with a ledger, and old bookings.
Your decision and next steps
Decide: bulk accept
Ctrl+Shift+Enter accepts every suggestion shown — now including weak guesses such as “Looks like a loan payment · 2% likely”. Keep it, or skip suggestions under some % in the bulk accept?
To reach the app
- Say “file the PR”: OpenQodex review, PR, squash-merge.
- Then, when you ask: server deploy (routing changes for installed apps at once).
- Then a desktop release for the Miracle history change.