The TallyGo field guide
Good questions.
Real choices.
A fairer tally.
Here’s how individual recommendations become a useful community signal—and how to make sense of that signal.
Let’s break it down01 Start with what matters to you
“Best” needs
a little context.
A board groups a type of product, such as headphones or air purifiers. Inside it, each comparison asks a specific question: best value, best sound, quietest, and so on.
Those questions have separate rankings. A product can be someone’s pick for sound without being their pick for price. Choose the question that fits your decision.
Find the exact product. Search within the comparison by name, brand, or model. Manufacturer and price filters help narrow the list; they do not give a product extra ranking weight.
Different question → separate ballot → its own ranking.
02 Put your experience into the mix
One choice.
Not set in stone.
A recommendation is your answer to that comparison’s question. Pick the product you would recommend, based on what you know and have experienced.
- Sign in and verify your email.
Browsing is open to everyone. Voting requires a verified account.
- Choose “Recommend” on your product.
You get one active choice on that comparison’s ballot.
- Keep it, move it, or withdraw it.
A new pick replaces your previous one. It never adds a second active vote to the same ballot.
Different comparisons can have different picks. Your best-value recommendation does not replace your best-sound recommendation.
Which would you recommend?
Try A. Then try B. Watch your vote move.
Pick a pair above to add your example vote.
03 Behind the position on the board
A ranking.
With its working shown.
Products are ordered by integrity-weighted community support within the selected comparison. The original vote count stays visible so you can see the size of the sample.
- 1 Votes
- The number of active recommendations for this product. A raw count, before integrity weighting.
- 2 Weighted support
- The influence left after integrity checks and limits on correlated votes. It can be a fractional number.
- 3 Support share
- The product’s weighted support divided by the total weighted support in that comparison. It is not a star rating.
Headphones A
10.8 ÷ 18 total weighted votes = 60% support
Keep your curiosity.
No votes means no community leader. A handful of votes is an early signal, not a settled verdict. Look at the count before you lean on the rank.
04 People, not piles of accounts
Every vote is a voice.
Not every pattern is natural.
Verification is the starting point. TallyGo also checks for suspicious behavior and coordinated voting before calculating how much influence a vote has.
Account signals
Account age, authentication signals, and clean voting history help provide context.
Shared patterns
Device, network, and signup patterns can reveal groups of votes that may be related.
Limited influence
Suspicious votes can carry less weight. Correlated groups have a cap on their combined influence.
Rules calculate the weight. A language model does not decide the numerical weight of a vote. The current integrity model is experimental and can be recalibrated as real voting grows.
Verification has limits. A verified email is not proof of ownership, expertise, or one unique person. These checks reduce manipulation; they cannot guarantee its absence.
The exact scoring rulesIntegrity model v1 · weights, thresholds, and cluster limits
Each input signal is normalized between 0 and 1. A deterministic calculation estimates risk and turns it into a base weight:
logit = -3.0
+ 0.8 × new_account + 0.8 × network_risk
+ 1.4 × device_cluster + 1.3 × signup_cohort
+ 1.7 × vote_pattern
- 0.4 × strong_auth - 0.6 × clean_history
fraud_risk = sigmoid(logit)
base_weight = 1 − fraud_riskPublic weight is 1 − risk.
Public weight is 1 − risk.
Flagged for review; retains its reduced weight.
Public weight is zero.
Coordinated votes have diminishing returns.
For a correlated group of n votes, the combined weight is capped at 1 + log₂(n). If the group exceeds the cap, every member’s weight is reduced by the same proportion. For example, 8 related votes with total base weight 7.2 have a combined cap of 4.
cluster_weight = sum(member_weights)
cluster_cap = 1 + log₂(member_count)
scale = min(1, cluster_cap / cluster_weight)
effective_member_weight = member_weight × scaleQuarantined records remain available for review and replay. Neither retailer ratings nor payments are inputs to these calculations.
