Override the engine — manual ID, valuation and re-identification for serious collectors

DetectID’s engine handles the heavy lifting on identification and valuation, but serious collectors and experienced detectorists often have information the engine doesn’t. This guide covers the four override tools that put you in the driver’s seat — promoting an alternative attribution, manually correcting the ID, re-running the engine with hints, and setting your own valuation. All are advanced features. Use them when you really know what you’re talking about.

At a glance — the four override tools

Promote alternative — one-click correction

The cheapest correction. When the engine returns its top candidate plus 1–2 alternatives, the alternatives often contain the right answer — they just didn’t score highest. Promoting an alternative does three things at once:

• Copies the alternative’s attribution (ruler, denomination, period, date range) onto the find itself.
• Updates your collection so search, sort, and filter all work against the corrected attribution.
• Records a “close — promoted alt #N” feedback row that’s sent back to the methodology team. Pattern recognition: if 14 users promote the same type from #2 to #1, that’s a methodology fix waiting to happen.

How to use it

1. Open the find result page: /identify/result/[id]
2. Scroll to the Top candidate(s) section. Below the highlighted top candidate, each alternative card now has a small “This one was right →” button.
3. Click. A browser confirm dialog appears — you can’t accidentally promote.
4. Confirm. The page reloads with the alternative now in the top slot, your collection re-labelled, and the edit-history timeline at the bottom showing the field-by-field change.

When NOT to use it

• The engine got it completely wrong (no alternative matches). Use Manual correction instead.
• You’re between two alternatives. Take a closer look at the coin before committing — the engine’s reasoning chain will tell you which diagnostic features it relied on.
• You’re guessing. The methodology team relies on these feedback signals; bad ones make the engine worse for everyone.

Manual correction — type the right answer

When none of the engine’s candidates match, you can override the ID by typing the correct attribution. Useful when:

• You’ve had an FLO confirm something the engine missed.
• The find is a series the engine doesn’t cover yet (Phase 1 of methodology) and you know the type.
• A trusted source on a forum or in person has identified it.
• You’re a numismatist who knows the type cold.

How to use it

1. On the result page, scroll below the candidates list.
2. Click “None of these are right — tell us what it is →”. An inline form opens.
3. Fill at least one of: ruler, denomination, period, date range. More is better — but one field is the minimum.
4. Optional notes field — say how you know (“FLO Sarah Davies confirmed in person at Northants Treasure Open Day 2026”, “matches Spink 2342”, “ex Tony Abramson collection”). These notes go to the methodology team for future engine tuning.
5. Click Save. The find is re-labelled across your collection; the result page reflects the new attribution; the edit history records the change.

What the methodology team sees

Your manual correction is recorded as “wrong — actually was X” on the original identification. We aggregate these signals to surface systematic blind spots — types or sub-types the engine consistently misses. Your notes field is invaluable here. The more detail you give about how you know, the faster the methodology improves.

Re-identify with hints — fresh engine run with your input

The two tools above let you tell the engine the answer. The re-identify flow goes the other way: you give the engine new context and let it have another go. Useful when you’ve learned something since the original ID but you still want the engine’s opinion rather than imposing yours.

• You finally measured the coin with a tape measure and have accurate diameter and weight figures.
• You found out from a forum or club mate roughly what type it is (“probably Henry VIII second-coinage groat”) but want the engine to confirm the specific issue / mint / bust style.
• The engine asked for better photos and you’ve since taken some — but rather than uploading them, you want the engine to have another go at the originals with the new context.
• A reverse-mark or legend has become clearer after cleaning, and you want to feed that as a hint.

How to use it

1. On the result page, click “Re-identify with hints →” link in the same row as the correction buttons.
2. The form pre-fills with what we already have on the find (existing measurements, material, region). Edit any of these, and add a free-text Notes / suspected attribution field.
3. Click Re-run engine. We download your original photos from storage, re-feed them to the engine with your new hints, and create a fresh identification row.
4. Wait ~10–20 seconds. The page redirects back to the result page, now showing the new attribution. The old identification is marked superseded but kept in the database.

Manual valuation — your own £ figure overrides the auction comp (Premium)

Auction-comparable pricing is one of Premium’s headline features — for every Premium result, the engine looks up real hammer prices from auction archives at the relevant grade. But sometimes you have a value you trust more than a generic comp. Manual valuation lets you commit your own figure.

