PUBLIC SCORING CHARTER

Public Scoring Charter

This page discloses our scoring method in full: how scores are computed, when they are shown, when they are frozen, how seats are allocated, and how mistakes are corrected. You don't need to trust us — you can verify it yourself.

Charter version：charter v2.0 Weights version：weights 1.0.0 Monthly public audit summary，corrections and audit records are published at /audit/

Article 1

Six Commitments

We keep these six commitments for every single score — and where we can't, we would rather not show a score at all.

Scores are reproducibleUsing the same public rules and the same raw data, anyone can recompute the same score.

Scores are decomposableEvery total score can be broken back down into its four dimension sub-scores and the calculation behind them — never a black-box number with no explanation.

Scores cannot be boughtAdvertising spend, affiliate partnerships and sponsorships never raise a score by a single point.

If we don't know, we say soWhen data is missing we show "pending verification"; missing data is never treated as a score of 0.

Every change leaves a recordEvery score change is logged and kept for at least 24 months for review.

Mistakes are fixed, and fixed in publicErrors we find are corrected, and the correction process is public — nothing is quietly changed.

Article 2

Four-Dimension Weights

AMPM Score is a weighted combination of four dimensions; the full calculation formula and parameters for each dimension are as set out in the full text of the Scoring Specification v2.0.

30%

Utility

25%

Value

Worth it or not

15%

Accessibility

Ease of access

30%

Reliability

Reliability can't be bought back: When Reliability is below 40, no total score is shown — only a warning. And no matter how high the other three dimensions are, the total can be at most 25 points above Reliability. If reliability is lacking, the other dimensions cannot drag the score up.

Article 3

How Scores Are Computed

Peer-group comparison: tools are compared only against tools of the same type, at the sub-category level (L2); if a peer group has fewer than 8 tools, comparison falls back to the parent category (L1) to avoid distortion from a tiny sample.
Percentile ranking: rank is determined by percentile within the peer group; before computing, the most extreme values below the 5th and above the 95th percentile are trimmed, so a single spike or crash cannot pollute the overall ranking.
Small-sample shrinkage: when a peer group has too few samples, scores are shrunk towards the group median, so an over-confident score is never given from a handful of samples.
Time decay: with a half-life of 180 days, the influence of older data and events on the score fades over time, so the score keeps reflecting the present rather than being stuck on old records.

Article 4

Seven Statuses

At any point in time, every tool is in exactly one of the following seven statuses — and the status itself is part of the public information.

Status	What you see	What it means
Not included	Does not appear on any list or page	Not yet on the monitoring list. This does not mean the tool is bad — it simply hasn't been included for evaluation yet.
Under evaluation	Labelled "Under evaluation"	Just added to the watch list and in its cooling-off period; data is still accumulating and is not yet sufficient for a formal score.
Provisional	Shows a score range plus a "Provisional" label	There is just enough data for an estimate, but not enough confidence to finalise it — so a range is given rather than a single number.
Rated	Shows a definite score plus a confidence indicator	Sufficient data, passed verification, formally included in rankings and scores.
Under review	Labelled "Under review", score frozen	An anomaly rule under Article 5 was triggered and a re-check is in progress; during the review the score is frozen and no points are deducted; the case is closed within 14 days at most.
Downgrade	Shows a warning only, no score	A key indicator such as reliability is below the threshold; only a warning message is shown and the total score is no longer displayed (per the hard rule in Article 2).
Archived	Shows the historical score, labelled "Archived"	The tool is no longer tracked or has been discontinued, but its historical score record is kept permanently and remains reviewable.

Article 5

Three Anti-Manipulation Rules

Three common ranking-manipulation tactics, each blocked by a dedicated rule.

Anti-floodingIf a single source shows an abnormal surge within 72 hours (more than 3 standard deviations), that source's data is quarantined and excluded from the score.

Anti-self-dealingReviews from parties with a vested interest have their weight set straight to zero and cannot move the score.

Anti-timing manipulationAn abnormal spike just before a key ranking moment is frozen and sent for review rather than accepted at face value.

Due process: Triggering a rule above does not mean a violation has been found. A review is closed within 14 days at most; if not closed by then it is automatically unfrozen; if no evidence is found the original score is restored retroactively; only 3 substantiated cases within 12 months lead to removal.

Article 6

Allocation of the 100 Seats

The 100 seats of AMPM100 are allocated by the Hamilton largest-remainder method, in proportion to the number of formally rated (RATED) tools in each category; every category is guaranteed at least 1 seat.

When seats are tied, the following are compared in order: ① unrounded total score ② reliability score ③ confidence ④ which tool was formally rated first ⑤ lexical order of the tool ID — item by item until a winner emerges.

Popularity is not quality: Page views, clicks and traffic are never counted towards the quality score — a score does not rise because more people view or search for a tool.

Article 7

Appeals & Audits

If you think a score was computed wrongly, or want to challenge a decision, these are the response times we commit to.

working days to acknowledge

calendar days to close

hours to publish an erroneous score

Monthly

public audit summary

All correction records and audit summaries are published at /audit/。