05 Useful context. Clear boundaries.
What moves the ranking.
And what doesn’t.
Different kinds of information do different jobs. Keeping those jobs separate makes the result easier to trust.
Community recommendations.
Active votes in that specific comparison, adjusted by the integrity rules. Support in another comparison is a separate answer.
- Paid placements and sponsorshipsA budget cannot buy a better position.
- Retailer stars and review countsShown separately as retailer context.
- Prices and product specificationsUseful for research and filtering, not extra votes.
- Owner reports and generated summariesMore context, with no automatic voting weight.
What about a product’s broader reputation score?How support across comparisons is summarized.
A product profile can summarize support across published, all-price comparisons. That summary is different from its position on a single board.
For each comparison, weighted vote share is multiplied by n / (n + 20), where n is the smaller of the distinct voter count and the total effective vote weight. This reduces the influence of tiny samples. The strongest contribution in each dimension is retained, then averaged across published dimensions and scaled to 100.
Distinct supporters are counted once across comparisons. Draft comparisons and separate price-segment ballots do not contribute to this aggregate. A comparison needs at least 20 voters and 20 effective votes before an untied first place counts as a win. Product profiles show the contributions used in their score.
These are descriptions of community support, not measured performance scores or statistical confidence guarantees.
06 A little more clarity
Good questions.
Straight answers.
Do I need an account to browse?
No. You can explore boards, search products, compare specifications, and read rankings without signing in. You need a signed-in account with a verified email address to recommend a product.
Does a verified account mean a verified owner?
No. Account verification confirms an email address, not a purchase or product experience. Owner reports have their own disclosures. An ownership claim should only be treated as verified when that specific report carries an explicit verification label.
Can I choose different products for different questions?
Yes. You might recommend one pair of headphones for best sound and another for best value. Each comparison has its own ballot. Where a named price segment has its own comparison, that ballot is separate too. Changing a vote replaces only the choice on that ballot.
What happens if I change my mind?
Recommend another eligible product in the same comparison and your active vote moves to it. The old pick does not keep an extra vote. Use Withdraw on your selected product to remove your recommendation. Your account page also shows your voting activity.
Why can a product with more votes appear lower?
The ranking uses integrity-weighted support first. A larger raw count can carry less influence after integrity checks and limits on correlated votes. When weighted support is equal, raw vote count and then product title determine the displayed order. That order alone is not evidence that a tied product is better.
Does first place mean the product is right for me?
It means that product has the most weighted support in that particular comparison. Check how many votes support it, what the question is, and whether its price and features fit you. A tiny sample can change quickly. Community preferences are not independent lab tests or a guarantee of quality.
Where do product details and prices come from?
The catalog uses retailer data and product specifications. Prices are observations and may change at the retailer. Product summaries can be assisted by automation; they are not invented hands-on reviews. Retailer links, ratings, and prices do not add community-ranking weight.
Grow with the community
Your experience.
A little recognition.
Earn badges for useful contributions and help decide which products we add next.
Verify your email. Share your picks and product experience.
10 eligible votes + 2 approved reviews.
25 votes + 5 reviews + 30 days. Request up to 3 products per rolling 30 days.
Fair credit. Thoughtful additions.
Vote credit counts once per comparison ballot, across price segments. Only eligible active votes count toward your current tier. Reviews must be substantive, original product feedback approved through moderation. One credit per product; edits and reposts add none. Questions and generic comments can remain useful discussion without earning review credit.
Product requests are suggestions for the catalog. Our team checks duplicates, searches the provider, reviews a preview and approves the import. You can follow progress in your account. Up to 3 requests may remain unresolved at once. Duplicate matches use no allowance; declined requests stay within your 30-day allowance.
Badges recognize contributions. They do not prove ownership or expertise, increase voting weight, or unlock free products. Removed spam and disqualified contributions cannot earn credit. Request privileges may be paused for abuse.