When to set a manual valuation:

• A dealer has quoted you a price for the coin (Spink, DNW, Noonans, Baldwin’s, an independent specialist).
• You’ve had an insurance appraisal (Hiscox, Aviva, Direct Line Plus, T.H. March Jewellers etc.).
• You sold an identical or near-identical example recently and know what it actually realised.
• The auction-comp range from our archive is too wide for your purposes (e.g. £50–£500 for a rare type) and you have a tighter, dealer-confirmed mid-point.
• The piece is a one-off (modified coin, love token, pierced pendant) where the auction archive has no comp and you have a specialist opinion.

How to set it

1. Open the find: /identify/result/[id]
2. Click Edit (top-right of the “What you measured” panel). The full edit form opens.
3. Scroll to the gold-bordered Your valuation (Premium) section.
4. Enter the £ figure (decimals allowed: 450.00 or 450 both work).
5. Add a short source note — “Spink valuation Aug 2025”, “Hiscox insured at £450”, “Tony Abramson at BNFG 2026 reckoned this would hammer £400–500”. The note is private to you but appears alongside the figure on the result page so future-you remembers where it came from.
6. Save.

What changes downstream

• Result page: A gold-bordered card at the top of the page shows your figure, source note, and the date you set it. The auction-comp panel still appears below — your override and the engine’s estimate are visible side-by-side, but yours is the headline.
• Collection valuation (/finds/valuation): Your figure replaces the auction-comp range for this find. The find counts as a single-point value (not a range) when summing the collection total.
• Insurance PDF report: Your manual figure is the headline value on the find’s line in the PDF, with the source note printed alongside. Insurance underwriters tend to prefer concrete dealer/appraisal sources over auction ranges, so the override is usually the right answer for an insurance schedule.
• Edit history: Each change to the manual figure is logged in the find’s edit timeline, so you can see the value drift over time as appraisals are renewed.

Clearing the override

Leave the £ field blank when editing and save — the find falls back to the auction comparable. No data loss; the figure goes to NULL and we restore the auction-comp display.

Putting them together — a workflow example

A real example from one of our testers, lightly fictionalised:

1. Uploaded a hammered silver coin. Engine’s top attribution: Edward III silver penny, post-treaty period, high confidence. Two alternatives: Richard II penny, and Henry IV light coinage penny.
2. Tester took the find to their local FLO who said it was actually a Richard II — the RICARD legend is visible if you look at the obverse closely; the engine had misread the letterforms.
3. Tester clicked “This one was right →” on the Richard II alternative card. The find re-labelled, the engine feedback recorded. Edit-history showed the field-by-field change.
4. A week later, the tester took the coin to Spink for a valuation (it was a rarer variety of Richard II York mint penny). Spink quoted £180. Tester opened the find on DetectID, clicked Edit, scrolled to the gold “Your valuation” section, entered 180.00 with the source note “Spink quote 12 Aug 2025”.
5. The collection valuation total updated to reflect the new £180 figure, and the next monthly insurance PDF the tester ran showed the Spink-sourced valuation rather than the wider auction-comp range.

What overrides are NOT

• Not retroactive to the methodology.Your override changes your collection’s record. It does NOT retroactively change the engine’s methodology for other users — that happens through the methodology team’s review of aggregate feedback over time.
• Not a workaround for missing photos.If the engine returned “needs better photos”, that’s a signal that the diagnostic features genuinely aren’t visible. Adding hints via re-identify often unblocks this, but skipping straight to manual override risks committing the wrong attribution.
• Not a way to share your private valuation publicly. Manual valuations are private to the owner. They’re not surfaced on the community feed or on shared find pages — your valuation stays between you and your insurance schedule.
• Not for guessing.The whole point of these tools is to act on KNOWLEDGE you have that the engine doesn’t. If you’re not sure, leave the engine’s call in place and add a note instead — that doesn’t touch the attribution or the valuation.

Reverting changes

Every override is logged in the edit history at the bottom of the result page (Basic+). You can see what changed and when. To revert:

• Promoted alternative — re-edit the find via the full edit page and type the original attribution back in. The edit history records both changes.
• Manual correction — same. Re-edit.
• Re-identification — currently no UI for rolling back. The old identification row is still in the database (marked superseded). If you need to roll back a re-run, email admin@detectid.co.uk with the find ID and we’ll do it manually.
• Manual valuation — re-edit, clear the £ field, save.

Related guides

• What to do with an unidentified coin — the workflow upstream of all of this.
• The most common UK detector find misidentifications — knowing what the engine usually gets wrong is the first step to using these tools well.
• UK coin grading guide — grade affects valuation by an order of magnitude. Know the scale before you set a manual figure.